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/InstructionSimplify.h" 32 #include "llvm/Analysis/TargetLibraryInfo.h" 33 #include "llvm/Analysis/ValueLattice.h" 34 #include "llvm/Analysis/ValueLatticeUtils.h" 35 #include "llvm/IR/BasicBlock.h" 36 #include "llvm/IR/CallSite.h" 37 #include "llvm/IR/Constant.h" 38 #include "llvm/IR/Constants.h" 39 #include "llvm/IR/DataLayout.h" 40 #include "llvm/IR/DerivedTypes.h" 41 #include "llvm/IR/Function.h" 42 #include "llvm/IR/GlobalVariable.h" 43 #include "llvm/IR/InstVisitor.h" 44 #include "llvm/IR/InstrTypes.h" 45 #include "llvm/IR/Instruction.h" 46 #include "llvm/IR/Instructions.h" 47 #include "llvm/IR/Module.h" 48 #include "llvm/IR/PassManager.h" 49 #include "llvm/IR/Type.h" 50 #include "llvm/IR/User.h" 51 #include "llvm/IR/Value.h" 52 #include "llvm/InitializePasses.h" 53 #include "llvm/Pass.h" 54 #include "llvm/Support/Casting.h" 55 #include "llvm/Support/Debug.h" 56 #include "llvm/Support/ErrorHandling.h" 57 #include "llvm/Support/raw_ostream.h" 58 #include "llvm/Transforms/Scalar.h" 59 #include "llvm/Transforms/Utils/Local.h" 60 #include "llvm/Transforms/Utils/PredicateInfo.h" 61 #include <cassert> 62 #include <utility> 63 #include <vector> 64 65 using namespace llvm; 66 67 #define DEBUG_TYPE "sccp" 68 69 STATISTIC(NumInstRemoved, "Number of instructions removed"); 70 STATISTIC(NumDeadBlocks , "Number of basic blocks unreachable"); 71 72 STATISTIC(IPNumInstRemoved, "Number of instructions removed by IPSCCP"); 73 STATISTIC(IPNumArgsElimed ,"Number of arguments constant propagated by IPSCCP"); 74 STATISTIC(IPNumGlobalConst, "Number of globals found to be constant by IPSCCP"); 75 76 namespace { 77 78 // Helper to check if \p LV is either a constant or a constant 79 // range with a single element. This should cover exactly the same cases as the 80 // old ValueLatticeElement::isConstant() and is intended to be used in the 81 // transition to ValueLatticeElement. 82 bool isConstant(const ValueLatticeElement &LV) { 83 return LV.isConstant() || 84 (LV.isConstantRange() && LV.getConstantRange().isSingleElement()); 85 } 86 87 // Helper to check if \p LV is either overdefined or a constant range with more 88 // than a single element. This should cover exactly the same cases as the old 89 // ValueLatticeElement::isOverdefined() and is intended to be used in the 90 // transition to ValueLatticeElement. 91 bool isOverdefined(const ValueLatticeElement &LV) { 92 return LV.isOverdefined() || 93 (LV.isConstantRange() && !LV.getConstantRange().isSingleElement()); 94 } 95 96 //===----------------------------------------------------------------------===// 97 // 98 /// SCCPSolver - This class is a general purpose solver for Sparse Conditional 99 /// Constant Propagation. 100 /// 101 class SCCPSolver : public InstVisitor<SCCPSolver> { 102 const DataLayout &DL; 103 std::function<const TargetLibraryInfo &(Function &)> GetTLI; 104 SmallPtrSet<BasicBlock *, 8> BBExecutable; // The BBs that are executable. 105 DenseMap<Value *, ValueLatticeElement> 106 ValueState; // The state each value is in. 107 108 /// StructValueState - This maintains ValueState for values that have 109 /// StructType, for example for formal arguments, calls, insertelement, etc. 110 DenseMap<std::pair<Value *, unsigned>, ValueLatticeElement> StructValueState; 111 112 /// GlobalValue - If we are tracking any values for the contents of a global 113 /// variable, we keep a mapping from the constant accessor to the element of 114 /// the global, to the currently known value. If the value becomes 115 /// overdefined, it's entry is simply removed from this map. 116 DenseMap<GlobalVariable *, ValueLatticeElement> TrackedGlobals; 117 118 /// TrackedRetVals - If we are tracking arguments into and the return 119 /// value out of a function, it will have an entry in this map, indicating 120 /// what the known return value for the function is. 121 MapVector<Function *, ValueLatticeElement> TrackedRetVals; 122 123 /// TrackedMultipleRetVals - Same as TrackedRetVals, but used for functions 124 /// that return multiple values. 125 MapVector<std::pair<Function *, unsigned>, ValueLatticeElement> 126 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 ValueLatticeElement &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( 221 std::make_pair(std::make_pair(F, i), ValueLatticeElement())); 222 } else 223 TrackedRetVals.insert(std::make_pair(F, ValueLatticeElement())); 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<ValueLatticeElement> getStructLatticeValueFor(Value *V) const { 267 std::vector<ValueLatticeElement> 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 ValueLatticeElement &getLatticeValueFor(Value *V) const { 279 assert(!V->getType()->isStructTy() && 280 "Should use getStructLatticeValueFor"); 281 DenseMap<Value *, ValueLatticeElement>::const_iterator I = 282 ValueState.find(V); 283 assert(I != ValueState.end() && 284 "V not found in ValueState nor Paramstate map!"); 285 return I->second; 286 } 287 288 /// getTrackedRetVals - Get the inferred return value map. 289 const MapVector<Function *, ValueLatticeElement> &getTrackedRetVals() { 290 return TrackedRetVals; 291 } 292 293 /// getTrackedGlobals - Get and return the set of inferred initializers for 294 /// global variables. 295 const DenseMap<GlobalVariable *, ValueLatticeElement> &getTrackedGlobals() { 296 return TrackedGlobals; 297 } 298 299 /// getMRVFunctionsTracked - Get the set of functions which return multiple 300 /// values tracked by the pass. 301 const SmallPtrSet<Function *, 16> getMRVFunctionsTracked() { 302 return MRVFunctionsTracked; 303 } 304 305 /// getMustTailCallees - Get the set of functions which are called 306 /// from non-removable musttail call sites. 307 const SmallPtrSet<Function *, 16> getMustTailCallees() { 308 return MustTailCallees; 309 } 310 311 /// markOverdefined - Mark the specified value overdefined. This 312 /// works with both scalars and structs. 313 void markOverdefined(Value *V) { 314 if (auto *STy = dyn_cast<StructType>(V->getType())) 315 for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) 316 markOverdefined(getStructValueState(V, i), V); 317 else 318 markOverdefined(ValueState[V], V); 319 } 320 321 // isStructLatticeConstant - Return true if all the lattice values 322 // corresponding to elements of the structure are constants, 323 // false otherwise. 324 bool isStructLatticeConstant(Function *F, StructType *STy) { 325 for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) { 326 const auto &It = TrackedMultipleRetVals.find(std::make_pair(F, i)); 327 assert(It != TrackedMultipleRetVals.end()); 328 ValueLatticeElement LV = It->second; 329 if (!isConstant(LV)) 330 return false; 331 } 332 return true; 333 } 334 335 /// Helper to return a Constant if \p LV is either a constant or a constant 336 /// range with a single element. 337 Constant *getConstant(const ValueLatticeElement &LV) const { 338 if (LV.isConstant()) 339 return LV.getConstant(); 340 341 if (LV.isConstantRange()) { 342 auto &CR = LV.getConstantRange(); 343 if (CR.getSingleElement()) 344 return ConstantInt::get(Ctx, *CR.getSingleElement()); 345 } 346 return nullptr; 347 } 348 349 private: 350 ConstantInt *getConstantInt(const ValueLatticeElement &IV) const { 351 return dyn_cast_or_null<ConstantInt>(getConstant(IV)); 352 } 353 354 // pushToWorkList - Helper for markConstant/markOverdefined 355 void pushToWorkList(ValueLatticeElement &IV, Value *V) { 356 if (IV.isOverdefined()) 357 return OverdefinedInstWorkList.push_back(V); 358 InstWorkList.push_back(V); 359 } 360 361 // Helper to push \p V to the worklist, after updating it to \p IV. Also 362 // prints a debug message with the updated value. 363 void pushToWorkListMsg(ValueLatticeElement &IV, Value *V) { 364 LLVM_DEBUG(dbgs() << "updated " << IV << ": " << *V << '\n'); 365 if (IV.isOverdefined()) 366 return OverdefinedInstWorkList.push_back(V); 367 InstWorkList.push_back(V); 368 } 369 370 // markConstant - Make a value be marked as "constant". If the value 371 // is not already a constant, add it to the instruction work list so that 372 // the users of the instruction are updated later. 373 bool markConstant(ValueLatticeElement &IV, Value *V, Constant *C, 374 bool MayIncludeUndef = false) { 375 if (!IV.markConstant(C, MayIncludeUndef)) 376 return false; 377 LLVM_DEBUG(dbgs() << "markConstant: " << *C << ": " << *V << '\n'); 378 pushToWorkList(IV, V); 379 return true; 380 } 381 382 bool markConstant(Value *V, Constant *C) { 383 assert(!V->getType()->isStructTy() && "structs should use mergeInValue"); 384 return markConstant(ValueState[V], V, C); 385 } 386 387 // markOverdefined - Make a value be marked as "overdefined". If the 388 // value is not already overdefined, add it to the overdefined instruction 389 // work list so that the users of the instruction are updated later. 390 bool markOverdefined(ValueLatticeElement &IV, Value *V) { 391 if (!IV.markOverdefined()) return false; 392 393 LLVM_DEBUG(dbgs() << "markOverdefined: "; 394 if (auto *F = dyn_cast<Function>(V)) dbgs() 395 << "Function '" << F->getName() << "'\n"; 396 else dbgs() << *V << '\n'); 397 // Only instructions go on the work list 398 pushToWorkList(IV, V); 399 return true; 400 } 401 402 bool mergeInValue(ValueLatticeElement &IV, Value *V, 403 ValueLatticeElement MergeWithV, bool Widen = true) { 404 if (IV.mergeIn(MergeWithV, Widen)) { 405 pushToWorkList(IV, V); 406 LLVM_DEBUG(dbgs() << "Merged " << MergeWithV << " into " << *V << " : " 407 << IV << "\n"); 408 return true; 409 } 410 return false; 411 } 412 413 bool mergeInValue(Value *V, ValueLatticeElement MergeWithV, 414 bool Widen = true) { 415 assert(!V->getType()->isStructTy() && 416 "non-structs should use markConstant"); 417 return mergeInValue(ValueState[V], V, MergeWithV, Widen); 418 } 419 420 /// getValueState - Return the ValueLatticeElement object that corresponds to 421 /// the value. This function handles the case when the value hasn't been seen 422 /// yet by properly seeding constants etc. 423 ValueLatticeElement &getValueState(Value *V) { 424 assert(!V->getType()->isStructTy() && "Should use getStructValueState"); 425 426 auto I = ValueState.insert(std::make_pair(V, ValueLatticeElement())); 427 ValueLatticeElement &LV = I.first->second; 428 429 if (!I.second) 430 return LV; // Common case, already in the map. 431 432 if (auto *C = dyn_cast<Constant>(V)) 433 LV.markConstant(C); // Constants are constant 434 435 // All others are unknown by default. 436 return LV; 437 } 438 439 /// getStructValueState - Return the ValueLatticeElement object that 440 /// corresponds to the value/field pair. This function handles the case when 441 /// the value hasn't been seen yet by properly seeding constants etc. 442 ValueLatticeElement &getStructValueState(Value *V, unsigned i) { 443 assert(V->getType()->isStructTy() && "Should use getValueState"); 444 assert(i < cast<StructType>(V->getType())->getNumElements() && 445 "Invalid element #"); 446 447 auto I = StructValueState.insert( 448 std::make_pair(std::make_pair(V, i), ValueLatticeElement())); 449 ValueLatticeElement &LV = I.first->second; 450 451 if (!I.second) 452 return LV; // Common case, already in the map. 453 454 if (auto *C = dyn_cast<Constant>(V)) { 455 Constant *Elt = C->getAggregateElement(i); 456 457 if (!Elt) 458 LV.markOverdefined(); // Unknown sort of constant. 459 else if (isa<UndefValue>(Elt)) 460 ; // Undef values remain unknown. 461 else 462 LV.markConstant(Elt); // Constants are constant. 463 } 464 465 // All others are underdefined by default. 466 return LV; 467 } 468 469 /// markEdgeExecutable - Mark a basic block as executable, adding it to the BB 470 /// work list if it is not already executable. 471 bool markEdgeExecutable(BasicBlock *Source, BasicBlock *Dest) { 472 if (!KnownFeasibleEdges.insert(Edge(Source, Dest)).second) 473 return false; // This edge is already known to be executable! 474 475 if (!MarkBlockExecutable(Dest)) { 476 // If the destination is already executable, we just made an *edge* 477 // feasible that wasn't before. Revisit the PHI nodes in the block 478 // because they have potentially new operands. 479 LLVM_DEBUG(dbgs() << "Marking Edge Executable: " << Source->getName() 480 << " -> " << Dest->getName() << '\n'); 481 482 for (PHINode &PN : Dest->phis()) 483 visitPHINode(PN); 484 } 485 return true; 486 } 487 488 // getFeasibleSuccessors - Return a vector of booleans to indicate which 489 // successors are reachable from a given terminator instruction. 490 void getFeasibleSuccessors(Instruction &TI, SmallVectorImpl<bool> &Succs); 491 492 // OperandChangedState - This method is invoked on all of the users of an 493 // instruction that was just changed state somehow. Based on this 494 // information, we need to update the specified user of this instruction. 495 void OperandChangedState(Instruction *I) { 496 if (BBExecutable.count(I->getParent())) // Inst is executable? 497 visit(*I); 498 } 499 500 // Add U as additional user of V. 501 void addAdditionalUser(Value *V, User *U) { 502 auto Iter = AdditionalUsers.insert({V, {}}); 503 Iter.first->second.insert(U); 504 } 505 506 // Mark I's users as changed, including AdditionalUsers. 507 void markUsersAsChanged(Value *I) { 508 // Functions include their arguments in the use-list. Changed function 509 // values mean that the result of the function changed. We only need to 510 // update the call sites with the new function result and do not have to 511 // propagate the call arguments. 512 if (isa<Function>(I)) { 513 for (User *U : I->users()) { 514 if (auto CS = CallSite(U)) 515 handleCallResult(CS); 516 } 517 } else { 518 for (User *U : I->users()) 519 if (auto *UI = dyn_cast<Instruction>(U)) 520 OperandChangedState(UI); 521 } 522 523 auto Iter = AdditionalUsers.find(I); 524 if (Iter != AdditionalUsers.end()) { 525 for (User *U : Iter->second) 526 if (auto *UI = dyn_cast<Instruction>(U)) 527 OperandChangedState(UI); 528 } 529 } 530 void handleCallOverdefined(CallSite CS); 531 void handleCallResult(CallSite CS); 532 void handleCallArguments(CallSite CS); 533 534 private: 535 friend class InstVisitor<SCCPSolver>; 536 537 // visit implementations - Something changed in this instruction. Either an 538 // operand made a transition, or the instruction is newly executable. Change 539 // the value type of I to reflect these changes if appropriate. 540 void visitPHINode(PHINode &I); 541 542 // Terminators 543 544 void visitReturnInst(ReturnInst &I); 545 void visitTerminator(Instruction &TI); 546 547 void visitCastInst(CastInst &I); 548 void visitSelectInst(SelectInst &I); 549 void visitUnaryOperator(Instruction &I); 550 void visitBinaryOperator(Instruction &I); 551 void visitCmpInst(CmpInst &I); 552 void visitExtractValueInst(ExtractValueInst &EVI); 553 void visitInsertValueInst(InsertValueInst &IVI); 554 555 void visitCatchSwitchInst(CatchSwitchInst &CPI) { 556 markOverdefined(&CPI); 557 visitTerminator(CPI); 558 } 559 560 // Instructions that cannot be folded away. 561 562 void visitStoreInst (StoreInst &I); 563 void visitLoadInst (LoadInst &I); 564 void visitGetElementPtrInst(GetElementPtrInst &I); 565 566 void visitCallInst (CallInst &I) { 567 visitCallSite(&I); 568 } 569 570 void visitInvokeInst (InvokeInst &II) { 571 visitCallSite(&II); 572 visitTerminator(II); 573 } 574 575 void visitCallBrInst (CallBrInst &CBI) { 576 visitCallSite(&CBI); 577 visitTerminator(CBI); 578 } 579 580 void visitCallSite (CallSite CS); 581 void visitResumeInst (ResumeInst &I) { /*returns void*/ } 582 void visitUnreachableInst(UnreachableInst &I) { /*returns void*/ } 583 void visitFenceInst (FenceInst &I) { /*returns void*/ } 584 585 void visitInstruction(Instruction &I) { 586 // All the instructions we don't do any special handling for just 587 // go to overdefined. 588 LLVM_DEBUG(dbgs() << "SCCP: Don't know how to handle: " << I << '\n'); 589 markOverdefined(&I); 590 } 591 }; 592 593 } // end anonymous namespace 594 595 // getFeasibleSuccessors - Return a vector of booleans to indicate which 596 // successors are reachable from a given terminator instruction. 597 void SCCPSolver::getFeasibleSuccessors(Instruction &TI, 598 SmallVectorImpl<bool> &Succs) { 599 Succs.resize(TI.getNumSuccessors()); 600 if (auto *BI = dyn_cast<BranchInst>(&TI)) { 601 if (BI->isUnconditional()) { 602 Succs[0] = true; 603 return; 604 } 605 606 ValueLatticeElement BCValue = getValueState(BI->getCondition()); 607 ConstantInt *CI = getConstantInt(BCValue); 608 if (!CI) { 609 // Overdefined condition variables, and branches on unfoldable constant 610 // conditions, mean the branch could go either way. 611 if (!BCValue.isUnknownOrUndef()) 612 Succs[0] = Succs[1] = true; 613 return; 614 } 615 616 // Constant condition variables mean the branch can only go a single way. 617 Succs[CI->isZero()] = true; 618 return; 619 } 620 621 // Unwinding instructions successors are always executable. 622 if (TI.isExceptionalTerminator()) { 623 Succs.assign(TI.getNumSuccessors(), true); 624 return; 625 } 626 627 if (auto *SI = dyn_cast<SwitchInst>(&TI)) { 628 if (!SI->getNumCases()) { 629 Succs[0] = true; 630 return; 631 } 632 ValueLatticeElement SCValue = getValueState(SI->getCondition()); 633 ConstantInt *CI = getConstantInt(SCValue); 634 635 if (!CI) { // Overdefined or unknown condition? 636 // All destinations are executable! 637 if (!SCValue.isUnknownOrUndef()) 638 Succs.assign(TI.getNumSuccessors(), true); 639 return; 640 } 641 642 Succs[SI->findCaseValue(CI)->getSuccessorIndex()] = true; 643 return; 644 } 645 646 // In case of indirect branch and its address is a blockaddress, we mark 647 // the target as executable. 648 if (auto *IBR = dyn_cast<IndirectBrInst>(&TI)) { 649 // Casts are folded by visitCastInst. 650 ValueLatticeElement IBRValue = getValueState(IBR->getAddress()); 651 BlockAddress *Addr = dyn_cast_or_null<BlockAddress>(getConstant(IBRValue)); 652 if (!Addr) { // Overdefined or unknown condition? 653 // All destinations are executable! 654 if (!IBRValue.isUnknownOrUndef()) 655 Succs.assign(TI.getNumSuccessors(), true); 656 return; 657 } 658 659 BasicBlock* T = Addr->getBasicBlock(); 660 assert(Addr->getFunction() == T->getParent() && 661 "Block address of a different function ?"); 662 for (unsigned i = 0; i < IBR->getNumSuccessors(); ++i) { 663 // This is the target. 664 if (IBR->getDestination(i) == T) { 665 Succs[i] = true; 666 return; 667 } 668 } 669 670 // If we didn't find our destination in the IBR successor list, then we 671 // have undefined behavior. Its ok to assume no successor is executable. 672 return; 673 } 674 675 // In case of callbr, we pessimistically assume that all successors are 676 // feasible. 677 if (isa<CallBrInst>(&TI)) { 678 Succs.assign(TI.getNumSuccessors(), true); 679 return; 680 } 681 682 LLVM_DEBUG(dbgs() << "Unknown terminator instruction: " << TI << '\n'); 683 llvm_unreachable("SCCP: Don't know how to handle this terminator!"); 684 } 685 686 // isEdgeFeasible - Return true if the control flow edge from the 'From' basic 687 // block to the 'To' basic block is currently feasible. 688 bool SCCPSolver::isEdgeFeasible(BasicBlock *From, BasicBlock *To) { 689 // Check if we've called markEdgeExecutable on the edge yet. (We could 690 // be more aggressive and try to consider edges which haven't been marked 691 // yet, but there isn't any need.) 692 return KnownFeasibleEdges.count(Edge(From, To)); 693 } 694 695 // visit Implementations - Something changed in this instruction, either an 696 // operand made a transition, or the instruction is newly executable. Change 697 // the value type of I to reflect these changes if appropriate. This method 698 // makes sure to do the following actions: 699 // 700 // 1. If a phi node merges two constants in, and has conflicting value coming 701 // from different branches, or if the PHI node merges in an overdefined 702 // value, then the PHI node becomes overdefined. 703 // 2. If a phi node merges only constants in, and they all agree on value, the 704 // PHI node becomes a constant value equal to that. 705 // 3. If V <- x (op) y && isConstant(x) && isConstant(y) V = Constant 706 // 4. If V <- x (op) y && (isOverdefined(x) || isOverdefined(y)) V = Overdefined 707 // 5. If V <- MEM or V <- CALL or V <- (unknown) then V = Overdefined 708 // 6. If a conditional branch has a value that is constant, make the selected 709 // destination executable 710 // 7. If a conditional branch has a value that is overdefined, make all 711 // successors executable. 712 void SCCPSolver::visitPHINode(PHINode &PN) { 713 // If this PN returns a struct, just mark the result overdefined. 714 // TODO: We could do a lot better than this if code actually uses this. 715 if (PN.getType()->isStructTy()) 716 return (void)markOverdefined(&PN); 717 718 if (getValueState(&PN).isOverdefined()) 719 return; // Quick exit 720 721 // Super-extra-high-degree PHI nodes are unlikely to ever be marked constant, 722 // and slow us down a lot. Just mark them overdefined. 723 if (PN.getNumIncomingValues() > 64) 724 return (void)markOverdefined(&PN); 725 726 // Look at all of the executable operands of the PHI node. If any of them 727 // are overdefined, the PHI becomes overdefined as well. If they are all 728 // constant, and they agree with each other, the PHI becomes the identical 729 // constant. If they are constant and don't agree, the PHI is overdefined. 730 // If there are no executable operands, the PHI remains unknown. 731 bool Changed = false; 732 for (unsigned i = 0, e = PN.getNumIncomingValues(); i != e; ++i) { 733 ValueLatticeElement IV = getValueState(PN.getIncomingValue(i)); 734 if (!isEdgeFeasible(PN.getIncomingBlock(i), PN.getParent())) 735 continue; 736 737 ValueLatticeElement &Res = getValueState(&PN); 738 Changed |= Res.mergeIn(IV); 739 if (Res.isOverdefined()) 740 break; 741 } 742 if (Changed) 743 pushToWorkListMsg(ValueState[&PN], &PN); 744 } 745 746 void SCCPSolver::visitReturnInst(ReturnInst &I) { 747 if (I.getNumOperands() == 0) return; // ret void 748 749 // ResolvedUndefsIn might mark I as overdefined. Bail out, even if we would 750 // discover a concrete value later. 751 if (isOverdefined(ValueState[&I])) 752 return (void)markOverdefined(&I); 753 754 Function *F = I.getParent()->getParent(); 755 Value *ResultOp = I.getOperand(0); 756 757 // If we are tracking the return value of this function, merge it in. 758 if (!TrackedRetVals.empty() && !ResultOp->getType()->isStructTy()) { 759 auto TFRVI = TrackedRetVals.find(F); 760 if (TFRVI != TrackedRetVals.end()) { 761 mergeInValue(TFRVI->second, F, getValueState(ResultOp)); 762 return; 763 } 764 } 765 766 // Handle functions that return multiple values. 767 if (!TrackedMultipleRetVals.empty()) { 768 if (auto *STy = dyn_cast<StructType>(ResultOp->getType())) 769 if (MRVFunctionsTracked.count(F)) 770 for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) 771 mergeInValue(TrackedMultipleRetVals[std::make_pair(F, i)], F, 772 getStructValueState(ResultOp, i)); 773 } 774 } 775 776 void SCCPSolver::visitTerminator(Instruction &TI) { 777 SmallVector<bool, 16> SuccFeasible; 778 getFeasibleSuccessors(TI, SuccFeasible); 779 780 BasicBlock *BB = TI.getParent(); 781 782 // Mark all feasible successors executable. 783 for (unsigned i = 0, e = SuccFeasible.size(); i != e; ++i) 784 if (SuccFeasible[i]) 785 markEdgeExecutable(BB, TI.getSuccessor(i)); 786 } 787 788 void SCCPSolver::visitCastInst(CastInst &I) { 789 // ResolvedUndefsIn might mark I as overdefined. Bail out, even if we would 790 // discover a concrete value later. 791 if (ValueState[&I].isOverdefined()) 792 return; 793 794 ValueLatticeElement OpSt = getValueState(I.getOperand(0)); 795 if (Constant *OpC = getConstant(OpSt)) { 796 // Fold the constant as we build. 797 Constant *C = ConstantFoldCastOperand(I.getOpcode(), OpC, I.getType(), DL); 798 if (isa<UndefValue>(C)) 799 return; 800 // Propagate constant value 801 markConstant(&I, C); 802 } else if (OpSt.isConstantRange() && I.getDestTy()->isIntegerTy()) { 803 auto &LV = getValueState(&I); 804 ConstantRange OpRange = OpSt.getConstantRange(); 805 Type *DestTy = I.getDestTy(); 806 ConstantRange Res = 807 OpRange.castOp(I.getOpcode(), DL.getTypeSizeInBits(DestTy)); 808 mergeInValue(LV, &I, ValueLatticeElement::getRange(Res)); 809 } else if (!OpSt.isUnknownOrUndef()) 810 markOverdefined(&I); 811 } 812 813 void SCCPSolver::visitExtractValueInst(ExtractValueInst &EVI) { 814 // ResolvedUndefsIn might mark I as overdefined. Bail out, even if we would 815 // discover a concrete value later. 816 if (isOverdefined(ValueState[&EVI])) 817 return (void)markOverdefined(&EVI); 818 819 // If this returns a struct, mark all elements over defined, we don't track 820 // structs in structs. 821 if (EVI.getType()->isStructTy()) 822 return (void)markOverdefined(&EVI); 823 824 // If this is extracting from more than one level of struct, we don't know. 825 if (EVI.getNumIndices() != 1) 826 return (void)markOverdefined(&EVI); 827 828 Value *AggVal = EVI.getAggregateOperand(); 829 if (AggVal->getType()->isStructTy()) { 830 unsigned i = *EVI.idx_begin(); 831 ValueLatticeElement EltVal = getStructValueState(AggVal, i); 832 mergeInValue(getValueState(&EVI), &EVI, EltVal); 833 } else { 834 // Otherwise, must be extracting from an array. 835 return (void)markOverdefined(&EVI); 836 } 837 } 838 839 void SCCPSolver::visitInsertValueInst(InsertValueInst &IVI) { 840 auto *STy = dyn_cast<StructType>(IVI.getType()); 841 if (!STy) 842 return (void)markOverdefined(&IVI); 843 844 // ResolvedUndefsIn might mark I as overdefined. Bail out, even if we would 845 // discover a concrete value later. 846 if (isOverdefined(ValueState[&IVI])) 847 return (void)markOverdefined(&IVI); 848 849 // If this has more than one index, we can't handle it, drive all results to 850 // undef. 851 if (IVI.getNumIndices() != 1) 852 return (void)markOverdefined(&IVI); 853 854 Value *Aggr = IVI.getAggregateOperand(); 855 unsigned Idx = *IVI.idx_begin(); 856 857 // Compute the result based on what we're inserting. 858 for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) { 859 // This passes through all values that aren't the inserted element. 860 if (i != Idx) { 861 ValueLatticeElement EltVal = getStructValueState(Aggr, i); 862 mergeInValue(getStructValueState(&IVI, i), &IVI, EltVal); 863 continue; 864 } 865 866 Value *Val = IVI.getInsertedValueOperand(); 867 if (Val->getType()->isStructTy()) 868 // We don't track structs in structs. 869 markOverdefined(getStructValueState(&IVI, i), &IVI); 870 else { 871 ValueLatticeElement InVal = getValueState(Val); 872 mergeInValue(getStructValueState(&IVI, i), &IVI, InVal); 873 } 874 } 875 } 876 877 void SCCPSolver::visitSelectInst(SelectInst &I) { 878 // If this select returns a struct, just mark the result overdefined. 879 // TODO: We could do a lot better than this if code actually uses this. 880 if (I.getType()->isStructTy()) 881 return (void)markOverdefined(&I); 882 883 // ResolvedUndefsIn might mark I as overdefined. Bail out, even if we would 884 // discover a concrete value later. 885 if (ValueState[&I].isOverdefined()) 886 return (void)markOverdefined(&I); 887 888 ValueLatticeElement CondValue = getValueState(I.getCondition()); 889 if (CondValue.isUnknownOrUndef()) 890 return; 891 892 if (ConstantInt *CondCB = getConstantInt(CondValue)) { 893 Value *OpVal = CondCB->isZero() ? I.getFalseValue() : I.getTrueValue(); 894 mergeInValue(&I, getValueState(OpVal)); 895 return; 896 } 897 898 // Otherwise, the condition is overdefined or a constant we can't evaluate. 899 // See if we can produce something better than overdefined based on the T/F 900 // value. 901 ValueLatticeElement TVal = getValueState(I.getTrueValue()); 902 ValueLatticeElement FVal = getValueState(I.getFalseValue()); 903 904 bool Changed = ValueState[&I].mergeIn(TVal); 905 Changed |= ValueState[&I].mergeIn(FVal); 906 if (Changed) 907 pushToWorkListMsg(ValueState[&I], &I); 908 } 909 910 // Handle Unary Operators. 911 void SCCPSolver::visitUnaryOperator(Instruction &I) { 912 ValueLatticeElement V0State = getValueState(I.getOperand(0)); 913 914 ValueLatticeElement &IV = ValueState[&I]; 915 // ResolvedUndefsIn might mark I as overdefined. Bail out, even if we would 916 // discover a concrete value later. 917 if (isOverdefined(IV)) 918 return (void)markOverdefined(&I); 919 920 if (isConstant(V0State)) { 921 Constant *C = ConstantExpr::get(I.getOpcode(), getConstant(V0State)); 922 923 // op Y -> undef. 924 if (isa<UndefValue>(C)) 925 return; 926 return (void)markConstant(IV, &I, C); 927 } 928 929 // If something is undef, wait for it to resolve. 930 if (!isOverdefined(V0State)) 931 return; 932 933 markOverdefined(&I); 934 } 935 936 // Handle Binary Operators. 937 void SCCPSolver::visitBinaryOperator(Instruction &I) { 938 ValueLatticeElement V1State = getValueState(I.getOperand(0)); 939 ValueLatticeElement V2State = getValueState(I.getOperand(1)); 940 941 ValueLatticeElement &IV = ValueState[&I]; 942 if (IV.isOverdefined()) 943 return; 944 945 // If something is undef, wait for it to resolve. 946 if (V1State.isUnknownOrUndef() || V2State.isUnknownOrUndef()) 947 return; 948 949 if (V1State.isOverdefined() && V2State.isOverdefined()) 950 return (void)markOverdefined(&I); 951 952 // If either of the operands is a constant, try to fold it to a constant. 953 // TODO: Use information from notconstant better. 954 if ((V1State.isConstant() || V2State.isConstant())) { 955 Value *V1 = isConstant(V1State) ? getConstant(V1State) : I.getOperand(0); 956 Value *V2 = isConstant(V2State) ? getConstant(V2State) : I.getOperand(1); 957 Value *R = SimplifyBinOp(I.getOpcode(), V1, V2, SimplifyQuery(DL)); 958 auto *C = dyn_cast_or_null<Constant>(R); 959 if (C) { 960 // X op Y -> undef. 961 if (isa<UndefValue>(C)) 962 return; 963 // Conservatively assume that the result may be based on operands that may 964 // be undef. Note that we use mergeInValue to combine the constant with 965 // the existing lattice value for I, as different constants might be found 966 // after one of the operands go to overdefined, e.g. due to one operand 967 // being a special floating value. 968 ValueLatticeElement NewV; 969 NewV.markConstant(C, /*MayIncludeUndef=*/true); 970 return (void)mergeInValue(&I, NewV); 971 } 972 } 973 974 // Only use ranges for binary operators on integers. 975 if (!I.getType()->isIntegerTy()) 976 return markOverdefined(&I); 977 978 // Try to simplify to a constant range. 979 ConstantRange A = ConstantRange::getFull(I.getType()->getScalarSizeInBits()); 980 ConstantRange B = ConstantRange::getFull(I.getType()->getScalarSizeInBits()); 981 if (V1State.isConstantRange()) 982 A = V1State.getConstantRange(); 983 if (V2State.isConstantRange()) 984 B = V2State.getConstantRange(); 985 986 ConstantRange R = A.binaryOp(cast<BinaryOperator>(&I)->getOpcode(), B); 987 mergeInValue(&I, ValueLatticeElement::getRange(R)); 988 989 // TODO: Currently we do not exploit special values that produce something 990 // better than overdefined with an overdefined operand for vector or floating 991 // point types, like and <4 x i32> overdefined, zeroinitializer. 992 } 993 994 // Handle ICmpInst instruction. 995 void SCCPSolver::visitCmpInst(CmpInst &I) { 996 // Do not cache this lookup, getValueState calls later in the function might 997 // invalidate the reference. 998 if (isOverdefined(ValueState[&I])) 999 return (void)markOverdefined(&I); 1000 1001 Value *Op1 = I.getOperand(0); 1002 Value *Op2 = I.getOperand(1); 1003 1004 // For parameters, use ParamState which includes constant range info if 1005 // available. 1006 auto V1State = getValueState(Op1); 1007 auto V2State = getValueState(Op2); 1008 1009 Constant *C = V1State.getCompare(I.getPredicate(), I.getType(), V2State); 1010 if (C) { 1011 if (isa<UndefValue>(C)) 1012 return; 1013 ValueLatticeElement CV; 1014 CV.markConstant(C); 1015 mergeInValue(&I, CV); 1016 return; 1017 } 1018 1019 // If operands are still unknown, wait for it to resolve. 1020 if ((V1State.isUnknownOrUndef() || V2State.isUnknownOrUndef()) && 1021 !isConstant(ValueState[&I])) 1022 return; 1023 1024 markOverdefined(&I); 1025 } 1026 1027 // Handle getelementptr instructions. If all operands are constants then we 1028 // can turn this into a getelementptr ConstantExpr. 1029 void SCCPSolver::visitGetElementPtrInst(GetElementPtrInst &I) { 1030 if (isOverdefined(ValueState[&I])) 1031 return (void)markOverdefined(&I); 1032 1033 SmallVector<Constant*, 8> Operands; 1034 Operands.reserve(I.getNumOperands()); 1035 1036 for (unsigned i = 0, e = I.getNumOperands(); i != e; ++i) { 1037 ValueLatticeElement State = getValueState(I.getOperand(i)); 1038 if (State.isUnknownOrUndef()) 1039 return; // Operands are not resolved yet. 1040 1041 if (isOverdefined(State)) 1042 return (void)markOverdefined(&I); 1043 1044 if (Constant *C = getConstant(State)) { 1045 Operands.push_back(C); 1046 continue; 1047 } 1048 1049 return (void)markOverdefined(&I); 1050 } 1051 1052 Constant *Ptr = Operands[0]; 1053 auto Indices = makeArrayRef(Operands.begin() + 1, Operands.end()); 1054 Constant *C = 1055 ConstantExpr::getGetElementPtr(I.getSourceElementType(), Ptr, Indices); 1056 if (isa<UndefValue>(C)) 1057 return; 1058 markConstant(&I, C); 1059 } 1060 1061 void SCCPSolver::visitStoreInst(StoreInst &SI) { 1062 // If this store is of a struct, ignore it. 1063 if (SI.getOperand(0)->getType()->isStructTy()) 1064 return; 1065 1066 if (TrackedGlobals.empty() || !isa<GlobalVariable>(SI.getOperand(1))) 1067 return; 1068 1069 // ResolvedUndefsIn might mark I as overdefined. Bail out, even if we would 1070 // discover a concrete value later. 1071 if (isOverdefined(ValueState[&SI])) 1072 return (void)markOverdefined(&SI); 1073 1074 GlobalVariable *GV = cast<GlobalVariable>(SI.getOperand(1)); 1075 auto I = TrackedGlobals.find(GV); 1076 if (I == TrackedGlobals.end()) 1077 return; 1078 1079 // Get the value we are storing into the global, then merge it. 1080 mergeInValue(I->second, GV, getValueState(SI.getOperand(0))); 1081 if (isOverdefined(I->second)) 1082 TrackedGlobals.erase(I); // No need to keep tracking this! 1083 } 1084 1085 // Handle load instructions. If the operand is a constant pointer to a constant 1086 // global, we can replace the load with the loaded constant value! 1087 void SCCPSolver::visitLoadInst(LoadInst &I) { 1088 // If this load is of a struct, just mark the result overdefined. 1089 if (I.getType()->isStructTy()) 1090 return (void)markOverdefined(&I); 1091 1092 // ResolvedUndefsIn might mark I as overdefined. Bail out, even if we would 1093 // discover a concrete value later. 1094 if (isOverdefined(ValueState[&I])) 1095 return (void)markOverdefined(&I); 1096 1097 ValueLatticeElement PtrVal = getValueState(I.getOperand(0)); 1098 if (PtrVal.isUnknownOrUndef()) 1099 return; // The pointer is not resolved yet! 1100 1101 ValueLatticeElement &IV = ValueState[&I]; 1102 1103 if (!isConstant(PtrVal) || I.isVolatile()) 1104 return (void)markOverdefined(IV, &I); 1105 1106 Constant *Ptr = getConstant(PtrVal); 1107 1108 // load null is undefined. 1109 if (isa<ConstantPointerNull>(Ptr)) { 1110 if (NullPointerIsDefined(I.getFunction(), I.getPointerAddressSpace())) 1111 return (void)markOverdefined(IV, &I); 1112 else 1113 return; 1114 } 1115 1116 // Transform load (constant global) into the value loaded. 1117 if (auto *GV = dyn_cast<GlobalVariable>(Ptr)) { 1118 if (!TrackedGlobals.empty()) { 1119 // If we are tracking this global, merge in the known value for it. 1120 auto It = TrackedGlobals.find(GV); 1121 if (It != TrackedGlobals.end()) { 1122 mergeInValue(IV, &I, It->second); 1123 return; 1124 } 1125 } 1126 } 1127 1128 // Transform load from a constant into a constant if possible. 1129 if (Constant *C = ConstantFoldLoadFromConstPtr(Ptr, I.getType(), DL)) { 1130 if (isa<UndefValue>(C)) 1131 return; 1132 return (void)markConstant(IV, &I, C); 1133 } 1134 1135 // Otherwise we cannot say for certain what value this load will produce. 1136 // Bail out. 1137 markOverdefined(IV, &I); 1138 } 1139 1140 void SCCPSolver::visitCallSite(CallSite CS) { 1141 handleCallResult(CS); 1142 handleCallArguments(CS); 1143 } 1144 1145 void SCCPSolver::handleCallOverdefined(CallSite CS) { 1146 Function *F = CS.getCalledFunction(); 1147 Instruction *I = CS.getInstruction(); 1148 1149 // Void return and not tracking callee, just bail. 1150 if (I->getType()->isVoidTy()) 1151 return; 1152 1153 // Otherwise, if we have a single return value case, and if the function is 1154 // a declaration, maybe we can constant fold it. 1155 if (F && F->isDeclaration() && !I->getType()->isStructTy() && 1156 canConstantFoldCallTo(cast<CallBase>(CS.getInstruction()), F)) { 1157 SmallVector<Constant *, 8> Operands; 1158 for (CallSite::arg_iterator AI = CS.arg_begin(), E = CS.arg_end(); AI != E; 1159 ++AI) { 1160 if (AI->get()->getType()->isStructTy()) 1161 return markOverdefined(I); // Can't handle struct args. 1162 ValueLatticeElement State = getValueState(*AI); 1163 1164 if (State.isUnknownOrUndef()) 1165 return; // Operands are not resolved yet. 1166 if (isOverdefined(State)) 1167 return (void)markOverdefined(I); 1168 assert(isConstant(State) && "Unknown state!"); 1169 Operands.push_back(getConstant(State)); 1170 } 1171 1172 if (isOverdefined(getValueState(I))) 1173 return (void)markOverdefined(I); 1174 1175 // If we can constant fold this, mark the result of the call as a 1176 // constant. 1177 if (Constant *C = ConstantFoldCall(cast<CallBase>(CS.getInstruction()), F, 1178 Operands, &GetTLI(*F))) { 1179 // call -> undef. 1180 if (isa<UndefValue>(C)) 1181 return; 1182 return (void)markConstant(I, C); 1183 } 1184 } 1185 1186 // Otherwise, we don't know anything about this call, mark it overdefined. 1187 return (void)markOverdefined(I); 1188 } 1189 1190 void SCCPSolver::handleCallArguments(CallSite CS) { 1191 Function *F = CS.getCalledFunction(); 1192 // If this is a local function that doesn't have its address taken, mark its 1193 // entry block executable and merge in the actual arguments to the call into 1194 // the formal arguments of the function. 1195 if (!TrackingIncomingArguments.empty() && 1196 TrackingIncomingArguments.count(F)) { 1197 MarkBlockExecutable(&F->front()); 1198 1199 // Propagate information from this call site into the callee. 1200 CallSite::arg_iterator CAI = CS.arg_begin(); 1201 for (Function::arg_iterator AI = F->arg_begin(), E = F->arg_end(); AI != E; 1202 ++AI, ++CAI) { 1203 // If this argument is byval, and if the function is not readonly, there 1204 // will be an implicit copy formed of the input aggregate. 1205 if (AI->hasByValAttr() && !F->onlyReadsMemory()) { 1206 markOverdefined(&*AI); 1207 continue; 1208 } 1209 1210 if (auto *STy = dyn_cast<StructType>(AI->getType())) { 1211 for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) { 1212 ValueLatticeElement CallArg = getStructValueState(*CAI, i); 1213 mergeInValue(getStructValueState(&*AI, i), &*AI, CallArg); 1214 } 1215 } else 1216 mergeInValue(&*AI, getValueState(*CAI), false); 1217 } 1218 } 1219 } 1220 1221 void SCCPSolver::handleCallResult(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 Value *CopyOf = I->getOperand(0); 1231 auto *PI = getPredicateInfoFor(I); 1232 auto *PBranch = dyn_cast_or_null<PredicateBranch>(PI); 1233 if (!PI || !PBranch) { 1234 mergeInValue(ValueState[I], I, getValueState(CopyOf)); 1235 return; 1236 } 1237 1238 // Everything below relies on the condition being a comparison. 1239 auto *Cmp = dyn_cast<CmpInst>(PBranch->Condition); 1240 if (!Cmp) { 1241 mergeInValue(ValueState[I], I, getValueState(CopyOf)); 1242 return; 1243 } 1244 1245 Value *CmpOp0 = Cmp->getOperand(0); 1246 Value *CmpOp1 = Cmp->getOperand(1); 1247 if (CopyOf != CmpOp0 && CopyOf != CmpOp1) { 1248 mergeInValue(ValueState[I], I, getValueState(CopyOf)); 1249 return; 1250 } 1251 1252 auto Pred = Cmp->getPredicate(); 1253 if (CmpOp0 != CopyOf) { 1254 std::swap(CmpOp0, CmpOp1); 1255 Pred = Cmp->getSwappedPredicate(); 1256 } 1257 1258 // Wait until CmpOp1 is resolved. 1259 if (getValueState(CmpOp1).isUnknown()) { 1260 addAdditionalUser(CmpOp1, I); 1261 return; 1262 } 1263 1264 if (!PBranch->TrueEdge) 1265 Pred = CmpInst::getInversePredicate(Pred); 1266 1267 ValueLatticeElement CondVal = getValueState(CmpOp1); 1268 ValueLatticeElement &IV = ValueState[I]; 1269 ValueLatticeElement OriginalVal = getValueState(CopyOf); 1270 if (CondVal.isConstantRange() || OriginalVal.isConstantRange()) { 1271 auto NewCR = 1272 ConstantRange::getFull(DL.getTypeSizeInBits(CopyOf->getType())); 1273 1274 // Get the range imposed by the condition. 1275 if (CondVal.isConstantRange()) 1276 NewCR = ConstantRange::makeAllowedICmpRegion( 1277 Pred, CondVal.getConstantRange()); 1278 1279 // Combine range info for the original value with the new range from the 1280 // condition. 1281 auto OriginalCR = OriginalVal.isConstantRange() 1282 ? OriginalVal.getConstantRange() 1283 : ConstantRange::getFull( 1284 DL.getTypeSizeInBits(CopyOf->getType())); 1285 NewCR = NewCR.intersectWith(OriginalCR); 1286 1287 addAdditionalUser(CmpOp1, I); 1288 // TODO: Actually filp MayIncludeUndef for the created range to false, 1289 // once most places in the optimizer respect the branches on 1290 // undef/poison are UB rule. The reason why the new range cannot be 1291 // undef is as follows below: 1292 // The new range is based on a branch condition. That guarantees that 1293 // neither of the compare operands can be undef in the branch targets, 1294 // unless we have conditions that are always true/false (e.g. icmp ule 1295 // i32, %a, i32_max). For the latter overdefined/empty range will be 1296 // inferred, but the branch will get folded accordingly anyways. 1297 mergeInValue( 1298 IV, I, 1299 ValueLatticeElement::getRange(NewCR, /*MayIncludeUndef=*/true)); 1300 return; 1301 } else if (Pred == CmpInst::ICMP_EQ && CondVal.isConstant()) { 1302 // For non-integer values or integer constant expressions, only 1303 // propagate equal constants. 1304 addAdditionalUser(CmpOp1, I); 1305 mergeInValue(IV, I, CondVal); 1306 return; 1307 } 1308 1309 return (void)mergeInValue(IV, I, getValueState(CopyOf)); 1310 } 1311 } 1312 1313 // The common case is that we aren't tracking the callee, either because we 1314 // are not doing interprocedural analysis or the callee is indirect, or is 1315 // external. Handle these cases first. 1316 if (!F || F->isDeclaration()) 1317 return handleCallOverdefined(CS); 1318 1319 // If this is a single/zero retval case, see if we're tracking the function. 1320 if (auto *STy = dyn_cast<StructType>(F->getReturnType())) { 1321 if (!MRVFunctionsTracked.count(F)) 1322 return handleCallOverdefined(CS); // Not tracking this callee. 1323 1324 // If we are tracking this callee, propagate the result of the function 1325 // into this call site. 1326 for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) 1327 mergeInValue(getStructValueState(I, i), I, 1328 TrackedMultipleRetVals[std::make_pair(F, i)]); 1329 } else { 1330 auto TFRVI = TrackedRetVals.find(F); 1331 if (TFRVI == TrackedRetVals.end()) 1332 return handleCallOverdefined(CS); // Not tracking this callee. 1333 1334 // If so, propagate the return value of the callee into this call result. 1335 mergeInValue(I, TFRVI->second); 1336 } 1337 } 1338 1339 void SCCPSolver::Solve() { 1340 // Process the work lists until they are empty! 1341 while (!BBWorkList.empty() || !InstWorkList.empty() || 1342 !OverdefinedInstWorkList.empty()) { 1343 // Process the overdefined instruction's work list first, which drives other 1344 // things to overdefined more quickly. 1345 while (!OverdefinedInstWorkList.empty()) { 1346 Value *I = OverdefinedInstWorkList.pop_back_val(); 1347 1348 LLVM_DEBUG(dbgs() << "\nPopped off OI-WL: " << *I << '\n'); 1349 1350 // "I" got into the work list because it either made the transition from 1351 // bottom to constant, or to overdefined. 1352 // 1353 // Anything on this worklist that is overdefined need not be visited 1354 // since all of its users will have already been marked as overdefined 1355 // Update all of the users of this instruction's value. 1356 // 1357 markUsersAsChanged(I); 1358 } 1359 1360 // Process the instruction work list. 1361 while (!InstWorkList.empty()) { 1362 Value *I = InstWorkList.pop_back_val(); 1363 1364 LLVM_DEBUG(dbgs() << "\nPopped off I-WL: " << *I << '\n'); 1365 1366 // "I" got into the work list because it made the transition from undef to 1367 // constant. 1368 // 1369 // Anything on this worklist that is overdefined need not be visited 1370 // since all of its users will have already been marked as overdefined. 1371 // Update all of the users of this instruction's value. 1372 // 1373 if (I->getType()->isStructTy() || !getValueState(I).isOverdefined()) 1374 markUsersAsChanged(I); 1375 } 1376 1377 // Process the basic block work list. 1378 while (!BBWorkList.empty()) { 1379 BasicBlock *BB = BBWorkList.back(); 1380 BBWorkList.pop_back(); 1381 1382 LLVM_DEBUG(dbgs() << "\nPopped off BBWL: " << *BB << '\n'); 1383 1384 // Notify all instructions in this basic block that they are newly 1385 // executable. 1386 visit(BB); 1387 } 1388 } 1389 } 1390 1391 /// ResolvedUndefsIn - While solving the dataflow for a function, we assume 1392 /// that branches on undef values cannot reach any of their successors. 1393 /// However, this is not a safe assumption. After we solve dataflow, this 1394 /// method should be use to handle this. If this returns true, the solver 1395 /// should be rerun. 1396 /// 1397 /// This method handles this by finding an unresolved branch and marking it one 1398 /// of the edges from the block as being feasible, even though the condition 1399 /// doesn't say it would otherwise be. This allows SCCP to find the rest of the 1400 /// CFG and only slightly pessimizes the analysis results (by marking one, 1401 /// potentially infeasible, edge feasible). This cannot usefully modify the 1402 /// constraints on the condition of the branch, as that would impact other users 1403 /// of the value. 1404 /// 1405 /// This scan also checks for values that use undefs. It conservatively marks 1406 /// them as overdefined. 1407 bool SCCPSolver::ResolvedUndefsIn(Function &F) { 1408 for (BasicBlock &BB : F) { 1409 if (!BBExecutable.count(&BB)) 1410 continue; 1411 1412 for (Instruction &I : BB) { 1413 // Look for instructions which produce undef values. 1414 if (I.getType()->isVoidTy()) continue; 1415 1416 if (auto *STy = dyn_cast<StructType>(I.getType())) { 1417 // Only a few things that can be structs matter for undef. 1418 1419 // Tracked calls must never be marked overdefined in ResolvedUndefsIn. 1420 if (CallSite CS = CallSite(&I)) 1421 if (Function *F = CS.getCalledFunction()) 1422 if (MRVFunctionsTracked.count(F)) 1423 continue; 1424 1425 // extractvalue and insertvalue don't need to be marked; they are 1426 // tracked as precisely as their operands. 1427 if (isa<ExtractValueInst>(I) || isa<InsertValueInst>(I)) 1428 continue; 1429 // Send the results of everything else to overdefined. We could be 1430 // more precise than this but it isn't worth bothering. 1431 for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) { 1432 ValueLatticeElement &LV = getStructValueState(&I, i); 1433 if (LV.isUnknownOrUndef()) 1434 markOverdefined(LV, &I); 1435 } 1436 continue; 1437 } 1438 1439 ValueLatticeElement &LV = getValueState(&I); 1440 if (!LV.isUnknownOrUndef()) 1441 continue; 1442 1443 // There are two reasons a call can have an undef result 1444 // 1. It could be tracked. 1445 // 2. It could be constant-foldable. 1446 // Because of the way we solve return values, tracked calls must 1447 // never be marked overdefined in ResolvedUndefsIn. 1448 if (CallSite CS = CallSite(&I)) 1449 if (Function *F = CS.getCalledFunction()) 1450 if (TrackedRetVals.count(F)) 1451 continue; 1452 1453 if (isa<LoadInst>(I)) { 1454 // A load here means one of two things: a load of undef from a global, 1455 // a load from an unknown pointer. Either way, having it return undef 1456 // is okay. 1457 continue; 1458 } 1459 1460 markOverdefined(&I); 1461 return true; 1462 } 1463 1464 // Check to see if we have a branch or switch on an undefined value. If so 1465 // we force the branch to go one way or the other to make the successor 1466 // values live. It doesn't really matter which way we force it. 1467 Instruction *TI = BB.getTerminator(); 1468 if (auto *BI = dyn_cast<BranchInst>(TI)) { 1469 if (!BI->isConditional()) continue; 1470 if (!getValueState(BI->getCondition()).isUnknownOrUndef()) 1471 continue; 1472 1473 // If the input to SCCP is actually branch on undef, fix the undef to 1474 // false. 1475 if (isa<UndefValue>(BI->getCondition())) { 1476 BI->setCondition(ConstantInt::getFalse(BI->getContext())); 1477 markEdgeExecutable(&BB, TI->getSuccessor(1)); 1478 return true; 1479 } 1480 1481 // Otherwise, it is a branch on a symbolic value which is currently 1482 // considered to be undef. Make sure some edge is executable, so a 1483 // branch on "undef" always flows somewhere. 1484 // FIXME: Distinguish between dead code and an LLVM "undef" value. 1485 BasicBlock *DefaultSuccessor = TI->getSuccessor(1); 1486 if (markEdgeExecutable(&BB, DefaultSuccessor)) 1487 return true; 1488 1489 continue; 1490 } 1491 1492 if (auto *IBR = dyn_cast<IndirectBrInst>(TI)) { 1493 // Indirect branch with no successor ?. Its ok to assume it branches 1494 // to no target. 1495 if (IBR->getNumSuccessors() < 1) 1496 continue; 1497 1498 if (!getValueState(IBR->getAddress()).isUnknownOrUndef()) 1499 continue; 1500 1501 // If the input to SCCP is actually branch on undef, fix the undef to 1502 // the first successor of the indirect branch. 1503 if (isa<UndefValue>(IBR->getAddress())) { 1504 IBR->setAddress(BlockAddress::get(IBR->getSuccessor(0))); 1505 markEdgeExecutable(&BB, IBR->getSuccessor(0)); 1506 return true; 1507 } 1508 1509 // Otherwise, it is a branch on a symbolic value which is currently 1510 // considered to be undef. Make sure some edge is executable, so a 1511 // branch on "undef" always flows somewhere. 1512 // FIXME: IndirectBr on "undef" doesn't actually need to go anywhere: 1513 // we can assume the branch has undefined behavior instead. 1514 BasicBlock *DefaultSuccessor = IBR->getSuccessor(0); 1515 if (markEdgeExecutable(&BB, DefaultSuccessor)) 1516 return true; 1517 1518 continue; 1519 } 1520 1521 if (auto *SI = dyn_cast<SwitchInst>(TI)) { 1522 if (!SI->getNumCases() || 1523 !getValueState(SI->getCondition()).isUnknownOrUndef()) 1524 continue; 1525 1526 // If the input to SCCP is actually switch on undef, fix the undef to 1527 // the first constant. 1528 if (isa<UndefValue>(SI->getCondition())) { 1529 SI->setCondition(SI->case_begin()->getCaseValue()); 1530 markEdgeExecutable(&BB, SI->case_begin()->getCaseSuccessor()); 1531 return true; 1532 } 1533 1534 // Otherwise, it is a branch on a symbolic value which is currently 1535 // considered to be undef. Make sure some edge is executable, so a 1536 // branch on "undef" always flows somewhere. 1537 // FIXME: Distinguish between dead code and an LLVM "undef" value. 1538 BasicBlock *DefaultSuccessor = SI->case_begin()->getCaseSuccessor(); 1539 if (markEdgeExecutable(&BB, DefaultSuccessor)) 1540 return true; 1541 1542 continue; 1543 } 1544 } 1545 1546 return false; 1547 } 1548 1549 static bool tryToReplaceWithConstant(SCCPSolver &Solver, Value *V) { 1550 Constant *Const = nullptr; 1551 if (V->getType()->isStructTy()) { 1552 std::vector<ValueLatticeElement> IVs = Solver.getStructLatticeValueFor(V); 1553 if (any_of(IVs, 1554 [](const ValueLatticeElement &LV) { return isOverdefined(LV); })) 1555 return false; 1556 std::vector<Constant *> ConstVals; 1557 auto *ST = cast<StructType>(V->getType()); 1558 for (unsigned i = 0, e = ST->getNumElements(); i != e; ++i) { 1559 ValueLatticeElement V = IVs[i]; 1560 ConstVals.push_back(isConstant(V) 1561 ? Solver.getConstant(V) 1562 : UndefValue::get(ST->getElementType(i))); 1563 } 1564 Const = ConstantStruct::get(ST, ConstVals); 1565 } else { 1566 const ValueLatticeElement &IV = Solver.getLatticeValueFor(V); 1567 if (isOverdefined(IV)) 1568 return false; 1569 1570 Const = 1571 isConstant(IV) ? Solver.getConstant(IV) : UndefValue::get(V->getType()); 1572 } 1573 assert(Const && "Constant is nullptr here!"); 1574 1575 // Replacing `musttail` instructions with constant breaks `musttail` invariant 1576 // unless the call itself can be removed 1577 CallInst *CI = dyn_cast<CallInst>(V); 1578 if (CI && CI->isMustTailCall() && !CI->isSafeToRemove()) { 1579 CallSite CS(CI); 1580 Function *F = CS.getCalledFunction(); 1581 1582 // Don't zap returns of the callee 1583 if (F) 1584 Solver.AddMustTailCallee(F); 1585 1586 LLVM_DEBUG(dbgs() << " Can\'t treat the result of musttail call : " << *CI 1587 << " as a constant\n"); 1588 return false; 1589 } 1590 1591 LLVM_DEBUG(dbgs() << " Constant: " << *Const << " = " << *V << '\n'); 1592 1593 // Replaces all of the uses of a variable with uses of the constant. 1594 V->replaceAllUsesWith(Const); 1595 return true; 1596 } 1597 1598 // runSCCP() - Run the Sparse Conditional Constant Propagation algorithm, 1599 // and return true if the function was modified. 1600 static bool runSCCP(Function &F, const DataLayout &DL, 1601 const TargetLibraryInfo *TLI) { 1602 LLVM_DEBUG(dbgs() << "SCCP on function '" << F.getName() << "'\n"); 1603 SCCPSolver Solver( 1604 DL, [TLI](Function &F) -> const TargetLibraryInfo & { return *TLI; }, 1605 F.getContext()); 1606 1607 // Mark the first block of the function as being executable. 1608 Solver.MarkBlockExecutable(&F.front()); 1609 1610 // Mark all arguments to the function as being overdefined. 1611 for (Argument &AI : F.args()) 1612 Solver.markOverdefined(&AI); 1613 1614 // Solve for constants. 1615 bool ResolvedUndefs = true; 1616 while (ResolvedUndefs) { 1617 Solver.Solve(); 1618 LLVM_DEBUG(dbgs() << "RESOLVING UNDEFs\n"); 1619 ResolvedUndefs = Solver.ResolvedUndefsIn(F); 1620 } 1621 1622 bool MadeChanges = false; 1623 1624 // If we decided that there are basic blocks that are dead in this function, 1625 // delete their contents now. Note that we cannot actually delete the blocks, 1626 // as we cannot modify the CFG of the function. 1627 1628 for (BasicBlock &BB : F) { 1629 if (!Solver.isBlockExecutable(&BB)) { 1630 LLVM_DEBUG(dbgs() << " BasicBlock Dead:" << BB); 1631 1632 ++NumDeadBlocks; 1633 NumInstRemoved += removeAllNonTerminatorAndEHPadInstructions(&BB); 1634 1635 MadeChanges = true; 1636 continue; 1637 } 1638 1639 // Iterate over all of the instructions in a function, replacing them with 1640 // constants if we have found them to be of constant values. 1641 for (BasicBlock::iterator BI = BB.begin(), E = BB.end(); BI != E;) { 1642 Instruction *Inst = &*BI++; 1643 if (Inst->getType()->isVoidTy() || Inst->isTerminator()) 1644 continue; 1645 1646 if (tryToReplaceWithConstant(Solver, Inst)) { 1647 if (isInstructionTriviallyDead(Inst)) 1648 Inst->eraseFromParent(); 1649 // Hey, we just changed something! 1650 MadeChanges = true; 1651 ++NumInstRemoved; 1652 } 1653 } 1654 } 1655 1656 return MadeChanges; 1657 } 1658 1659 PreservedAnalyses SCCPPass::run(Function &F, FunctionAnalysisManager &AM) { 1660 const DataLayout &DL = F.getParent()->getDataLayout(); 1661 auto &TLI = AM.getResult<TargetLibraryAnalysis>(F); 1662 if (!runSCCP(F, DL, &TLI)) 1663 return PreservedAnalyses::all(); 1664 1665 auto PA = PreservedAnalyses(); 1666 PA.preserve<GlobalsAA>(); 1667 PA.preserveSet<CFGAnalyses>(); 1668 return PA; 1669 } 1670 1671 namespace { 1672 1673 //===--------------------------------------------------------------------===// 1674 // 1675 /// SCCP Class - This class uses the SCCPSolver to implement a per-function 1676 /// Sparse Conditional Constant Propagator. 1677 /// 1678 class SCCPLegacyPass : public FunctionPass { 1679 public: 1680 // Pass identification, replacement for typeid 1681 static char ID; 1682 1683 SCCPLegacyPass() : FunctionPass(ID) { 1684 initializeSCCPLegacyPassPass(*PassRegistry::getPassRegistry()); 1685 } 1686 1687 void getAnalysisUsage(AnalysisUsage &AU) const override { 1688 AU.addRequired<TargetLibraryInfoWrapperPass>(); 1689 AU.addPreserved<GlobalsAAWrapperPass>(); 1690 AU.setPreservesCFG(); 1691 } 1692 1693 // runOnFunction - Run the Sparse Conditional Constant Propagation 1694 // algorithm, and return true if the function was modified. 1695 bool runOnFunction(Function &F) override { 1696 if (skipFunction(F)) 1697 return false; 1698 const DataLayout &DL = F.getParent()->getDataLayout(); 1699 const TargetLibraryInfo *TLI = 1700 &getAnalysis<TargetLibraryInfoWrapperPass>().getTLI(F); 1701 return runSCCP(F, DL, TLI); 1702 } 1703 }; 1704 1705 } // end anonymous namespace 1706 1707 char SCCPLegacyPass::ID = 0; 1708 1709 INITIALIZE_PASS_BEGIN(SCCPLegacyPass, "sccp", 1710 "Sparse Conditional Constant Propagation", false, false) 1711 INITIALIZE_PASS_DEPENDENCY(TargetLibraryInfoWrapperPass) 1712 INITIALIZE_PASS_END(SCCPLegacyPass, "sccp", 1713 "Sparse Conditional Constant Propagation", false, false) 1714 1715 // createSCCPPass - This is the public interface to this file. 1716 FunctionPass *llvm::createSCCPPass() { return new SCCPLegacyPass(); } 1717 1718 static void findReturnsToZap(Function &F, 1719 SmallVector<ReturnInst *, 8> &ReturnsToZap, 1720 SCCPSolver &Solver) { 1721 // We can only do this if we know that nothing else can call the function. 1722 if (!Solver.isArgumentTrackedFunction(&F)) 1723 return; 1724 1725 // There is a non-removable musttail call site of this function. Zapping 1726 // returns is not allowed. 1727 if (Solver.isMustTailCallee(&F)) { 1728 LLVM_DEBUG(dbgs() << "Can't zap returns of the function : " << F.getName() 1729 << " due to present musttail call of it\n"); 1730 return; 1731 } 1732 1733 assert( 1734 all_of(F.users(), 1735 [&Solver](User *U) { 1736 if (isa<Instruction>(U) && 1737 !Solver.isBlockExecutable(cast<Instruction>(U)->getParent())) 1738 return true; 1739 // Non-callsite uses are not impacted by zapping. Also, constant 1740 // uses (like blockaddresses) could stuck around, without being 1741 // used in the underlying IR, meaning we do not have lattice 1742 // values for them. 1743 if (!CallSite(U)) 1744 return true; 1745 if (U->getType()->isStructTy()) { 1746 return all_of(Solver.getStructLatticeValueFor(U), 1747 [](const ValueLatticeElement &LV) { 1748 return !isOverdefined(LV); 1749 }); 1750 } 1751 return !isOverdefined(Solver.getLatticeValueFor(U)); 1752 }) && 1753 "We can only zap functions where all live users have a concrete value"); 1754 1755 for (BasicBlock &BB : F) { 1756 if (CallInst *CI = BB.getTerminatingMustTailCall()) { 1757 LLVM_DEBUG(dbgs() << "Can't zap return of the block due to present " 1758 << "musttail call : " << *CI << "\n"); 1759 (void)CI; 1760 return; 1761 } 1762 1763 if (auto *RI = dyn_cast<ReturnInst>(BB.getTerminator())) 1764 if (!isa<UndefValue>(RI->getOperand(0))) 1765 ReturnsToZap.push_back(RI); 1766 } 1767 } 1768 1769 // Update the condition for terminators that are branching on indeterminate 1770 // values, forcing them to use a specific edge. 1771 static void forceIndeterminateEdge(Instruction* I, SCCPSolver &Solver) { 1772 BasicBlock *Dest = nullptr; 1773 Constant *C = nullptr; 1774 if (SwitchInst *SI = dyn_cast<SwitchInst>(I)) { 1775 if (!isa<ConstantInt>(SI->getCondition())) { 1776 // Indeterminate switch; use first case value. 1777 Dest = SI->case_begin()->getCaseSuccessor(); 1778 C = SI->case_begin()->getCaseValue(); 1779 } 1780 } else if (BranchInst *BI = dyn_cast<BranchInst>(I)) { 1781 if (!isa<ConstantInt>(BI->getCondition())) { 1782 // Indeterminate branch; use false. 1783 Dest = BI->getSuccessor(1); 1784 C = ConstantInt::getFalse(BI->getContext()); 1785 } 1786 } else if (IndirectBrInst *IBR = dyn_cast<IndirectBrInst>(I)) { 1787 if (!isa<BlockAddress>(IBR->getAddress()->stripPointerCasts())) { 1788 // Indeterminate indirectbr; use successor 0. 1789 Dest = IBR->getSuccessor(0); 1790 C = BlockAddress::get(IBR->getSuccessor(0)); 1791 } 1792 } else { 1793 llvm_unreachable("Unexpected terminator instruction"); 1794 } 1795 if (C) { 1796 assert(Solver.isEdgeFeasible(I->getParent(), Dest) && 1797 "Didn't find feasible edge?"); 1798 (void)Dest; 1799 1800 I->setOperand(0, C); 1801 } 1802 } 1803 1804 bool llvm::runIPSCCP( 1805 Module &M, const DataLayout &DL, 1806 std::function<const TargetLibraryInfo &(Function &)> GetTLI, 1807 function_ref<AnalysisResultsForFn(Function &)> getAnalysis) { 1808 SCCPSolver Solver(DL, GetTLI, M.getContext()); 1809 1810 // Loop over all functions, marking arguments to those with their addresses 1811 // taken or that are external as overdefined. 1812 for (Function &F : M) { 1813 if (F.isDeclaration()) 1814 continue; 1815 1816 Solver.addAnalysis(F, getAnalysis(F)); 1817 1818 // Determine if we can track the function's return values. If so, add the 1819 // function to the solver's set of return-tracked functions. 1820 if (canTrackReturnsInterprocedurally(&F)) 1821 Solver.AddTrackedFunction(&F); 1822 1823 // Determine if we can track the function's arguments. If so, add the 1824 // function to the solver's set of argument-tracked functions. 1825 if (canTrackArgumentsInterprocedurally(&F)) { 1826 Solver.AddArgumentTrackedFunction(&F); 1827 continue; 1828 } 1829 1830 // Assume the function is called. 1831 Solver.MarkBlockExecutable(&F.front()); 1832 1833 // Assume nothing about the incoming arguments. 1834 for (Argument &AI : F.args()) 1835 Solver.markOverdefined(&AI); 1836 } 1837 1838 // Determine if we can track any of the module's global variables. If so, add 1839 // the global variables we can track to the solver's set of tracked global 1840 // variables. 1841 for (GlobalVariable &G : M.globals()) { 1842 G.removeDeadConstantUsers(); 1843 if (canTrackGlobalVariableInterprocedurally(&G)) 1844 Solver.TrackValueOfGlobalVariable(&G); 1845 } 1846 1847 // Solve for constants. 1848 bool ResolvedUndefs = true; 1849 Solver.Solve(); 1850 while (ResolvedUndefs) { 1851 LLVM_DEBUG(dbgs() << "RESOLVING UNDEFS\n"); 1852 ResolvedUndefs = false; 1853 for (Function &F : M) 1854 if (Solver.ResolvedUndefsIn(F)) { 1855 // We run Solve() after we resolved an undef in a function, because 1856 // we might deduce a fact that eliminates an undef in another function. 1857 Solver.Solve(); 1858 ResolvedUndefs = true; 1859 } 1860 } 1861 1862 bool MadeChanges = false; 1863 1864 // Iterate over all of the instructions in the module, replacing them with 1865 // constants if we have found them to be of constant values. 1866 1867 for (Function &F : M) { 1868 if (F.isDeclaration()) 1869 continue; 1870 1871 SmallVector<BasicBlock *, 512> BlocksToErase; 1872 1873 if (Solver.isBlockExecutable(&F.front())) 1874 for (Function::arg_iterator AI = F.arg_begin(), E = F.arg_end(); AI != E; 1875 ++AI) { 1876 if (!AI->use_empty() && tryToReplaceWithConstant(Solver, &*AI)) { 1877 ++IPNumArgsElimed; 1878 continue; 1879 } 1880 } 1881 1882 for (Function::iterator BB = F.begin(), E = F.end(); BB != E; ++BB) { 1883 if (!Solver.isBlockExecutable(&*BB)) { 1884 LLVM_DEBUG(dbgs() << " BasicBlock Dead:" << *BB); 1885 ++NumDeadBlocks; 1886 1887 MadeChanges = true; 1888 1889 if (&*BB != &F.front()) 1890 BlocksToErase.push_back(&*BB); 1891 continue; 1892 } 1893 1894 for (BasicBlock::iterator BI = BB->begin(), E = BB->end(); BI != E; ) { 1895 Instruction *Inst = &*BI++; 1896 if (Inst->getType()->isVoidTy()) 1897 continue; 1898 if (tryToReplaceWithConstant(Solver, Inst)) { 1899 if (Inst->isSafeToRemove()) 1900 Inst->eraseFromParent(); 1901 // Hey, we just changed something! 1902 MadeChanges = true; 1903 ++IPNumInstRemoved; 1904 } 1905 } 1906 } 1907 1908 DomTreeUpdater DTU = Solver.getDTU(F); 1909 // Change dead blocks to unreachable. We do it after replacing constants 1910 // in all executable blocks, because changeToUnreachable may remove PHI 1911 // nodes in executable blocks we found values for. The function's entry 1912 // block is not part of BlocksToErase, so we have to handle it separately. 1913 for (BasicBlock *BB : BlocksToErase) { 1914 NumInstRemoved += 1915 changeToUnreachable(BB->getFirstNonPHI(), /*UseLLVMTrap=*/false, 1916 /*PreserveLCSSA=*/false, &DTU); 1917 } 1918 if (!Solver.isBlockExecutable(&F.front())) 1919 NumInstRemoved += changeToUnreachable(F.front().getFirstNonPHI(), 1920 /*UseLLVMTrap=*/false, 1921 /*PreserveLCSSA=*/false, &DTU); 1922 1923 // Now that all instructions in the function are constant folded, 1924 // use ConstantFoldTerminator to get rid of in-edges, record DT updates and 1925 // delete dead BBs. 1926 for (BasicBlock *DeadBB : BlocksToErase) { 1927 // If there are any PHI nodes in this successor, drop entries for BB now. 1928 for (Value::user_iterator UI = DeadBB->user_begin(), 1929 UE = DeadBB->user_end(); 1930 UI != UE;) { 1931 // Grab the user and then increment the iterator early, as the user 1932 // will be deleted. Step past all adjacent uses from the same user. 1933 auto *I = dyn_cast<Instruction>(*UI); 1934 do { ++UI; } while (UI != UE && *UI == I); 1935 1936 // Ignore blockaddress users; BasicBlock's dtor will handle them. 1937 if (!I) continue; 1938 1939 // If we have forced an edge for an indeterminate value, then force the 1940 // terminator to fold to that edge. 1941 forceIndeterminateEdge(I, Solver); 1942 BasicBlock *InstBB = I->getParent(); 1943 bool Folded = ConstantFoldTerminator(InstBB, 1944 /*DeleteDeadConditions=*/false, 1945 /*TLI=*/nullptr, &DTU); 1946 assert(Folded && 1947 "Expect TermInst on constantint or blockaddress to be folded"); 1948 (void) Folded; 1949 // If we folded the terminator to an unconditional branch to another 1950 // dead block, replace it with Unreachable, to avoid trying to fold that 1951 // branch again. 1952 BranchInst *BI = cast<BranchInst>(InstBB->getTerminator()); 1953 if (BI && BI->isUnconditional() && 1954 !Solver.isBlockExecutable(BI->getSuccessor(0))) { 1955 InstBB->getTerminator()->eraseFromParent(); 1956 new UnreachableInst(InstBB->getContext(), InstBB); 1957 } 1958 } 1959 // Mark dead BB for deletion. 1960 DTU.deleteBB(DeadBB); 1961 } 1962 1963 for (BasicBlock &BB : F) { 1964 for (BasicBlock::iterator BI = BB.begin(), E = BB.end(); BI != E;) { 1965 Instruction *Inst = &*BI++; 1966 if (Solver.getPredicateInfoFor(Inst)) { 1967 if (auto *II = dyn_cast<IntrinsicInst>(Inst)) { 1968 if (II->getIntrinsicID() == Intrinsic::ssa_copy) { 1969 Value *Op = II->getOperand(0); 1970 Inst->replaceAllUsesWith(Op); 1971 Inst->eraseFromParent(); 1972 } 1973 } 1974 } 1975 } 1976 } 1977 } 1978 1979 // If we inferred constant or undef return values for a function, we replaced 1980 // all call uses with the inferred value. This means we don't need to bother 1981 // actually returning anything from the function. Replace all return 1982 // instructions with return undef. 1983 // 1984 // Do this in two stages: first identify the functions we should process, then 1985 // actually zap their returns. This is important because we can only do this 1986 // if the address of the function isn't taken. In cases where a return is the 1987 // last use of a function, the order of processing functions would affect 1988 // whether other functions are optimizable. 1989 SmallVector<ReturnInst*, 8> ReturnsToZap; 1990 1991 for (const auto &I : Solver.getTrackedRetVals()) { 1992 Function *F = I.first; 1993 if (isOverdefined(I.second) || F->getReturnType()->isVoidTy()) 1994 continue; 1995 findReturnsToZap(*F, ReturnsToZap, Solver); 1996 } 1997 1998 for (auto F : Solver.getMRVFunctionsTracked()) { 1999 assert(F->getReturnType()->isStructTy() && 2000 "The return type should be a struct"); 2001 StructType *STy = cast<StructType>(F->getReturnType()); 2002 if (Solver.isStructLatticeConstant(F, STy)) 2003 findReturnsToZap(*F, ReturnsToZap, Solver); 2004 } 2005 2006 // Zap all returns which we've identified as zap to change. 2007 for (unsigned i = 0, e = ReturnsToZap.size(); i != e; ++i) { 2008 Function *F = ReturnsToZap[i]->getParent()->getParent(); 2009 ReturnsToZap[i]->setOperand(0, UndefValue::get(F->getReturnType())); 2010 } 2011 2012 // If we inferred constant or undef values for globals variables, we can 2013 // delete the global and any stores that remain to it. 2014 for (auto &I : make_early_inc_range(Solver.getTrackedGlobals())) { 2015 GlobalVariable *GV = I.first; 2016 if (isOverdefined(I.second)) 2017 continue; 2018 LLVM_DEBUG(dbgs() << "Found that GV '" << GV->getName() 2019 << "' is constant!\n"); 2020 while (!GV->use_empty()) { 2021 StoreInst *SI = cast<StoreInst>(GV->user_back()); 2022 SI->eraseFromParent(); 2023 } 2024 M.getGlobalList().erase(GV); 2025 ++IPNumGlobalConst; 2026 } 2027 2028 return MadeChanges; 2029 } 2030