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