1 //===- SCCP.cpp - Sparse Conditional Constant Propagation -----------------===// 2 // 3 // The LLVM Compiler Infrastructure 4 // 5 // This file is distributed under the University of Illinois Open Source 6 // License. See LICENSE.TXT for details. 7 // 8 //===----------------------------------------------------------------------===// 9 // 10 // This file implements sparse conditional constant propagation and merging: 11 // 12 // Specifically, this: 13 // * Assumes values are constant unless proven otherwise 14 // * Assumes BasicBlocks are dead unless proven otherwise 15 // * Proves values to be constant, and replaces them with constants 16 // * Proves conditional branches to be unconditional 17 // 18 //===----------------------------------------------------------------------===// 19 20 #include "llvm/Transforms/Scalar/SCCP.h" 21 #include "llvm/ADT/ArrayRef.h" 22 #include "llvm/ADT/DenseMap.h" 23 #include "llvm/ADT/DenseSet.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/Transforms/Utils/Local.h" 33 #include "llvm/Analysis/ValueLattice.h" 34 #include "llvm/Analysis/ValueLatticeUtils.h" 35 #include "llvm/IR/BasicBlock.h" 36 #include "llvm/IR/CallSite.h" 37 #include "llvm/IR/Constant.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/GlobalVariable.h" 43 #include "llvm/IR/InstVisitor.h" 44 #include "llvm/IR/InstrTypes.h" 45 #include "llvm/IR/Instruction.h" 46 #include "llvm/IR/Instructions.h" 47 #include "llvm/IR/Module.h" 48 #include "llvm/IR/PassManager.h" 49 #include "llvm/IR/Type.h" 50 #include "llvm/IR/User.h" 51 #include "llvm/IR/Value.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/PredicateInfo.h" 59 #include <cassert> 60 #include <utility> 61 #include <vector> 62 63 using namespace llvm; 64 65 #define DEBUG_TYPE "sccp" 66 67 STATISTIC(NumInstRemoved, "Number of instructions removed"); 68 STATISTIC(NumDeadBlocks , "Number of basic blocks unreachable"); 69 70 STATISTIC(IPNumInstRemoved, "Number of instructions removed by IPSCCP"); 71 STATISTIC(IPNumArgsElimed ,"Number of arguments constant propagated by IPSCCP"); 72 STATISTIC(IPNumGlobalConst, "Number of globals found to be constant by IPSCCP"); 73 74 namespace { 75 76 /// LatticeVal class - This class represents the different lattice values that 77 /// an LLVM value may occupy. It is a simple class with value semantics. 78 /// 79 class LatticeVal { 80 enum LatticeValueTy { 81 /// unknown - This LLVM Value has no known value yet. 82 unknown, 83 84 /// constant - This LLVM Value has a specific constant value. 85 constant, 86 87 /// forcedconstant - This LLVM Value was thought to be undef until 88 /// ResolvedUndefsIn. This is treated just like 'constant', but if merged 89 /// with another (different) constant, it goes to overdefined, instead of 90 /// asserting. 91 forcedconstant, 92 93 /// overdefined - This instruction is not known to be constant, and we know 94 /// it has a value. 95 overdefined 96 }; 97 98 /// Val: This stores the current lattice value along with the Constant* for 99 /// the constant if this is a 'constant' or 'forcedconstant' value. 100 PointerIntPair<Constant *, 2, LatticeValueTy> Val; 101 102 LatticeValueTy getLatticeValue() const { 103 return Val.getInt(); 104 } 105 106 public: 107 LatticeVal() : Val(nullptr, unknown) {} 108 109 bool isUnknown() const { return getLatticeValue() == unknown; } 110 111 bool isConstant() const { 112 return getLatticeValue() == constant || getLatticeValue() == forcedconstant; 113 } 114 115 bool isOverdefined() const { return getLatticeValue() == overdefined; } 116 117 Constant *getConstant() const { 118 assert(isConstant() && "Cannot get the constant of a non-constant!"); 119 return Val.getPointer(); 120 } 121 122 /// markOverdefined - Return true if this is a change in status. 123 bool markOverdefined() { 124 if (isOverdefined()) 125 return false; 126 127 Val.setInt(overdefined); 128 return true; 129 } 130 131 /// markConstant - Return true if this is a change in status. 132 bool markConstant(Constant *V) { 133 if (getLatticeValue() == constant) { // Constant but not forcedconstant. 134 assert(getConstant() == V && "Marking constant with different value"); 135 return false; 136 } 137 138 if (isUnknown()) { 139 Val.setInt(constant); 140 assert(V && "Marking constant with NULL"); 141 Val.setPointer(V); 142 } else { 143 assert(getLatticeValue() == forcedconstant && 144 "Cannot move from overdefined to constant!"); 145 // Stay at forcedconstant if the constant is the same. 146 if (V == getConstant()) return false; 147 148 // Otherwise, we go to overdefined. Assumptions made based on the 149 // forced value are possibly wrong. Assuming this is another constant 150 // could expose a contradiction. 151 Val.setInt(overdefined); 152 } 153 return true; 154 } 155 156 /// getConstantInt - If this is a constant with a ConstantInt value, return it 157 /// otherwise return null. 158 ConstantInt *getConstantInt() const { 159 if (isConstant()) 160 return dyn_cast<ConstantInt>(getConstant()); 161 return nullptr; 162 } 163 164 /// getBlockAddress - If this is a constant with a BlockAddress value, return 165 /// it, otherwise return null. 166 BlockAddress *getBlockAddress() const { 167 if (isConstant()) 168 return dyn_cast<BlockAddress>(getConstant()); 169 return nullptr; 170 } 171 172 void markForcedConstant(Constant *V) { 173 assert(isUnknown() && "Can't force a defined value!"); 174 Val.setInt(forcedconstant); 175 Val.setPointer(V); 176 } 177 178 ValueLatticeElement toValueLattice() const { 179 if (isOverdefined()) 180 return ValueLatticeElement::getOverdefined(); 181 if (isConstant()) 182 return ValueLatticeElement::get(getConstant()); 183 return ValueLatticeElement(); 184 } 185 }; 186 187 //===----------------------------------------------------------------------===// 188 // 189 /// SCCPSolver - This class is a general purpose solver for Sparse Conditional 190 /// Constant Propagation. 191 /// 192 class SCCPSolver : public InstVisitor<SCCPSolver> { 193 const DataLayout &DL; 194 const TargetLibraryInfo *TLI; 195 SmallPtrSet<BasicBlock *, 8> BBExecutable; // The BBs that are executable. 196 DenseMap<Value *, LatticeVal> ValueState; // The state each value is in. 197 // The state each parameter is in. 198 DenseMap<Value *, ValueLatticeElement> ParamState; 199 200 /// StructValueState - This maintains ValueState for values that have 201 /// StructType, for example for formal arguments, calls, insertelement, etc. 202 DenseMap<std::pair<Value *, unsigned>, LatticeVal> StructValueState; 203 204 /// GlobalValue - If we are tracking any values for the contents of a global 205 /// variable, we keep a mapping from the constant accessor to the element of 206 /// the global, to the currently known value. If the value becomes 207 /// overdefined, it's entry is simply removed from this map. 208 DenseMap<GlobalVariable *, LatticeVal> TrackedGlobals; 209 210 /// TrackedRetVals - If we are tracking arguments into and the return 211 /// value out of a function, it will have an entry in this map, indicating 212 /// what the known return value for the function is. 213 DenseMap<Function *, LatticeVal> TrackedRetVals; 214 215 /// TrackedMultipleRetVals - Same as TrackedRetVals, but used for functions 216 /// that return multiple values. 217 DenseMap<std::pair<Function *, unsigned>, LatticeVal> TrackedMultipleRetVals; 218 219 /// MRVFunctionsTracked - Each function in TrackedMultipleRetVals is 220 /// represented here for efficient lookup. 221 SmallPtrSet<Function *, 16> MRVFunctionsTracked; 222 223 /// MustTailFunctions - Each function here is a callee of non-removable 224 /// musttail call site. 225 SmallPtrSet<Function *, 16> MustTailCallees; 226 227 /// TrackingIncomingArguments - This is the set of functions for whose 228 /// arguments we make optimistic assumptions about and try to prove as 229 /// constants. 230 SmallPtrSet<Function *, 16> TrackingIncomingArguments; 231 232 /// The reason for two worklists is that overdefined is the lowest state 233 /// on the lattice, and moving things to overdefined as fast as possible 234 /// makes SCCP converge much faster. 235 /// 236 /// By having a separate worklist, we accomplish this because everything 237 /// possibly overdefined will become overdefined at the soonest possible 238 /// point. 239 SmallVector<Value *, 64> OverdefinedInstWorkList; 240 SmallVector<Value *, 64> InstWorkList; 241 242 // The BasicBlock work list 243 SmallVector<BasicBlock *, 64> BBWorkList; 244 245 /// KnownFeasibleEdges - Entries in this set are edges which have already had 246 /// PHI nodes retriggered. 247 using Edge = std::pair<BasicBlock *, BasicBlock *>; 248 DenseSet<Edge> KnownFeasibleEdges; 249 250 DenseMap<Function *, std::unique_ptr<PredicateInfo>> PredInfos; 251 DenseMap<Value *, SmallPtrSet<User *, 2>> AdditionalUsers; 252 253 public: 254 void addPredInfo(Function &F, std::unique_ptr<PredicateInfo> PI) { 255 PredInfos[&F] = std::move(PI); 256 } 257 258 const PredicateBase *getPredicateInfoFor(Instruction *I) { 259 auto PI = PredInfos.find(I->getFunction()); 260 if (PI == PredInfos.end()) 261 return nullptr; 262 return PI->second->getPredicateInfoFor(I); 263 } 264 265 SCCPSolver(const DataLayout &DL, const TargetLibraryInfo *tli) 266 : DL(DL), TLI(tli) {} 267 268 /// MarkBlockExecutable - This method can be used by clients to mark all of 269 /// the blocks that are known to be intrinsically live in the processed unit. 270 /// 271 /// This returns true if the block was not considered live before. 272 bool MarkBlockExecutable(BasicBlock *BB) { 273 if (!BBExecutable.insert(BB).second) 274 return false; 275 LLVM_DEBUG(dbgs() << "Marking Block Executable: " << BB->getName() << '\n'); 276 BBWorkList.push_back(BB); // Add the block to the work list! 277 return true; 278 } 279 280 /// TrackValueOfGlobalVariable - Clients can use this method to 281 /// inform the SCCPSolver that it should track loads and stores to the 282 /// specified global variable if it can. This is only legal to call if 283 /// performing Interprocedural SCCP. 284 void TrackValueOfGlobalVariable(GlobalVariable *GV) { 285 // We only track the contents of scalar globals. 286 if (GV->getValueType()->isSingleValueType()) { 287 LatticeVal &IV = TrackedGlobals[GV]; 288 if (!isa<UndefValue>(GV->getInitializer())) 289 IV.markConstant(GV->getInitializer()); 290 } 291 } 292 293 /// AddTrackedFunction - If the SCCP solver is supposed to track calls into 294 /// and out of the specified function (which cannot have its address taken), 295 /// this method must be called. 296 void AddTrackedFunction(Function *F) { 297 // Add an entry, F -> undef. 298 if (auto *STy = dyn_cast<StructType>(F->getReturnType())) { 299 MRVFunctionsTracked.insert(F); 300 for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) 301 TrackedMultipleRetVals.insert(std::make_pair(std::make_pair(F, i), 302 LatticeVal())); 303 } else 304 TrackedRetVals.insert(std::make_pair(F, LatticeVal())); 305 } 306 307 /// AddMustTailCallee - If the SCCP solver finds that this function is called 308 /// from non-removable musttail call site. 309 void AddMustTailCallee(Function *F) { 310 MustTailCallees.insert(F); 311 } 312 313 /// Returns true if the given function is called from non-removable musttail 314 /// call site. 315 bool isMustTailCallee(Function *F) { 316 return MustTailCallees.count(F); 317 } 318 319 void AddArgumentTrackedFunction(Function *F) { 320 TrackingIncomingArguments.insert(F); 321 } 322 323 /// Returns true if the given function is in the solver's set of 324 /// argument-tracked functions. 325 bool isArgumentTrackedFunction(Function *F) { 326 return TrackingIncomingArguments.count(F); 327 } 328 329 /// Solve - Solve for constants and executable blocks. 330 void Solve(); 331 332 /// ResolvedUndefsIn - While solving the dataflow for a function, we assume 333 /// that branches on undef values cannot reach any of their successors. 334 /// However, this is not a safe assumption. After we solve dataflow, this 335 /// method should be use to handle this. If this returns true, the solver 336 /// should be rerun. 337 bool ResolvedUndefsIn(Function &F); 338 339 bool isBlockExecutable(BasicBlock *BB) const { 340 return BBExecutable.count(BB); 341 } 342 343 // isEdgeFeasible - Return true if the control flow edge from the 'From' basic 344 // block to the 'To' basic block is currently feasible. 345 bool isEdgeFeasible(BasicBlock *From, BasicBlock *To); 346 347 std::vector<LatticeVal> getStructLatticeValueFor(Value *V) const { 348 std::vector<LatticeVal> StructValues; 349 auto *STy = dyn_cast<StructType>(V->getType()); 350 assert(STy && "getStructLatticeValueFor() can be called only on structs"); 351 for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) { 352 auto I = StructValueState.find(std::make_pair(V, i)); 353 assert(I != StructValueState.end() && "Value not in valuemap!"); 354 StructValues.push_back(I->second); 355 } 356 return StructValues; 357 } 358 359 const LatticeVal &getLatticeValueFor(Value *V) const { 360 assert(!V->getType()->isStructTy() && 361 "Should use getStructLatticeValueFor"); 362 DenseMap<Value *, LatticeVal>::const_iterator I = ValueState.find(V); 363 assert(I != ValueState.end() && 364 "V not found in ValueState nor Paramstate map!"); 365 return I->second; 366 } 367 368 /// getTrackedRetVals - Get the inferred return value map. 369 const DenseMap<Function*, LatticeVal> &getTrackedRetVals() { 370 return TrackedRetVals; 371 } 372 373 /// getTrackedGlobals - Get and return the set of inferred initializers for 374 /// global variables. 375 const DenseMap<GlobalVariable*, LatticeVal> &getTrackedGlobals() { 376 return TrackedGlobals; 377 } 378 379 /// getMRVFunctionsTracked - Get the set of functions which return multiple 380 /// values tracked by the pass. 381 const SmallPtrSet<Function *, 16> getMRVFunctionsTracked() { 382 return MRVFunctionsTracked; 383 } 384 385 /// getMustTailCallees - Get the set of functions which are called 386 /// from non-removable musttail call sites. 387 const SmallPtrSet<Function *, 16> getMustTailCallees() { 388 return MustTailCallees; 389 } 390 391 /// markOverdefined - Mark the specified value overdefined. This 392 /// works with both scalars and structs. 393 void markOverdefined(Value *V) { 394 if (auto *STy = dyn_cast<StructType>(V->getType())) 395 for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) 396 markOverdefined(getStructValueState(V, i), V); 397 else 398 markOverdefined(ValueState[V], V); 399 } 400 401 // isStructLatticeConstant - Return true if all the lattice values 402 // corresponding to elements of the structure are not overdefined, 403 // false otherwise. 404 bool isStructLatticeConstant(Function *F, StructType *STy) { 405 for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) { 406 const auto &It = TrackedMultipleRetVals.find(std::make_pair(F, i)); 407 assert(It != TrackedMultipleRetVals.end()); 408 LatticeVal LV = It->second; 409 if (LV.isOverdefined()) 410 return false; 411 } 412 return true; 413 } 414 415 private: 416 // pushToWorkList - Helper for markConstant/markForcedConstant/markOverdefined 417 void pushToWorkList(LatticeVal &IV, Value *V) { 418 if (IV.isOverdefined()) 419 return OverdefinedInstWorkList.push_back(V); 420 InstWorkList.push_back(V); 421 } 422 423 // markConstant - Make a value be marked as "constant". If the value 424 // is not already a constant, add it to the instruction work list so that 425 // the users of the instruction are updated later. 426 bool markConstant(LatticeVal &IV, Value *V, Constant *C) { 427 if (!IV.markConstant(C)) return false; 428 LLVM_DEBUG(dbgs() << "markConstant: " << *C << ": " << *V << '\n'); 429 pushToWorkList(IV, V); 430 return true; 431 } 432 433 bool markConstant(Value *V, Constant *C) { 434 assert(!V->getType()->isStructTy() && "structs should use mergeInValue"); 435 return markConstant(ValueState[V], V, C); 436 } 437 438 void markForcedConstant(Value *V, Constant *C) { 439 assert(!V->getType()->isStructTy() && "structs should use mergeInValue"); 440 LatticeVal &IV = ValueState[V]; 441 IV.markForcedConstant(C); 442 LLVM_DEBUG(dbgs() << "markForcedConstant: " << *C << ": " << *V << '\n'); 443 pushToWorkList(IV, V); 444 } 445 446 // markOverdefined - Make a value be marked as "overdefined". If the 447 // value is not already overdefined, add it to the overdefined instruction 448 // work list so that the users of the instruction are updated later. 449 bool markOverdefined(LatticeVal &IV, Value *V) { 450 if (!IV.markOverdefined()) return false; 451 452 LLVM_DEBUG(dbgs() << "markOverdefined: "; 453 if (auto *F = dyn_cast<Function>(V)) dbgs() 454 << "Function '" << F->getName() << "'\n"; 455 else dbgs() << *V << '\n'); 456 // Only instructions go on the work list 457 pushToWorkList(IV, V); 458 return true; 459 } 460 461 bool mergeInValue(LatticeVal &IV, Value *V, LatticeVal MergeWithV) { 462 if (IV.isOverdefined() || MergeWithV.isUnknown()) 463 return false; // Noop. 464 if (MergeWithV.isOverdefined()) 465 return markOverdefined(IV, V); 466 if (IV.isUnknown()) 467 return markConstant(IV, V, MergeWithV.getConstant()); 468 if (IV.getConstant() != MergeWithV.getConstant()) 469 return markOverdefined(IV, V); 470 return false; 471 } 472 473 bool mergeInValue(Value *V, LatticeVal MergeWithV) { 474 assert(!V->getType()->isStructTy() && 475 "non-structs should use markConstant"); 476 return mergeInValue(ValueState[V], V, MergeWithV); 477 } 478 479 /// getValueState - Return the LatticeVal object that corresponds to the 480 /// value. This function handles the case when the value hasn't been seen yet 481 /// by properly seeding constants etc. 482 LatticeVal &getValueState(Value *V) { 483 assert(!V->getType()->isStructTy() && "Should use getStructValueState"); 484 485 std::pair<DenseMap<Value*, LatticeVal>::iterator, bool> I = 486 ValueState.insert(std::make_pair(V, LatticeVal())); 487 LatticeVal &LV = I.first->second; 488 489 if (!I.second) 490 return LV; // Common case, already in the map. 491 492 if (auto *C = dyn_cast<Constant>(V)) { 493 // Undef values remain unknown. 494 if (!isa<UndefValue>(V)) 495 LV.markConstant(C); // Constants are constant 496 } 497 498 // All others are underdefined by default. 499 return LV; 500 } 501 502 ValueLatticeElement &getParamState(Value *V) { 503 assert(!V->getType()->isStructTy() && "Should use getStructValueState"); 504 505 std::pair<DenseMap<Value*, ValueLatticeElement>::iterator, bool> 506 PI = ParamState.insert(std::make_pair(V, ValueLatticeElement())); 507 ValueLatticeElement &LV = PI.first->second; 508 if (PI.second) 509 LV = getValueState(V).toValueLattice(); 510 511 return LV; 512 } 513 514 /// getStructValueState - Return the LatticeVal object that corresponds to the 515 /// value/field pair. This function handles the case when the value hasn't 516 /// been seen yet by properly seeding constants etc. 517 LatticeVal &getStructValueState(Value *V, unsigned i) { 518 assert(V->getType()->isStructTy() && "Should use getValueState"); 519 assert(i < cast<StructType>(V->getType())->getNumElements() && 520 "Invalid element #"); 521 522 std::pair<DenseMap<std::pair<Value*, unsigned>, LatticeVal>::iterator, 523 bool> I = StructValueState.insert( 524 std::make_pair(std::make_pair(V, i), LatticeVal())); 525 LatticeVal &LV = I.first->second; 526 527 if (!I.second) 528 return LV; // Common case, already in the map. 529 530 if (auto *C = dyn_cast<Constant>(V)) { 531 Constant *Elt = C->getAggregateElement(i); 532 533 if (!Elt) 534 LV.markOverdefined(); // Unknown sort of constant. 535 else if (isa<UndefValue>(Elt)) 536 ; // Undef values remain unknown. 537 else 538 LV.markConstant(Elt); // Constants are constant. 539 } 540 541 // All others are underdefined by default. 542 return LV; 543 } 544 545 /// markEdgeExecutable - Mark a basic block as executable, adding it to the BB 546 /// work list if it is not already executable. 547 bool markEdgeExecutable(BasicBlock *Source, BasicBlock *Dest) { 548 if (!KnownFeasibleEdges.insert(Edge(Source, Dest)).second) 549 return false; // This edge is already known to be executable! 550 551 if (!MarkBlockExecutable(Dest)) { 552 // If the destination is already executable, we just made an *edge* 553 // feasible that wasn't before. Revisit the PHI nodes in the block 554 // because they have potentially new operands. 555 LLVM_DEBUG(dbgs() << "Marking Edge Executable: " << Source->getName() 556 << " -> " << Dest->getName() << '\n'); 557 558 for (PHINode &PN : Dest->phis()) 559 visitPHINode(PN); 560 } 561 return true; 562 } 563 564 // getFeasibleSuccessors - Return a vector of booleans to indicate which 565 // successors are reachable from a given terminator instruction. 566 void getFeasibleSuccessors(TerminatorInst &TI, SmallVectorImpl<bool> &Succs); 567 568 // OperandChangedState - This method is invoked on all of the users of an 569 // instruction that was just changed state somehow. Based on this 570 // information, we need to update the specified user of this instruction. 571 void OperandChangedState(Instruction *I) { 572 if (BBExecutable.count(I->getParent())) // Inst is executable? 573 visit(*I); 574 } 575 576 // Add U as additional user of V. 577 void addAdditionalUser(Value *V, User *U) { 578 auto Iter = AdditionalUsers.insert({V, {}}); 579 Iter.first->second.insert(U); 580 } 581 582 // Mark I's users as changed, including AdditionalUsers. 583 void markUsersAsChanged(Value *I) { 584 for (User *U : I->users()) 585 if (auto *UI = dyn_cast<Instruction>(U)) 586 OperandChangedState(UI); 587 588 auto Iter = AdditionalUsers.find(I); 589 if (Iter != AdditionalUsers.end()) { 590 for (User *U : Iter->second) 591 if (auto *UI = dyn_cast<Instruction>(U)) 592 OperandChangedState(UI); 593 } 594 } 595 596 private: 597 friend class InstVisitor<SCCPSolver>; 598 599 // visit implementations - Something changed in this instruction. Either an 600 // operand made a transition, or the instruction is newly executable. Change 601 // the value type of I to reflect these changes if appropriate. 602 void visitPHINode(PHINode &I); 603 604 // Terminators 605 606 void visitReturnInst(ReturnInst &I); 607 void visitTerminatorInst(TerminatorInst &TI); 608 609 void visitCastInst(CastInst &I); 610 void visitSelectInst(SelectInst &I); 611 void visitBinaryOperator(Instruction &I); 612 void visitCmpInst(CmpInst &I); 613 void visitExtractValueInst(ExtractValueInst &EVI); 614 void visitInsertValueInst(InsertValueInst &IVI); 615 616 void visitCatchSwitchInst(CatchSwitchInst &CPI) { 617 markOverdefined(&CPI); 618 visitTerminatorInst(CPI); 619 } 620 621 // Instructions that cannot be folded away. 622 623 void visitStoreInst (StoreInst &I); 624 void visitLoadInst (LoadInst &I); 625 void visitGetElementPtrInst(GetElementPtrInst &I); 626 627 void visitCallInst (CallInst &I) { 628 visitCallSite(&I); 629 } 630 631 void visitInvokeInst (InvokeInst &II) { 632 visitCallSite(&II); 633 visitTerminatorInst(II); 634 } 635 636 void visitCallSite (CallSite CS); 637 void visitResumeInst (TerminatorInst &I) { /*returns void*/ } 638 void visitUnreachableInst(TerminatorInst &I) { /*returns void*/ } 639 void visitFenceInst (FenceInst &I) { /*returns void*/ } 640 641 void visitInstruction(Instruction &I) { 642 // All the instructions we don't do any special handling for just 643 // go to overdefined. 644 LLVM_DEBUG(dbgs() << "SCCP: Don't know how to handle: " << I << '\n'); 645 markOverdefined(&I); 646 } 647 }; 648 649 } // end anonymous namespace 650 651 // getFeasibleSuccessors - Return a vector of booleans to indicate which 652 // successors are reachable from a given terminator instruction. 653 void SCCPSolver::getFeasibleSuccessors(TerminatorInst &TI, 654 SmallVectorImpl<bool> &Succs) { 655 Succs.resize(TI.getNumSuccessors()); 656 if (auto *BI = dyn_cast<BranchInst>(&TI)) { 657 if (BI->isUnconditional()) { 658 Succs[0] = true; 659 return; 660 } 661 662 LatticeVal BCValue = getValueState(BI->getCondition()); 663 ConstantInt *CI = BCValue.getConstantInt(); 664 if (!CI) { 665 // Overdefined condition variables, and branches on unfoldable constant 666 // conditions, mean the branch could go either way. 667 if (!BCValue.isUnknown()) 668 Succs[0] = Succs[1] = true; 669 return; 670 } 671 672 // Constant condition variables mean the branch can only go a single way. 673 Succs[CI->isZero()] = true; 674 return; 675 } 676 677 // Unwinding instructions successors are always executable. 678 if (TI.isExceptional()) { 679 Succs.assign(TI.getNumSuccessors(), true); 680 return; 681 } 682 683 if (auto *SI = dyn_cast<SwitchInst>(&TI)) { 684 if (!SI->getNumCases()) { 685 Succs[0] = true; 686 return; 687 } 688 LatticeVal SCValue = getValueState(SI->getCondition()); 689 ConstantInt *CI = SCValue.getConstantInt(); 690 691 if (!CI) { // Overdefined or unknown condition? 692 // All destinations are executable! 693 if (!SCValue.isUnknown()) 694 Succs.assign(TI.getNumSuccessors(), true); 695 return; 696 } 697 698 Succs[SI->findCaseValue(CI)->getSuccessorIndex()] = true; 699 return; 700 } 701 702 // In case of indirect branch and its address is a blockaddress, we mark 703 // the target as executable. 704 if (auto *IBR = dyn_cast<IndirectBrInst>(&TI)) { 705 // Casts are folded by visitCastInst. 706 LatticeVal IBRValue = getValueState(IBR->getAddress()); 707 BlockAddress *Addr = IBRValue.getBlockAddress(); 708 if (!Addr) { // Overdefined or unknown condition? 709 // All destinations are executable! 710 if (!IBRValue.isUnknown()) 711 Succs.assign(TI.getNumSuccessors(), true); 712 return; 713 } 714 715 BasicBlock* T = Addr->getBasicBlock(); 716 assert(Addr->getFunction() == T->getParent() && 717 "Block address of a different function ?"); 718 for (unsigned i = 0; i < IBR->getNumSuccessors(); ++i) { 719 // This is the target. 720 if (IBR->getDestination(i) == T) { 721 Succs[i] = true; 722 return; 723 } 724 } 725 726 // If we didn't find our destination in the IBR successor list, then we 727 // have undefined behavior. Its ok to assume no successor is executable. 728 return; 729 } 730 731 LLVM_DEBUG(dbgs() << "Unknown terminator instruction: " << TI << '\n'); 732 llvm_unreachable("SCCP: Don't know how to handle this terminator!"); 733 } 734 735 // isEdgeFeasible - Return true if the control flow edge from the 'From' basic 736 // block to the 'To' basic block is currently feasible. 737 bool SCCPSolver::isEdgeFeasible(BasicBlock *From, BasicBlock *To) { 738 // Check if we've called markEdgeExecutable on the edge yet. (We could 739 // be more aggressive and try to consider edges which haven't been marked 740 // yet, but there isn't any need.) 741 return KnownFeasibleEdges.count(Edge(From, To)); 742 } 743 744 // visit Implementations - Something changed in this instruction, either an 745 // operand made a transition, or the instruction is newly executable. Change 746 // the value type of I to reflect these changes if appropriate. This method 747 // makes sure to do the following actions: 748 // 749 // 1. If a phi node merges two constants in, and has conflicting value coming 750 // from different branches, or if the PHI node merges in an overdefined 751 // value, then the PHI node becomes overdefined. 752 // 2. If a phi node merges only constants in, and they all agree on value, the 753 // PHI node becomes a constant value equal to that. 754 // 3. If V <- x (op) y && isConstant(x) && isConstant(y) V = Constant 755 // 4. If V <- x (op) y && (isOverdefined(x) || isOverdefined(y)) V = Overdefined 756 // 5. If V <- MEM or V <- CALL or V <- (unknown) then V = Overdefined 757 // 6. If a conditional branch has a value that is constant, make the selected 758 // destination executable 759 // 7. If a conditional branch has a value that is overdefined, make all 760 // successors executable. 761 void SCCPSolver::visitPHINode(PHINode &PN) { 762 // If this PN returns a struct, just mark the result overdefined. 763 // TODO: We could do a lot better than this if code actually uses this. 764 if (PN.getType()->isStructTy()) 765 return (void)markOverdefined(&PN); 766 767 if (getValueState(&PN).isOverdefined()) 768 return; // Quick exit 769 770 // Super-extra-high-degree PHI nodes are unlikely to ever be marked constant, 771 // and slow us down a lot. Just mark them overdefined. 772 if (PN.getNumIncomingValues() > 64) 773 return (void)markOverdefined(&PN); 774 775 // Look at all of the executable operands of the PHI node. If any of them 776 // are overdefined, the PHI becomes overdefined as well. If they are all 777 // constant, and they agree with each other, the PHI becomes the identical 778 // constant. If they are constant and don't agree, the PHI is overdefined. 779 // If there are no executable operands, the PHI remains unknown. 780 Constant *OperandVal = nullptr; 781 for (unsigned i = 0, e = PN.getNumIncomingValues(); i != e; ++i) { 782 LatticeVal IV = getValueState(PN.getIncomingValue(i)); 783 if (IV.isUnknown()) continue; // Doesn't influence PHI node. 784 785 if (!isEdgeFeasible(PN.getIncomingBlock(i), PN.getParent())) 786 continue; 787 788 if (IV.isOverdefined()) // PHI node becomes overdefined! 789 return (void)markOverdefined(&PN); 790 791 if (!OperandVal) { // Grab the first value. 792 OperandVal = IV.getConstant(); 793 continue; 794 } 795 796 // There is already a reachable operand. If we conflict with it, 797 // then the PHI node becomes overdefined. If we agree with it, we 798 // can continue on. 799 800 // Check to see if there are two different constants merging, if so, the PHI 801 // node is overdefined. 802 if (IV.getConstant() != OperandVal) 803 return (void)markOverdefined(&PN); 804 } 805 806 // If we exited the loop, this means that the PHI node only has constant 807 // arguments that agree with each other(and OperandVal is the constant) or 808 // OperandVal is null because there are no defined incoming arguments. If 809 // this is the case, the PHI remains unknown. 810 if (OperandVal) 811 markConstant(&PN, OperandVal); // Acquire operand value 812 } 813 814 void SCCPSolver::visitReturnInst(ReturnInst &I) { 815 if (I.getNumOperands() == 0) return; // ret void 816 817 Function *F = I.getParent()->getParent(); 818 Value *ResultOp = I.getOperand(0); 819 820 // If we are tracking the return value of this function, merge it in. 821 if (!TrackedRetVals.empty() && !ResultOp->getType()->isStructTy()) { 822 DenseMap<Function*, LatticeVal>::iterator TFRVI = 823 TrackedRetVals.find(F); 824 if (TFRVI != TrackedRetVals.end()) { 825 mergeInValue(TFRVI->second, F, getValueState(ResultOp)); 826 return; 827 } 828 } 829 830 // Handle functions that return multiple values. 831 if (!TrackedMultipleRetVals.empty()) { 832 if (auto *STy = dyn_cast<StructType>(ResultOp->getType())) 833 if (MRVFunctionsTracked.count(F)) 834 for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) 835 mergeInValue(TrackedMultipleRetVals[std::make_pair(F, i)], F, 836 getStructValueState(ResultOp, i)); 837 } 838 } 839 840 void SCCPSolver::visitTerminatorInst(TerminatorInst &TI) { 841 SmallVector<bool, 16> SuccFeasible; 842 getFeasibleSuccessors(TI, SuccFeasible); 843 844 BasicBlock *BB = TI.getParent(); 845 846 // Mark all feasible successors executable. 847 for (unsigned i = 0, e = SuccFeasible.size(); i != e; ++i) 848 if (SuccFeasible[i]) 849 markEdgeExecutable(BB, TI.getSuccessor(i)); 850 } 851 852 void SCCPSolver::visitCastInst(CastInst &I) { 853 LatticeVal OpSt = getValueState(I.getOperand(0)); 854 if (OpSt.isOverdefined()) // Inherit overdefinedness of operand 855 markOverdefined(&I); 856 else if (OpSt.isConstant()) { 857 // Fold the constant as we build. 858 Constant *C = ConstantFoldCastOperand(I.getOpcode(), OpSt.getConstant(), 859 I.getType(), DL); 860 if (isa<UndefValue>(C)) 861 return; 862 // Propagate constant value 863 markConstant(&I, C); 864 } 865 } 866 867 void SCCPSolver::visitExtractValueInst(ExtractValueInst &EVI) { 868 // If this returns a struct, mark all elements over defined, we don't track 869 // structs in structs. 870 if (EVI.getType()->isStructTy()) 871 return (void)markOverdefined(&EVI); 872 873 // If this is extracting from more than one level of struct, we don't know. 874 if (EVI.getNumIndices() != 1) 875 return (void)markOverdefined(&EVI); 876 877 Value *AggVal = EVI.getAggregateOperand(); 878 if (AggVal->getType()->isStructTy()) { 879 unsigned i = *EVI.idx_begin(); 880 LatticeVal EltVal = getStructValueState(AggVal, i); 881 mergeInValue(getValueState(&EVI), &EVI, EltVal); 882 } else { 883 // Otherwise, must be extracting from an array. 884 return (void)markOverdefined(&EVI); 885 } 886 } 887 888 void SCCPSolver::visitInsertValueInst(InsertValueInst &IVI) { 889 auto *STy = dyn_cast<StructType>(IVI.getType()); 890 if (!STy) 891 return (void)markOverdefined(&IVI); 892 893 // If this has more than one index, we can't handle it, drive all results to 894 // undef. 895 if (IVI.getNumIndices() != 1) 896 return (void)markOverdefined(&IVI); 897 898 Value *Aggr = IVI.getAggregateOperand(); 899 unsigned Idx = *IVI.idx_begin(); 900 901 // Compute the result based on what we're inserting. 902 for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) { 903 // This passes through all values that aren't the inserted element. 904 if (i != Idx) { 905 LatticeVal EltVal = getStructValueState(Aggr, i); 906 mergeInValue(getStructValueState(&IVI, i), &IVI, EltVal); 907 continue; 908 } 909 910 Value *Val = IVI.getInsertedValueOperand(); 911 if (Val->getType()->isStructTy()) 912 // We don't track structs in structs. 913 markOverdefined(getStructValueState(&IVI, i), &IVI); 914 else { 915 LatticeVal InVal = getValueState(Val); 916 mergeInValue(getStructValueState(&IVI, i), &IVI, InVal); 917 } 918 } 919 } 920 921 void SCCPSolver::visitSelectInst(SelectInst &I) { 922 // If this select returns a struct, just mark the result overdefined. 923 // TODO: We could do a lot better than this if code actually uses this. 924 if (I.getType()->isStructTy()) 925 return (void)markOverdefined(&I); 926 927 LatticeVal CondValue = getValueState(I.getCondition()); 928 if (CondValue.isUnknown()) 929 return; 930 931 if (ConstantInt *CondCB = CondValue.getConstantInt()) { 932 Value *OpVal = CondCB->isZero() ? I.getFalseValue() : I.getTrueValue(); 933 mergeInValue(&I, getValueState(OpVal)); 934 return; 935 } 936 937 // Otherwise, the condition is overdefined or a constant we can't evaluate. 938 // See if we can produce something better than overdefined based on the T/F 939 // value. 940 LatticeVal TVal = getValueState(I.getTrueValue()); 941 LatticeVal FVal = getValueState(I.getFalseValue()); 942 943 // select ?, C, C -> C. 944 if (TVal.isConstant() && FVal.isConstant() && 945 TVal.getConstant() == FVal.getConstant()) 946 return (void)markConstant(&I, FVal.getConstant()); 947 948 if (TVal.isUnknown()) // select ?, undef, X -> X. 949 return (void)mergeInValue(&I, FVal); 950 if (FVal.isUnknown()) // select ?, X, undef -> X. 951 return (void)mergeInValue(&I, TVal); 952 markOverdefined(&I); 953 } 954 955 // Handle Binary Operators. 956 void SCCPSolver::visitBinaryOperator(Instruction &I) { 957 LatticeVal V1State = getValueState(I.getOperand(0)); 958 LatticeVal V2State = getValueState(I.getOperand(1)); 959 960 LatticeVal &IV = ValueState[&I]; 961 if (IV.isOverdefined()) return; 962 963 if (V1State.isConstant() && V2State.isConstant()) { 964 Constant *C = ConstantExpr::get(I.getOpcode(), V1State.getConstant(), 965 V2State.getConstant()); 966 // X op Y -> undef. 967 if (isa<UndefValue>(C)) 968 return; 969 return (void)markConstant(IV, &I, C); 970 } 971 972 // If something is undef, wait for it to resolve. 973 if (!V1State.isOverdefined() && !V2State.isOverdefined()) 974 return; 975 976 // Otherwise, one of our operands is overdefined. Try to produce something 977 // better than overdefined with some tricks. 978 // If this is 0 / Y, it doesn't matter that the second operand is 979 // overdefined, and we can replace it with zero. 980 if (I.getOpcode() == Instruction::UDiv || I.getOpcode() == Instruction::SDiv) 981 if (V1State.isConstant() && V1State.getConstant()->isNullValue()) 982 return (void)markConstant(IV, &I, V1State.getConstant()); 983 984 // If this is: 985 // -> AND/MUL with 0 986 // -> OR with -1 987 // it doesn't matter that the other operand is overdefined. 988 if (I.getOpcode() == Instruction::And || I.getOpcode() == Instruction::Mul || 989 I.getOpcode() == Instruction::Or) { 990 LatticeVal *NonOverdefVal = nullptr; 991 if (!V1State.isOverdefined()) 992 NonOverdefVal = &V1State; 993 else if (!V2State.isOverdefined()) 994 NonOverdefVal = &V2State; 995 996 if (NonOverdefVal) { 997 if (NonOverdefVal->isUnknown()) 998 return; 999 1000 if (I.getOpcode() == Instruction::And || 1001 I.getOpcode() == Instruction::Mul) { 1002 // X and 0 = 0 1003 // X * 0 = 0 1004 if (NonOverdefVal->getConstant()->isNullValue()) 1005 return (void)markConstant(IV, &I, NonOverdefVal->getConstant()); 1006 } else { 1007 // X or -1 = -1 1008 if (ConstantInt *CI = NonOverdefVal->getConstantInt()) 1009 if (CI->isMinusOne()) 1010 return (void)markConstant(IV, &I, NonOverdefVal->getConstant()); 1011 } 1012 } 1013 } 1014 1015 markOverdefined(&I); 1016 } 1017 1018 // Handle ICmpInst instruction. 1019 void SCCPSolver::visitCmpInst(CmpInst &I) { 1020 LatticeVal &IV = ValueState[&I]; 1021 if (IV.isOverdefined()) return; 1022 1023 Value *Op1 = I.getOperand(0); 1024 Value *Op2 = I.getOperand(1); 1025 1026 // For parameters, use ParamState which includes constant range info if 1027 // available. 1028 auto V1Param = ParamState.find(Op1); 1029 ValueLatticeElement V1State = (V1Param != ParamState.end()) 1030 ? V1Param->second 1031 : getValueState(Op1).toValueLattice(); 1032 1033 auto V2Param = ParamState.find(Op2); 1034 ValueLatticeElement V2State = V2Param != ParamState.end() 1035 ? V2Param->second 1036 : getValueState(Op2).toValueLattice(); 1037 1038 Constant *C = V1State.getCompare(I.getPredicate(), I.getType(), V2State); 1039 if (C) { 1040 if (isa<UndefValue>(C)) 1041 return; 1042 LatticeVal CV; 1043 CV.markConstant(C); 1044 mergeInValue(&I, CV); 1045 return; 1046 } 1047 1048 // If operands are still unknown, wait for it to resolve. 1049 if (!V1State.isOverdefined() && !V2State.isOverdefined() && !IV.isConstant()) 1050 return; 1051 1052 markOverdefined(&I); 1053 } 1054 1055 // Handle getelementptr instructions. If all operands are constants then we 1056 // can turn this into a getelementptr ConstantExpr. 1057 void SCCPSolver::visitGetElementPtrInst(GetElementPtrInst &I) { 1058 if (ValueState[&I].isOverdefined()) return; 1059 1060 SmallVector<Constant*, 8> Operands; 1061 Operands.reserve(I.getNumOperands()); 1062 1063 for (unsigned i = 0, e = I.getNumOperands(); i != e; ++i) { 1064 LatticeVal State = getValueState(I.getOperand(i)); 1065 if (State.isUnknown()) 1066 return; // Operands are not resolved yet. 1067 1068 if (State.isOverdefined()) 1069 return (void)markOverdefined(&I); 1070 1071 assert(State.isConstant() && "Unknown state!"); 1072 Operands.push_back(State.getConstant()); 1073 } 1074 1075 Constant *Ptr = Operands[0]; 1076 auto Indices = makeArrayRef(Operands.begin() + 1, Operands.end()); 1077 Constant *C = 1078 ConstantExpr::getGetElementPtr(I.getSourceElementType(), Ptr, Indices); 1079 if (isa<UndefValue>(C)) 1080 return; 1081 markConstant(&I, C); 1082 } 1083 1084 void SCCPSolver::visitStoreInst(StoreInst &SI) { 1085 // If this store is of a struct, ignore it. 1086 if (SI.getOperand(0)->getType()->isStructTy()) 1087 return; 1088 1089 if (TrackedGlobals.empty() || !isa<GlobalVariable>(SI.getOperand(1))) 1090 return; 1091 1092 GlobalVariable *GV = cast<GlobalVariable>(SI.getOperand(1)); 1093 DenseMap<GlobalVariable*, LatticeVal>::iterator I = TrackedGlobals.find(GV); 1094 if (I == TrackedGlobals.end() || I->second.isOverdefined()) return; 1095 1096 // Get the value we are storing into the global, then merge it. 1097 mergeInValue(I->second, GV, getValueState(SI.getOperand(0))); 1098 if (I->second.isOverdefined()) 1099 TrackedGlobals.erase(I); // No need to keep tracking this! 1100 } 1101 1102 // Handle load instructions. If the operand is a constant pointer to a constant 1103 // global, we can replace the load with the loaded constant value! 1104 void SCCPSolver::visitLoadInst(LoadInst &I) { 1105 // If this load is of a struct, just mark the result overdefined. 1106 if (I.getType()->isStructTy()) 1107 return (void)markOverdefined(&I); 1108 1109 LatticeVal PtrVal = getValueState(I.getOperand(0)); 1110 if (PtrVal.isUnknown()) return; // The pointer is not resolved yet! 1111 1112 LatticeVal &IV = ValueState[&I]; 1113 if (IV.isOverdefined()) return; 1114 1115 if (!PtrVal.isConstant() || I.isVolatile()) 1116 return (void)markOverdefined(IV, &I); 1117 1118 Constant *Ptr = PtrVal.getConstant(); 1119 1120 // load null is undefined. 1121 if (isa<ConstantPointerNull>(Ptr)) { 1122 if (NullPointerIsDefined(I.getFunction(), I.getPointerAddressSpace())) 1123 return (void)markOverdefined(IV, &I); 1124 else 1125 return; 1126 } 1127 1128 // Transform load (constant global) into the value loaded. 1129 if (auto *GV = dyn_cast<GlobalVariable>(Ptr)) { 1130 if (!TrackedGlobals.empty()) { 1131 // If we are tracking this global, merge in the known value for it. 1132 DenseMap<GlobalVariable*, LatticeVal>::iterator It = 1133 TrackedGlobals.find(GV); 1134 if (It != TrackedGlobals.end()) { 1135 mergeInValue(IV, &I, It->second); 1136 return; 1137 } 1138 } 1139 } 1140 1141 // Transform load from a constant into a constant if possible. 1142 if (Constant *C = ConstantFoldLoadFromConstPtr(Ptr, I.getType(), DL)) { 1143 if (isa<UndefValue>(C)) 1144 return; 1145 return (void)markConstant(IV, &I, C); 1146 } 1147 1148 // Otherwise we cannot say for certain what value this load will produce. 1149 // Bail out. 1150 markOverdefined(IV, &I); 1151 } 1152 1153 void SCCPSolver::visitCallSite(CallSite CS) { 1154 Function *F = CS.getCalledFunction(); 1155 Instruction *I = CS.getInstruction(); 1156 1157 if (auto *II = dyn_cast<IntrinsicInst>(I)) { 1158 if (II->getIntrinsicID() == Intrinsic::ssa_copy) { 1159 if (ValueState[I].isOverdefined()) 1160 return; 1161 1162 auto *PI = getPredicateInfoFor(I); 1163 if (!PI) 1164 return; 1165 1166 auto *PBranch = dyn_cast<PredicateBranch>(getPredicateInfoFor(I)); 1167 if (!PBranch) { 1168 mergeInValue(ValueState[I], I, getValueState(PI->OriginalOp)); 1169 return; 1170 } 1171 1172 Value *CopyOf = I->getOperand(0); 1173 Value *Cond = PBranch->Condition; 1174 1175 // Everything below relies on the condition being a comparison. 1176 auto *Cmp = dyn_cast<CmpInst>(Cond); 1177 if (!Cmp) { 1178 mergeInValue(ValueState[I], I, getValueState(PI->OriginalOp)); 1179 return; 1180 } 1181 1182 Value *CmpOp0 = Cmp->getOperand(0); 1183 Value *CmpOp1 = Cmp->getOperand(1); 1184 if (CopyOf != CmpOp0 && CopyOf != CmpOp1) { 1185 mergeInValue(ValueState[I], I, getValueState(PI->OriginalOp)); 1186 return; 1187 } 1188 1189 if (CmpOp0 != CopyOf) 1190 std::swap(CmpOp0, CmpOp1); 1191 1192 LatticeVal OriginalVal = getValueState(CopyOf); 1193 LatticeVal EqVal = getValueState(CmpOp1); 1194 LatticeVal &IV = ValueState[I]; 1195 if (PBranch->TrueEdge && Cmp->getPredicate() == CmpInst::ICMP_EQ) { 1196 addAdditionalUser(CmpOp1, I); 1197 if (OriginalVal.isConstant()) 1198 mergeInValue(IV, I, OriginalVal); 1199 else 1200 mergeInValue(IV, I, EqVal); 1201 return; 1202 } 1203 if (!PBranch->TrueEdge && Cmp->getPredicate() == CmpInst::ICMP_NE) { 1204 addAdditionalUser(CmpOp1, I); 1205 if (OriginalVal.isConstant()) 1206 mergeInValue(IV, I, OriginalVal); 1207 else 1208 mergeInValue(IV, I, EqVal); 1209 return; 1210 } 1211 1212 return (void)mergeInValue(IV, I, getValueState(PBranch->OriginalOp)); 1213 } 1214 } 1215 1216 // The common case is that we aren't tracking the callee, either because we 1217 // are not doing interprocedural analysis or the callee is indirect, or is 1218 // external. Handle these cases first. 1219 if (!F || F->isDeclaration()) { 1220 CallOverdefined: 1221 // Void return and not tracking callee, just bail. 1222 if (I->getType()->isVoidTy()) return; 1223 1224 // Otherwise, if we have a single return value case, and if the function is 1225 // a declaration, maybe we can constant fold it. 1226 if (F && F->isDeclaration() && !I->getType()->isStructTy() && 1227 canConstantFoldCallTo(CS, F)) { 1228 SmallVector<Constant*, 8> Operands; 1229 for (CallSite::arg_iterator AI = CS.arg_begin(), E = CS.arg_end(); 1230 AI != E; ++AI) { 1231 LatticeVal State = getValueState(*AI); 1232 1233 if (State.isUnknown()) 1234 return; // Operands are not resolved yet. 1235 if (State.isOverdefined()) 1236 return (void)markOverdefined(I); 1237 assert(State.isConstant() && "Unknown state!"); 1238 Operands.push_back(State.getConstant()); 1239 } 1240 1241 if (getValueState(I).isOverdefined()) 1242 return; 1243 1244 // If we can constant fold this, mark the result of the call as a 1245 // constant. 1246 if (Constant *C = ConstantFoldCall(CS, F, Operands, TLI)) { 1247 // call -> undef. 1248 if (isa<UndefValue>(C)) 1249 return; 1250 return (void)markConstant(I, C); 1251 } 1252 } 1253 1254 // Otherwise, we don't know anything about this call, mark it overdefined. 1255 return (void)markOverdefined(I); 1256 } 1257 1258 // If this is a local function that doesn't have its address taken, mark its 1259 // entry block executable and merge in the actual arguments to the call into 1260 // the formal arguments of the function. 1261 if (!TrackingIncomingArguments.empty() && TrackingIncomingArguments.count(F)){ 1262 MarkBlockExecutable(&F->front()); 1263 1264 // Propagate information from this call site into the callee. 1265 CallSite::arg_iterator CAI = CS.arg_begin(); 1266 for (Function::arg_iterator AI = F->arg_begin(), E = F->arg_end(); 1267 AI != E; ++AI, ++CAI) { 1268 // If this argument is byval, and if the function is not readonly, there 1269 // will be an implicit copy formed of the input aggregate. 1270 if (AI->hasByValAttr() && !F->onlyReadsMemory()) { 1271 markOverdefined(&*AI); 1272 continue; 1273 } 1274 1275 if (auto *STy = dyn_cast<StructType>(AI->getType())) { 1276 for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) { 1277 LatticeVal CallArg = getStructValueState(*CAI, i); 1278 mergeInValue(getStructValueState(&*AI, i), &*AI, CallArg); 1279 } 1280 } else { 1281 // Most other parts of the Solver still only use the simpler value 1282 // lattice, so we propagate changes for parameters to both lattices. 1283 LatticeVal ConcreteArgument = getValueState(*CAI); 1284 bool ParamChanged = 1285 getParamState(&*AI).mergeIn(ConcreteArgument.toValueLattice(), DL); 1286 bool ValueChanged = mergeInValue(&*AI, ConcreteArgument); 1287 // Add argument to work list, if the state of a parameter changes but 1288 // ValueState does not change (because it is already overdefined there), 1289 // We have to take changes in ParamState into account, as it is used 1290 // when evaluating Cmp instructions. 1291 if (!ValueChanged && ParamChanged) 1292 pushToWorkList(ValueState[&*AI], &*AI); 1293 } 1294 } 1295 } 1296 1297 // If this is a single/zero retval case, see if we're tracking the function. 1298 if (auto *STy = dyn_cast<StructType>(F->getReturnType())) { 1299 if (!MRVFunctionsTracked.count(F)) 1300 goto CallOverdefined; // Not tracking this callee. 1301 1302 // If we are tracking this callee, propagate the result of the function 1303 // into this call site. 1304 for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) 1305 mergeInValue(getStructValueState(I, i), I, 1306 TrackedMultipleRetVals[std::make_pair(F, i)]); 1307 } else { 1308 DenseMap<Function*, LatticeVal>::iterator TFRVI = TrackedRetVals.find(F); 1309 if (TFRVI == TrackedRetVals.end()) 1310 goto CallOverdefined; // Not tracking this callee. 1311 1312 // If so, propagate the return value of the callee into this call result. 1313 mergeInValue(I, TFRVI->second); 1314 } 1315 } 1316 1317 void SCCPSolver::Solve() { 1318 // Process the work lists until they are empty! 1319 while (!BBWorkList.empty() || !InstWorkList.empty() || 1320 !OverdefinedInstWorkList.empty()) { 1321 // Process the overdefined instruction's work list first, which drives other 1322 // things to overdefined more quickly. 1323 while (!OverdefinedInstWorkList.empty()) { 1324 Value *I = OverdefinedInstWorkList.pop_back_val(); 1325 1326 LLVM_DEBUG(dbgs() << "\nPopped off OI-WL: " << *I << '\n'); 1327 1328 // "I" got into the work list because it either made the transition from 1329 // bottom to constant, or to overdefined. 1330 // 1331 // Anything on this worklist that is overdefined need not be visited 1332 // since all of its users will have already been marked as overdefined 1333 // Update all of the users of this instruction's value. 1334 // 1335 markUsersAsChanged(I); 1336 } 1337 1338 // Process the instruction work list. 1339 while (!InstWorkList.empty()) { 1340 Value *I = InstWorkList.pop_back_val(); 1341 1342 LLVM_DEBUG(dbgs() << "\nPopped off I-WL: " << *I << '\n'); 1343 1344 // "I" got into the work list because it made the transition from undef to 1345 // constant. 1346 // 1347 // Anything on this worklist that is overdefined need not be visited 1348 // since all of its users will have already been marked as overdefined. 1349 // Update all of the users of this instruction's value. 1350 // 1351 if (I->getType()->isStructTy() || !getValueState(I).isOverdefined()) 1352 markUsersAsChanged(I); 1353 } 1354 1355 // Process the basic block work list. 1356 while (!BBWorkList.empty()) { 1357 BasicBlock *BB = BBWorkList.back(); 1358 BBWorkList.pop_back(); 1359 1360 LLVM_DEBUG(dbgs() << "\nPopped off BBWL: " << *BB << '\n'); 1361 1362 // Notify all instructions in this basic block that they are newly 1363 // executable. 1364 visit(BB); 1365 } 1366 } 1367 } 1368 1369 /// ResolvedUndefsIn - While solving the dataflow for a function, we assume 1370 /// that branches on undef values cannot reach any of their successors. 1371 /// However, this is not a safe assumption. After we solve dataflow, this 1372 /// method should be use to handle this. If this returns true, the solver 1373 /// should be rerun. 1374 /// 1375 /// This method handles this by finding an unresolved branch and marking it one 1376 /// of the edges from the block as being feasible, even though the condition 1377 /// doesn't say it would otherwise be. This allows SCCP to find the rest of the 1378 /// CFG and only slightly pessimizes the analysis results (by marking one, 1379 /// potentially infeasible, edge feasible). This cannot usefully modify the 1380 /// constraints on the condition of the branch, as that would impact other users 1381 /// of the value. 1382 /// 1383 /// This scan also checks for values that use undefs, whose results are actually 1384 /// defined. For example, 'zext i8 undef to i32' should produce all zeros 1385 /// conservatively, as "(zext i8 X -> i32) & 0xFF00" must always return zero, 1386 /// even if X isn't defined. 1387 bool SCCPSolver::ResolvedUndefsIn(Function &F) { 1388 for (BasicBlock &BB : F) { 1389 if (!BBExecutable.count(&BB)) 1390 continue; 1391 1392 for (Instruction &I : BB) { 1393 // Look for instructions which produce undef values. 1394 if (I.getType()->isVoidTy()) continue; 1395 1396 if (auto *STy = dyn_cast<StructType>(I.getType())) { 1397 // Only a few things that can be structs matter for undef. 1398 1399 // Tracked calls must never be marked overdefined in ResolvedUndefsIn. 1400 if (CallSite CS = CallSite(&I)) 1401 if (Function *F = CS.getCalledFunction()) 1402 if (MRVFunctionsTracked.count(F)) 1403 continue; 1404 1405 // extractvalue and insertvalue don't need to be marked; they are 1406 // tracked as precisely as their operands. 1407 if (isa<ExtractValueInst>(I) || isa<InsertValueInst>(I)) 1408 continue; 1409 1410 // Send the results of everything else to overdefined. We could be 1411 // more precise than this but it isn't worth bothering. 1412 for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) { 1413 LatticeVal &LV = getStructValueState(&I, i); 1414 if (LV.isUnknown()) 1415 markOverdefined(LV, &I); 1416 } 1417 continue; 1418 } 1419 1420 LatticeVal &LV = getValueState(&I); 1421 if (!LV.isUnknown()) continue; 1422 1423 // extractvalue is safe; check here because the argument is a struct. 1424 if (isa<ExtractValueInst>(I)) 1425 continue; 1426 1427 // Compute the operand LatticeVals, for convenience below. 1428 // Anything taking a struct is conservatively assumed to require 1429 // overdefined markings. 1430 if (I.getOperand(0)->getType()->isStructTy()) { 1431 markOverdefined(&I); 1432 return true; 1433 } 1434 LatticeVal Op0LV = getValueState(I.getOperand(0)); 1435 LatticeVal Op1LV; 1436 if (I.getNumOperands() == 2) { 1437 if (I.getOperand(1)->getType()->isStructTy()) { 1438 markOverdefined(&I); 1439 return true; 1440 } 1441 1442 Op1LV = getValueState(I.getOperand(1)); 1443 } 1444 // If this is an instructions whose result is defined even if the input is 1445 // not fully defined, propagate the information. 1446 Type *ITy = I.getType(); 1447 switch (I.getOpcode()) { 1448 case Instruction::Add: 1449 case Instruction::Sub: 1450 case Instruction::Trunc: 1451 case Instruction::FPTrunc: 1452 case Instruction::BitCast: 1453 break; // Any undef -> undef 1454 case Instruction::FSub: 1455 case Instruction::FAdd: 1456 case Instruction::FMul: 1457 case Instruction::FDiv: 1458 case Instruction::FRem: 1459 // Floating-point binary operation: be conservative. 1460 if (Op0LV.isUnknown() && Op1LV.isUnknown()) 1461 markForcedConstant(&I, Constant::getNullValue(ITy)); 1462 else 1463 markOverdefined(&I); 1464 return true; 1465 case Instruction::ZExt: 1466 case Instruction::SExt: 1467 case Instruction::FPToUI: 1468 case Instruction::FPToSI: 1469 case Instruction::FPExt: 1470 case Instruction::PtrToInt: 1471 case Instruction::IntToPtr: 1472 case Instruction::SIToFP: 1473 case Instruction::UIToFP: 1474 // undef -> 0; some outputs are impossible 1475 markForcedConstant(&I, Constant::getNullValue(ITy)); 1476 return true; 1477 case Instruction::Mul: 1478 case Instruction::And: 1479 // Both operands undef -> undef 1480 if (Op0LV.isUnknown() && Op1LV.isUnknown()) 1481 break; 1482 // undef * X -> 0. X could be zero. 1483 // undef & X -> 0. X could be zero. 1484 markForcedConstant(&I, Constant::getNullValue(ITy)); 1485 return true; 1486 case Instruction::Or: 1487 // Both operands undef -> undef 1488 if (Op0LV.isUnknown() && Op1LV.isUnknown()) 1489 break; 1490 // undef | X -> -1. X could be -1. 1491 markForcedConstant(&I, Constant::getAllOnesValue(ITy)); 1492 return true; 1493 case Instruction::Xor: 1494 // undef ^ undef -> 0; strictly speaking, this is not strictly 1495 // necessary, but we try to be nice to people who expect this 1496 // behavior in simple cases 1497 if (Op0LV.isUnknown() && Op1LV.isUnknown()) { 1498 markForcedConstant(&I, Constant::getNullValue(ITy)); 1499 return true; 1500 } 1501 // undef ^ X -> undef 1502 break; 1503 case Instruction::SDiv: 1504 case Instruction::UDiv: 1505 case Instruction::SRem: 1506 case Instruction::URem: 1507 // X / undef -> undef. No change. 1508 // X % undef -> undef. No change. 1509 if (Op1LV.isUnknown()) break; 1510 1511 // X / 0 -> undef. No change. 1512 // X % 0 -> undef. No change. 1513 if (Op1LV.isConstant() && Op1LV.getConstant()->isZeroValue()) 1514 break; 1515 1516 // undef / X -> 0. X could be maxint. 1517 // undef % X -> 0. X could be 1. 1518 markForcedConstant(&I, Constant::getNullValue(ITy)); 1519 return true; 1520 case Instruction::AShr: 1521 // X >>a undef -> undef. 1522 if (Op1LV.isUnknown()) break; 1523 1524 // Shifting by the bitwidth or more is undefined. 1525 if (Op1LV.isConstant()) { 1526 if (auto *ShiftAmt = Op1LV.getConstantInt()) 1527 if (ShiftAmt->getLimitedValue() >= 1528 ShiftAmt->getType()->getScalarSizeInBits()) 1529 break; 1530 } 1531 1532 // undef >>a X -> 0 1533 markForcedConstant(&I, Constant::getNullValue(ITy)); 1534 return true; 1535 case Instruction::LShr: 1536 case Instruction::Shl: 1537 // X << undef -> undef. 1538 // X >> undef -> undef. 1539 if (Op1LV.isUnknown()) break; 1540 1541 // Shifting by the bitwidth or more is undefined. 1542 if (Op1LV.isConstant()) { 1543 if (auto *ShiftAmt = Op1LV.getConstantInt()) 1544 if (ShiftAmt->getLimitedValue() >= 1545 ShiftAmt->getType()->getScalarSizeInBits()) 1546 break; 1547 } 1548 1549 // undef << X -> 0 1550 // undef >> X -> 0 1551 markForcedConstant(&I, Constant::getNullValue(ITy)); 1552 return true; 1553 case Instruction::Select: 1554 Op1LV = getValueState(I.getOperand(1)); 1555 // undef ? X : Y -> X or Y. There could be commonality between X/Y. 1556 if (Op0LV.isUnknown()) { 1557 if (!Op1LV.isConstant()) // Pick the constant one if there is any. 1558 Op1LV = getValueState(I.getOperand(2)); 1559 } else if (Op1LV.isUnknown()) { 1560 // c ? undef : undef -> undef. No change. 1561 Op1LV = getValueState(I.getOperand(2)); 1562 if (Op1LV.isUnknown()) 1563 break; 1564 // Otherwise, c ? undef : x -> x. 1565 } else { 1566 // Leave Op1LV as Operand(1)'s LatticeValue. 1567 } 1568 1569 if (Op1LV.isConstant()) 1570 markForcedConstant(&I, Op1LV.getConstant()); 1571 else 1572 markOverdefined(&I); 1573 return true; 1574 case Instruction::Load: 1575 // A load here means one of two things: a load of undef from a global, 1576 // a load from an unknown pointer. Either way, having it return undef 1577 // is okay. 1578 break; 1579 case Instruction::ICmp: 1580 // X == undef -> undef. Other comparisons get more complicated. 1581 Op0LV = getValueState(I.getOperand(0)); 1582 Op1LV = getValueState(I.getOperand(1)); 1583 1584 if ((Op0LV.isUnknown() || Op1LV.isUnknown()) && 1585 cast<ICmpInst>(&I)->isEquality()) 1586 break; 1587 markOverdefined(&I); 1588 return true; 1589 case Instruction::Call: 1590 case Instruction::Invoke: 1591 // There are two reasons a call can have an undef result 1592 // 1. It could be tracked. 1593 // 2. It could be constant-foldable. 1594 // Because of the way we solve return values, tracked calls must 1595 // never be marked overdefined in ResolvedUndefsIn. 1596 if (Function *F = CallSite(&I).getCalledFunction()) 1597 if (TrackedRetVals.count(F)) 1598 break; 1599 1600 // If the call is constant-foldable, we mark it overdefined because 1601 // we do not know what return values are valid. 1602 markOverdefined(&I); 1603 return true; 1604 default: 1605 // If we don't know what should happen here, conservatively mark it 1606 // overdefined. 1607 markOverdefined(&I); 1608 return true; 1609 } 1610 } 1611 1612 // Check to see if we have a branch or switch on an undefined value. If so 1613 // we force the branch to go one way or the other to make the successor 1614 // values live. It doesn't really matter which way we force it. 1615 TerminatorInst *TI = BB.getTerminator(); 1616 if (auto *BI = dyn_cast<BranchInst>(TI)) { 1617 if (!BI->isConditional()) continue; 1618 if (!getValueState(BI->getCondition()).isUnknown()) 1619 continue; 1620 1621 // If the input to SCCP is actually branch on undef, fix the undef to 1622 // false. 1623 if (isa<UndefValue>(BI->getCondition())) { 1624 BI->setCondition(ConstantInt::getFalse(BI->getContext())); 1625 markEdgeExecutable(&BB, TI->getSuccessor(1)); 1626 return true; 1627 } 1628 1629 // Otherwise, it is a branch on a symbolic value which is currently 1630 // considered to be undef. Make sure some edge is executable, so a 1631 // branch on "undef" always flows somewhere. 1632 // FIXME: Distinguish between dead code and an LLVM "undef" value. 1633 BasicBlock *DefaultSuccessor = TI->getSuccessor(1); 1634 if (markEdgeExecutable(&BB, DefaultSuccessor)) 1635 return true; 1636 1637 continue; 1638 } 1639 1640 if (auto *IBR = dyn_cast<IndirectBrInst>(TI)) { 1641 // Indirect branch with no successor ?. Its ok to assume it branches 1642 // to no target. 1643 if (IBR->getNumSuccessors() < 1) 1644 continue; 1645 1646 if (!getValueState(IBR->getAddress()).isUnknown()) 1647 continue; 1648 1649 // If the input to SCCP is actually branch on undef, fix the undef to 1650 // the first successor of the indirect branch. 1651 if (isa<UndefValue>(IBR->getAddress())) { 1652 IBR->setAddress(BlockAddress::get(IBR->getSuccessor(0))); 1653 markEdgeExecutable(&BB, IBR->getSuccessor(0)); 1654 return true; 1655 } 1656 1657 // Otherwise, it is a branch on a symbolic value which is currently 1658 // considered to be undef. Make sure some edge is executable, so a 1659 // branch on "undef" always flows somewhere. 1660 // FIXME: IndirectBr on "undef" doesn't actually need to go anywhere: 1661 // we can assume the branch has undefined behavior instead. 1662 BasicBlock *DefaultSuccessor = IBR->getSuccessor(0); 1663 if (markEdgeExecutable(&BB, DefaultSuccessor)) 1664 return true; 1665 1666 continue; 1667 } 1668 1669 if (auto *SI = dyn_cast<SwitchInst>(TI)) { 1670 if (!SI->getNumCases() || !getValueState(SI->getCondition()).isUnknown()) 1671 continue; 1672 1673 // If the input to SCCP is actually switch on undef, fix the undef to 1674 // the first constant. 1675 if (isa<UndefValue>(SI->getCondition())) { 1676 SI->setCondition(SI->case_begin()->getCaseValue()); 1677 markEdgeExecutable(&BB, SI->case_begin()->getCaseSuccessor()); 1678 return true; 1679 } 1680 1681 // Otherwise, it is a branch on a symbolic value which is currently 1682 // considered to be undef. Make sure some edge is executable, so a 1683 // branch on "undef" always flows somewhere. 1684 // FIXME: Distinguish between dead code and an LLVM "undef" value. 1685 BasicBlock *DefaultSuccessor = SI->case_begin()->getCaseSuccessor(); 1686 if (markEdgeExecutable(&BB, DefaultSuccessor)) 1687 return true; 1688 1689 continue; 1690 } 1691 } 1692 1693 return false; 1694 } 1695 1696 static bool tryToReplaceWithConstant(SCCPSolver &Solver, Value *V) { 1697 Constant *Const = nullptr; 1698 if (V->getType()->isStructTy()) { 1699 std::vector<LatticeVal> IVs = Solver.getStructLatticeValueFor(V); 1700 if (llvm::any_of(IVs, 1701 [](const LatticeVal &LV) { return LV.isOverdefined(); })) 1702 return false; 1703 std::vector<Constant *> ConstVals; 1704 auto *ST = dyn_cast<StructType>(V->getType()); 1705 for (unsigned i = 0, e = ST->getNumElements(); i != e; ++i) { 1706 LatticeVal V = IVs[i]; 1707 ConstVals.push_back(V.isConstant() 1708 ? V.getConstant() 1709 : UndefValue::get(ST->getElementType(i))); 1710 } 1711 Const = ConstantStruct::get(ST, ConstVals); 1712 } else { 1713 const LatticeVal &IV = Solver.getLatticeValueFor(V); 1714 if (IV.isOverdefined()) 1715 return false; 1716 1717 Const = IV.isConstant() ? IV.getConstant() : UndefValue::get(V->getType()); 1718 } 1719 assert(Const && "Constant is nullptr here!"); 1720 1721 // Replacing `musttail` instructions with constant breaks `musttail` invariant 1722 // unless the call itself can be removed 1723 CallInst *CI = dyn_cast<CallInst>(V); 1724 if (CI && CI->isMustTailCall() && !CI->isSafeToRemove()) { 1725 CallSite CS(CI); 1726 Function *F = CS.getCalledFunction(); 1727 1728 // Don't zap returns of the callee 1729 if (F) 1730 Solver.AddMustTailCallee(F); 1731 1732 LLVM_DEBUG(dbgs() << " Can\'t treat the result of musttail call : " << *CI 1733 << " as a constant\n"); 1734 return false; 1735 } 1736 1737 LLVM_DEBUG(dbgs() << " Constant: " << *Const << " = " << *V << '\n'); 1738 1739 // Replaces all of the uses of a variable with uses of the constant. 1740 V->replaceAllUsesWith(Const); 1741 return true; 1742 } 1743 1744 // runSCCP() - Run the Sparse Conditional Constant Propagation algorithm, 1745 // and return true if the function was modified. 1746 static bool runSCCP(Function &F, const DataLayout &DL, 1747 const TargetLibraryInfo *TLI) { 1748 LLVM_DEBUG(dbgs() << "SCCP on function '" << F.getName() << "'\n"); 1749 SCCPSolver Solver(DL, TLI); 1750 1751 // Mark the first block of the function as being executable. 1752 Solver.MarkBlockExecutable(&F.front()); 1753 1754 // Mark all arguments to the function as being overdefined. 1755 for (Argument &AI : F.args()) 1756 Solver.markOverdefined(&AI); 1757 1758 // Solve for constants. 1759 bool ResolvedUndefs = true; 1760 while (ResolvedUndefs) { 1761 Solver.Solve(); 1762 LLVM_DEBUG(dbgs() << "RESOLVING UNDEFs\n"); 1763 ResolvedUndefs = Solver.ResolvedUndefsIn(F); 1764 } 1765 1766 bool MadeChanges = false; 1767 1768 // If we decided that there are basic blocks that are dead in this function, 1769 // delete their contents now. Note that we cannot actually delete the blocks, 1770 // as we cannot modify the CFG of the function. 1771 1772 for (BasicBlock &BB : F) { 1773 if (!Solver.isBlockExecutable(&BB)) { 1774 LLVM_DEBUG(dbgs() << " BasicBlock Dead:" << BB); 1775 1776 ++NumDeadBlocks; 1777 NumInstRemoved += removeAllNonTerminatorAndEHPadInstructions(&BB); 1778 1779 MadeChanges = true; 1780 continue; 1781 } 1782 1783 // Iterate over all of the instructions in a function, replacing them with 1784 // constants if we have found them to be of constant values. 1785 for (BasicBlock::iterator BI = BB.begin(), E = BB.end(); BI != E;) { 1786 Instruction *Inst = &*BI++; 1787 if (Inst->getType()->isVoidTy() || isa<TerminatorInst>(Inst)) 1788 continue; 1789 1790 if (tryToReplaceWithConstant(Solver, Inst)) { 1791 if (isInstructionTriviallyDead(Inst)) 1792 Inst->eraseFromParent(); 1793 // Hey, we just changed something! 1794 MadeChanges = true; 1795 ++NumInstRemoved; 1796 } 1797 } 1798 } 1799 1800 return MadeChanges; 1801 } 1802 1803 PreservedAnalyses SCCPPass::run(Function &F, FunctionAnalysisManager &AM) { 1804 const DataLayout &DL = F.getParent()->getDataLayout(); 1805 auto &TLI = AM.getResult<TargetLibraryAnalysis>(F); 1806 if (!runSCCP(F, DL, &TLI)) 1807 return PreservedAnalyses::all(); 1808 1809 auto PA = PreservedAnalyses(); 1810 PA.preserve<GlobalsAA>(); 1811 PA.preserveSet<CFGAnalyses>(); 1812 return PA; 1813 } 1814 1815 namespace { 1816 1817 //===--------------------------------------------------------------------===// 1818 // 1819 /// SCCP Class - This class uses the SCCPSolver to implement a per-function 1820 /// Sparse Conditional Constant Propagator. 1821 /// 1822 class SCCPLegacyPass : public FunctionPass { 1823 public: 1824 // Pass identification, replacement for typeid 1825 static char ID; 1826 1827 SCCPLegacyPass() : FunctionPass(ID) { 1828 initializeSCCPLegacyPassPass(*PassRegistry::getPassRegistry()); 1829 } 1830 1831 void getAnalysisUsage(AnalysisUsage &AU) const override { 1832 AU.addRequired<TargetLibraryInfoWrapperPass>(); 1833 AU.addPreserved<GlobalsAAWrapperPass>(); 1834 AU.setPreservesCFG(); 1835 } 1836 1837 // runOnFunction - Run the Sparse Conditional Constant Propagation 1838 // algorithm, and return true if the function was modified. 1839 bool runOnFunction(Function &F) override { 1840 if (skipFunction(F)) 1841 return false; 1842 const DataLayout &DL = F.getParent()->getDataLayout(); 1843 const TargetLibraryInfo *TLI = 1844 &getAnalysis<TargetLibraryInfoWrapperPass>().getTLI(); 1845 return runSCCP(F, DL, TLI); 1846 } 1847 }; 1848 1849 } // end anonymous namespace 1850 1851 char SCCPLegacyPass::ID = 0; 1852 1853 INITIALIZE_PASS_BEGIN(SCCPLegacyPass, "sccp", 1854 "Sparse Conditional Constant Propagation", false, false) 1855 INITIALIZE_PASS_DEPENDENCY(TargetLibraryInfoWrapperPass) 1856 INITIALIZE_PASS_END(SCCPLegacyPass, "sccp", 1857 "Sparse Conditional Constant Propagation", false, false) 1858 1859 // createSCCPPass - This is the public interface to this file. 1860 FunctionPass *llvm::createSCCPPass() { return new SCCPLegacyPass(); } 1861 1862 static void findReturnsToZap(Function &F, 1863 SmallVector<ReturnInst *, 8> &ReturnsToZap, 1864 SCCPSolver &Solver) { 1865 // We can only do this if we know that nothing else can call the function. 1866 if (!Solver.isArgumentTrackedFunction(&F)) 1867 return; 1868 1869 // There is a non-removable musttail call site of this function. Zapping 1870 // returns is not allowed. 1871 if (Solver.isMustTailCallee(&F)) { 1872 LLVM_DEBUG(dbgs() << "Can't zap returns of the function : " << F.getName() 1873 << " due to present musttail call of it\n"); 1874 return; 1875 } 1876 1877 for (BasicBlock &BB : F) { 1878 if (CallInst *CI = BB.getTerminatingMustTailCall()) { 1879 LLVM_DEBUG(dbgs() << "Can't zap return of the block due to present " 1880 << "musttail call : " << *CI << "\n"); 1881 (void)CI; 1882 return; 1883 } 1884 1885 if (auto *RI = dyn_cast<ReturnInst>(BB.getTerminator())) 1886 if (!isa<UndefValue>(RI->getOperand(0))) 1887 ReturnsToZap.push_back(RI); 1888 } 1889 } 1890 1891 bool llvm::runIPSCCP( 1892 Module &M, const DataLayout &DL, const TargetLibraryInfo *TLI, 1893 function_ref<std::unique_ptr<PredicateInfo>(Function &)> getPredicateInfo) { 1894 SCCPSolver Solver(DL, TLI); 1895 1896 // Loop over all functions, marking arguments to those with their addresses 1897 // taken or that are external as overdefined. 1898 for (Function &F : M) { 1899 if (F.isDeclaration()) 1900 continue; 1901 1902 Solver.addPredInfo(F, getPredicateInfo(F)); 1903 // Determine if we can track the function's return values. If so, add the 1904 // function to the solver's set of return-tracked functions. 1905 if (canTrackReturnsInterprocedurally(&F)) 1906 Solver.AddTrackedFunction(&F); 1907 1908 // Determine if we can track the function's arguments. If so, add the 1909 // function to the solver's set of argument-tracked functions. 1910 if (canTrackArgumentsInterprocedurally(&F)) { 1911 Solver.AddArgumentTrackedFunction(&F); 1912 continue; 1913 } 1914 1915 // Assume the function is called. 1916 Solver.MarkBlockExecutable(&F.front()); 1917 1918 // Assume nothing about the incoming arguments. 1919 for (Argument &AI : F.args()) 1920 Solver.markOverdefined(&AI); 1921 } 1922 1923 // Determine if we can track any of the module's global variables. If so, add 1924 // the global variables we can track to the solver's set of tracked global 1925 // variables. 1926 for (GlobalVariable &G : M.globals()) { 1927 G.removeDeadConstantUsers(); 1928 if (canTrackGlobalVariableInterprocedurally(&G)) 1929 Solver.TrackValueOfGlobalVariable(&G); 1930 } 1931 1932 // Solve for constants. 1933 bool ResolvedUndefs = true; 1934 Solver.Solve(); 1935 while (ResolvedUndefs) { 1936 LLVM_DEBUG(dbgs() << "RESOLVING UNDEFS\n"); 1937 ResolvedUndefs = false; 1938 for (Function &F : M) 1939 if (Solver.ResolvedUndefsIn(F)) { 1940 // We run Solve() after we resolved an undef in a function, because 1941 // we might deduce a fact that eliminates an undef in another function. 1942 Solver.Solve(); 1943 ResolvedUndefs = true; 1944 } 1945 } 1946 1947 bool MadeChanges = false; 1948 1949 // Iterate over all of the instructions in the module, replacing them with 1950 // constants if we have found them to be of constant values. 1951 SmallVector<BasicBlock*, 512> BlocksToErase; 1952 1953 for (Function &F : M) { 1954 if (F.isDeclaration()) 1955 continue; 1956 1957 if (Solver.isBlockExecutable(&F.front())) 1958 for (Function::arg_iterator AI = F.arg_begin(), E = F.arg_end(); AI != E; 1959 ++AI) { 1960 if (!AI->use_empty() && tryToReplaceWithConstant(Solver, &*AI)) { 1961 ++IPNumArgsElimed; 1962 continue; 1963 } 1964 } 1965 1966 for (Function::iterator BB = F.begin(), E = F.end(); BB != E; ++BB) { 1967 if (!Solver.isBlockExecutable(&*BB)) { 1968 LLVM_DEBUG(dbgs() << " BasicBlock Dead:" << *BB); 1969 ++NumDeadBlocks; 1970 1971 MadeChanges = true; 1972 1973 if (&*BB != &F.front()) 1974 BlocksToErase.push_back(&*BB); 1975 continue; 1976 } 1977 1978 for (BasicBlock::iterator BI = BB->begin(), E = BB->end(); BI != E; ) { 1979 Instruction *Inst = &*BI++; 1980 if (Inst->getType()->isVoidTy()) 1981 continue; 1982 if (tryToReplaceWithConstant(Solver, Inst)) { 1983 if (Inst->isSafeToRemove()) 1984 Inst->eraseFromParent(); 1985 // Hey, we just changed something! 1986 MadeChanges = true; 1987 ++IPNumInstRemoved; 1988 } 1989 } 1990 } 1991 1992 // Change dead blocks to unreachable. We do it after replacing constants in 1993 // all executable blocks, because changeToUnreachable may remove PHI nodes 1994 // in executable blocks we found values for. The function's entry block is 1995 // not part of BlocksToErase, so we have to handle it separately. 1996 for (BasicBlock *BB : BlocksToErase) 1997 NumInstRemoved += 1998 changeToUnreachable(BB->getFirstNonPHI(), /*UseLLVMTrap=*/false); 1999 if (!Solver.isBlockExecutable(&F.front())) 2000 NumInstRemoved += changeToUnreachable(F.front().getFirstNonPHI(), 2001 /*UseLLVMTrap=*/false); 2002 2003 // Now that all instructions in the function are constant folded, erase dead 2004 // blocks, because we can now use ConstantFoldTerminator to get rid of 2005 // in-edges. 2006 for (unsigned i = 0, e = BlocksToErase.size(); i != e; ++i) { 2007 // If there are any PHI nodes in this successor, drop entries for BB now. 2008 BasicBlock *DeadBB = BlocksToErase[i]; 2009 for (Value::user_iterator UI = DeadBB->user_begin(), 2010 UE = DeadBB->user_end(); 2011 UI != UE;) { 2012 // Grab the user and then increment the iterator early, as the user 2013 // will be deleted. Step past all adjacent uses from the same user. 2014 auto *I = dyn_cast<Instruction>(*UI); 2015 do { ++UI; } while (UI != UE && *UI == I); 2016 2017 // Ignore blockaddress users; BasicBlock's dtor will handle them. 2018 if (!I) continue; 2019 2020 bool Folded = ConstantFoldTerminator(I->getParent()); 2021 if (!Folded) { 2022 // If the branch can't be folded, we must have forced an edge 2023 // for an indeterminate value. Force the terminator to fold 2024 // to that edge. 2025 Constant *C; 2026 BasicBlock *Dest; 2027 if (SwitchInst *SI = dyn_cast<SwitchInst>(I)) { 2028 Dest = SI->case_begin()->getCaseSuccessor(); 2029 C = SI->case_begin()->getCaseValue(); 2030 } else if (BranchInst *BI = dyn_cast<BranchInst>(I)) { 2031 Dest = BI->getSuccessor(1); 2032 C = ConstantInt::getFalse(BI->getContext()); 2033 } else if (IndirectBrInst *IBR = dyn_cast<IndirectBrInst>(I)) { 2034 Dest = IBR->getSuccessor(0); 2035 C = BlockAddress::get(IBR->getSuccessor(0)); 2036 } else { 2037 llvm_unreachable("Unexpected terminator instruction"); 2038 } 2039 assert(Solver.isEdgeFeasible(I->getParent(), Dest) && 2040 "Didn't find feasible edge?"); 2041 (void)Dest; 2042 2043 I->setOperand(0, C); 2044 Folded = ConstantFoldTerminator(I->getParent()); 2045 } 2046 assert(Folded && 2047 "Expect TermInst on constantint or blockaddress to be folded"); 2048 (void) Folded; 2049 } 2050 2051 // Finally, delete the basic block. 2052 F.getBasicBlockList().erase(DeadBB); 2053 } 2054 BlocksToErase.clear(); 2055 2056 for (BasicBlock &BB : F) { 2057 for (BasicBlock::iterator BI = BB.begin(), E = BB.end(); BI != E;) { 2058 Instruction *Inst = &*BI++; 2059 if (Solver.getPredicateInfoFor(Inst)) { 2060 if (auto *II = dyn_cast<IntrinsicInst>(Inst)) { 2061 if (II->getIntrinsicID() == Intrinsic::ssa_copy) { 2062 Value *Op = II->getOperand(0); 2063 Inst->replaceAllUsesWith(Op); 2064 Inst->eraseFromParent(); 2065 } 2066 } 2067 } 2068 } 2069 } 2070 } 2071 2072 // If we inferred constant or undef return values for a function, we replaced 2073 // all call uses with the inferred value. This means we don't need to bother 2074 // actually returning anything from the function. Replace all return 2075 // instructions with return undef. 2076 // 2077 // Do this in two stages: first identify the functions we should process, then 2078 // actually zap their returns. This is important because we can only do this 2079 // if the address of the function isn't taken. In cases where a return is the 2080 // last use of a function, the order of processing functions would affect 2081 // whether other functions are optimizable. 2082 SmallVector<ReturnInst*, 8> ReturnsToZap; 2083 2084 const DenseMap<Function*, LatticeVal> &RV = Solver.getTrackedRetVals(); 2085 for (const auto &I : RV) { 2086 Function *F = I.first; 2087 if (I.second.isOverdefined() || F->getReturnType()->isVoidTy()) 2088 continue; 2089 findReturnsToZap(*F, ReturnsToZap, Solver); 2090 } 2091 2092 for (const auto &F : Solver.getMRVFunctionsTracked()) { 2093 assert(F->getReturnType()->isStructTy() && 2094 "The return type should be a struct"); 2095 StructType *STy = cast<StructType>(F->getReturnType()); 2096 if (Solver.isStructLatticeConstant(F, STy)) 2097 findReturnsToZap(*F, ReturnsToZap, Solver); 2098 } 2099 2100 // Zap all returns which we've identified as zap to change. 2101 for (unsigned i = 0, e = ReturnsToZap.size(); i != e; ++i) { 2102 Function *F = ReturnsToZap[i]->getParent()->getParent(); 2103 ReturnsToZap[i]->setOperand(0, UndefValue::get(F->getReturnType())); 2104 } 2105 2106 // If we inferred constant or undef values for globals variables, we can 2107 // delete the global and any stores that remain to it. 2108 const DenseMap<GlobalVariable*, LatticeVal> &TG = Solver.getTrackedGlobals(); 2109 for (DenseMap<GlobalVariable*, LatticeVal>::const_iterator I = TG.begin(), 2110 E = TG.end(); I != E; ++I) { 2111 GlobalVariable *GV = I->first; 2112 assert(!I->second.isOverdefined() && 2113 "Overdefined values should have been taken out of the map!"); 2114 LLVM_DEBUG(dbgs() << "Found that GV '" << GV->getName() 2115 << "' is constant!\n"); 2116 while (!GV->use_empty()) { 2117 StoreInst *SI = cast<StoreInst>(GV->user_back()); 2118 SI->eraseFromParent(); 2119 } 2120 M.getGlobalList().erase(GV); 2121 ++IPNumGlobalConst; 2122 } 2123 2124 return MadeChanges; 2125 } 2126