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