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 // Do a simple form of widening, to avoid extending a range repeatedly in a 405 // loop. If IV is a constant range, it means we already set it once. If 406 // MergeWithV would extend IV, mark V as overdefined. 407 if (Widen && IV.isConstantRange() && MergeWithV.isConstantRange() && 408 !IV.getConstantRange().contains(MergeWithV.getConstantRange())) { 409 markOverdefined(IV, V); 410 return true; 411 } 412 if (IV.mergeIn(MergeWithV)) { 413 pushToWorkList(IV, V); 414 LLVM_DEBUG(dbgs() << "Merged " << MergeWithV << " into " << *V << " : " 415 << IV << "\n"); 416 return true; 417 } 418 return false; 419 } 420 421 bool mergeInValue(Value *V, ValueLatticeElement MergeWithV, 422 bool Widen = true) { 423 assert(!V->getType()->isStructTy() && 424 "non-structs should use markConstant"); 425 return mergeInValue(ValueState[V], V, MergeWithV, Widen); 426 } 427 428 /// getValueState - Return the ValueLatticeElement object that corresponds to 429 /// the value. This function handles the case when the value hasn't been seen 430 /// yet by properly seeding constants etc. 431 ValueLatticeElement &getValueState(Value *V) { 432 assert(!V->getType()->isStructTy() && "Should use getStructValueState"); 433 434 auto I = ValueState.insert(std::make_pair(V, ValueLatticeElement())); 435 ValueLatticeElement &LV = I.first->second; 436 437 if (!I.second) 438 return LV; // Common case, already in the map. 439 440 if (auto *C = dyn_cast<Constant>(V)) 441 LV.markConstant(C); // Constants are constant 442 443 // All others are unknown by default. 444 return LV; 445 } 446 447 /// getStructValueState - Return the ValueLatticeElement object that 448 /// corresponds to the value/field pair. This function handles the case when 449 /// the value hasn't been seen yet by properly seeding constants etc. 450 ValueLatticeElement &getStructValueState(Value *V, unsigned i) { 451 assert(V->getType()->isStructTy() && "Should use getValueState"); 452 assert(i < cast<StructType>(V->getType())->getNumElements() && 453 "Invalid element #"); 454 455 auto I = StructValueState.insert( 456 std::make_pair(std::make_pair(V, i), ValueLatticeElement())); 457 ValueLatticeElement &LV = I.first->second; 458 459 if (!I.second) 460 return LV; // Common case, already in the map. 461 462 if (auto *C = dyn_cast<Constant>(V)) { 463 Constant *Elt = C->getAggregateElement(i); 464 465 if (!Elt) 466 LV.markOverdefined(); // Unknown sort of constant. 467 else if (isa<UndefValue>(Elt)) 468 ; // Undef values remain unknown. 469 else 470 LV.markConstant(Elt); // Constants are constant. 471 } 472 473 // All others are underdefined by default. 474 return LV; 475 } 476 477 /// markEdgeExecutable - Mark a basic block as executable, adding it to the BB 478 /// work list if it is not already executable. 479 bool markEdgeExecutable(BasicBlock *Source, BasicBlock *Dest) { 480 if (!KnownFeasibleEdges.insert(Edge(Source, Dest)).second) 481 return false; // This edge is already known to be executable! 482 483 if (!MarkBlockExecutable(Dest)) { 484 // If the destination is already executable, we just made an *edge* 485 // feasible that wasn't before. Revisit the PHI nodes in the block 486 // because they have potentially new operands. 487 LLVM_DEBUG(dbgs() << "Marking Edge Executable: " << Source->getName() 488 << " -> " << Dest->getName() << '\n'); 489 490 for (PHINode &PN : Dest->phis()) 491 visitPHINode(PN); 492 } 493 return true; 494 } 495 496 // getFeasibleSuccessors - Return a vector of booleans to indicate which 497 // successors are reachable from a given terminator instruction. 498 void getFeasibleSuccessors(Instruction &TI, SmallVectorImpl<bool> &Succs); 499 500 // OperandChangedState - This method is invoked on all of the users of an 501 // instruction that was just changed state somehow. Based on this 502 // information, we need to update the specified user of this instruction. 503 void OperandChangedState(Instruction *I) { 504 if (BBExecutable.count(I->getParent())) // Inst is executable? 505 visit(*I); 506 } 507 508 // Add U as additional user of V. 509 void addAdditionalUser(Value *V, User *U) { 510 auto Iter = AdditionalUsers.insert({V, {}}); 511 Iter.first->second.insert(U); 512 } 513 514 // Mark I's users as changed, including AdditionalUsers. 515 void markUsersAsChanged(Value *I) { 516 // Functions include their arguments in the use-list. Changed function 517 // values mean that the result of the function changed. We only need to 518 // update the call sites with the new function result and do not have to 519 // propagate the call arguments. 520 if (isa<Function>(I)) { 521 for (User *U : I->users()) { 522 if (auto CS = CallSite(U)) 523 handleCallResult(CS); 524 } 525 } else { 526 for (User *U : I->users()) 527 if (auto *UI = dyn_cast<Instruction>(U)) 528 OperandChangedState(UI); 529 } 530 531 auto Iter = AdditionalUsers.find(I); 532 if (Iter != AdditionalUsers.end()) { 533 for (User *U : Iter->second) 534 if (auto *UI = dyn_cast<Instruction>(U)) 535 OperandChangedState(UI); 536 } 537 } 538 void handleCallOverdefined(CallSite CS); 539 void handleCallResult(CallSite CS); 540 void handleCallArguments(CallSite CS); 541 542 private: 543 friend class InstVisitor<SCCPSolver>; 544 545 // visit implementations - Something changed in this instruction. Either an 546 // operand made a transition, or the instruction is newly executable. Change 547 // the value type of I to reflect these changes if appropriate. 548 void visitPHINode(PHINode &I); 549 550 // Terminators 551 552 void visitReturnInst(ReturnInst &I); 553 void visitTerminator(Instruction &TI); 554 555 void visitCastInst(CastInst &I); 556 void visitSelectInst(SelectInst &I); 557 void visitUnaryOperator(Instruction &I); 558 void visitBinaryOperator(Instruction &I); 559 void visitCmpInst(CmpInst &I); 560 void visitExtractValueInst(ExtractValueInst &EVI); 561 void visitInsertValueInst(InsertValueInst &IVI); 562 563 void visitCatchSwitchInst(CatchSwitchInst &CPI) { 564 markOverdefined(&CPI); 565 visitTerminator(CPI); 566 } 567 568 // Instructions that cannot be folded away. 569 570 void visitStoreInst (StoreInst &I); 571 void visitLoadInst (LoadInst &I); 572 void visitGetElementPtrInst(GetElementPtrInst &I); 573 574 void visitCallInst (CallInst &I) { 575 visitCallSite(&I); 576 } 577 578 void visitInvokeInst (InvokeInst &II) { 579 visitCallSite(&II); 580 visitTerminator(II); 581 } 582 583 void visitCallBrInst (CallBrInst &CBI) { 584 visitCallSite(&CBI); 585 visitTerminator(CBI); 586 } 587 588 void visitCallSite (CallSite CS); 589 void visitResumeInst (ResumeInst &I) { /*returns void*/ } 590 void visitUnreachableInst(UnreachableInst &I) { /*returns void*/ } 591 void visitFenceInst (FenceInst &I) { /*returns void*/ } 592 593 void visitInstruction(Instruction &I) { 594 // All the instructions we don't do any special handling for just 595 // go to overdefined. 596 LLVM_DEBUG(dbgs() << "SCCP: Don't know how to handle: " << I << '\n'); 597 markOverdefined(&I); 598 } 599 }; 600 601 } // end anonymous namespace 602 603 // getFeasibleSuccessors - Return a vector of booleans to indicate which 604 // successors are reachable from a given terminator instruction. 605 void SCCPSolver::getFeasibleSuccessors(Instruction &TI, 606 SmallVectorImpl<bool> &Succs) { 607 Succs.resize(TI.getNumSuccessors()); 608 if (auto *BI = dyn_cast<BranchInst>(&TI)) { 609 if (BI->isUnconditional()) { 610 Succs[0] = true; 611 return; 612 } 613 614 ValueLatticeElement BCValue = getValueState(BI->getCondition()); 615 ConstantInt *CI = getConstantInt(BCValue); 616 if (!CI) { 617 // Overdefined condition variables, and branches on unfoldable constant 618 // conditions, mean the branch could go either way. 619 if (!BCValue.isUnknownOrUndef()) 620 Succs[0] = Succs[1] = true; 621 return; 622 } 623 624 // Constant condition variables mean the branch can only go a single way. 625 Succs[CI->isZero()] = true; 626 return; 627 } 628 629 // Unwinding instructions successors are always executable. 630 if (TI.isExceptionalTerminator()) { 631 Succs.assign(TI.getNumSuccessors(), true); 632 return; 633 } 634 635 if (auto *SI = dyn_cast<SwitchInst>(&TI)) { 636 if (!SI->getNumCases()) { 637 Succs[0] = true; 638 return; 639 } 640 ValueLatticeElement SCValue = getValueState(SI->getCondition()); 641 ConstantInt *CI = getConstantInt(SCValue); 642 643 if (!CI) { // Overdefined or unknown condition? 644 // All destinations are executable! 645 if (!SCValue.isUnknownOrUndef()) 646 Succs.assign(TI.getNumSuccessors(), true); 647 return; 648 } 649 650 Succs[SI->findCaseValue(CI)->getSuccessorIndex()] = true; 651 return; 652 } 653 654 // In case of indirect branch and its address is a blockaddress, we mark 655 // the target as executable. 656 if (auto *IBR = dyn_cast<IndirectBrInst>(&TI)) { 657 // Casts are folded by visitCastInst. 658 ValueLatticeElement IBRValue = getValueState(IBR->getAddress()); 659 BlockAddress *Addr = dyn_cast_or_null<BlockAddress>(getConstant(IBRValue)); 660 if (!Addr) { // Overdefined or unknown condition? 661 // All destinations are executable! 662 if (!IBRValue.isUnknownOrUndef()) 663 Succs.assign(TI.getNumSuccessors(), true); 664 return; 665 } 666 667 BasicBlock* T = Addr->getBasicBlock(); 668 assert(Addr->getFunction() == T->getParent() && 669 "Block address of a different function ?"); 670 for (unsigned i = 0; i < IBR->getNumSuccessors(); ++i) { 671 // This is the target. 672 if (IBR->getDestination(i) == T) { 673 Succs[i] = true; 674 return; 675 } 676 } 677 678 // If we didn't find our destination in the IBR successor list, then we 679 // have undefined behavior. Its ok to assume no successor is executable. 680 return; 681 } 682 683 // In case of callbr, we pessimistically assume that all successors are 684 // feasible. 685 if (isa<CallBrInst>(&TI)) { 686 Succs.assign(TI.getNumSuccessors(), true); 687 return; 688 } 689 690 LLVM_DEBUG(dbgs() << "Unknown terminator instruction: " << TI << '\n'); 691 llvm_unreachable("SCCP: Don't know how to handle this terminator!"); 692 } 693 694 // isEdgeFeasible - Return true if the control flow edge from the 'From' basic 695 // block to the 'To' basic block is currently feasible. 696 bool SCCPSolver::isEdgeFeasible(BasicBlock *From, BasicBlock *To) { 697 // Check if we've called markEdgeExecutable on the edge yet. (We could 698 // be more aggressive and try to consider edges which haven't been marked 699 // yet, but there isn't any need.) 700 return KnownFeasibleEdges.count(Edge(From, To)); 701 } 702 703 // visit Implementations - Something changed in this instruction, either an 704 // operand made a transition, or the instruction is newly executable. Change 705 // the value type of I to reflect these changes if appropriate. This method 706 // makes sure to do the following actions: 707 // 708 // 1. If a phi node merges two constants in, and has conflicting value coming 709 // from different branches, or if the PHI node merges in an overdefined 710 // value, then the PHI node becomes overdefined. 711 // 2. If a phi node merges only constants in, and they all agree on value, the 712 // PHI node becomes a constant value equal to that. 713 // 3. If V <- x (op) y && isConstant(x) && isConstant(y) V = Constant 714 // 4. If V <- x (op) y && (isOverdefined(x) || isOverdefined(y)) V = Overdefined 715 // 5. If V <- MEM or V <- CALL or V <- (unknown) then V = Overdefined 716 // 6. If a conditional branch has a value that is constant, make the selected 717 // destination executable 718 // 7. If a conditional branch has a value that is overdefined, make all 719 // successors executable. 720 void SCCPSolver::visitPHINode(PHINode &PN) { 721 // If this PN returns a struct, just mark the result overdefined. 722 // TODO: We could do a lot better than this if code actually uses this. 723 if (PN.getType()->isStructTy()) 724 return (void)markOverdefined(&PN); 725 726 if (getValueState(&PN).isOverdefined()) 727 return; // Quick exit 728 729 // Super-extra-high-degree PHI nodes are unlikely to ever be marked constant, 730 // and slow us down a lot. Just mark them overdefined. 731 if (PN.getNumIncomingValues() > 64) 732 return (void)markOverdefined(&PN); 733 734 // Look at all of the executable operands of the PHI node. If any of them 735 // are overdefined, the PHI becomes overdefined as well. If they are all 736 // constant, and they agree with each other, the PHI becomes the identical 737 // constant. If they are constant and don't agree, the PHI is overdefined. 738 // If there are no executable operands, the PHI remains unknown. 739 bool Changed = false; 740 for (unsigned i = 0, e = PN.getNumIncomingValues(); i != e; ++i) { 741 ValueLatticeElement IV = getValueState(PN.getIncomingValue(i)); 742 if (!isEdgeFeasible(PN.getIncomingBlock(i), PN.getParent())) 743 continue; 744 745 ValueLatticeElement &Res = getValueState(&PN); 746 Changed |= Res.mergeIn(IV); 747 if (Res.isOverdefined()) 748 break; 749 } 750 if (Changed) 751 pushToWorkListMsg(ValueState[&PN], &PN); 752 } 753 754 void SCCPSolver::visitReturnInst(ReturnInst &I) { 755 if (I.getNumOperands() == 0) return; // ret void 756 757 // ResolvedUndefsIn might mark I as overdefined. Bail out, even if we would 758 // discover a concrete value later. 759 if (isOverdefined(ValueState[&I])) 760 return (void)markOverdefined(&I); 761 762 Function *F = I.getParent()->getParent(); 763 Value *ResultOp = I.getOperand(0); 764 765 // If we are tracking the return value of this function, merge it in. 766 if (!TrackedRetVals.empty() && !ResultOp->getType()->isStructTy()) { 767 auto TFRVI = TrackedRetVals.find(F); 768 if (TFRVI != TrackedRetVals.end()) { 769 mergeInValue(TFRVI->second, F, getValueState(ResultOp)); 770 return; 771 } 772 } 773 774 // Handle functions that return multiple values. 775 if (!TrackedMultipleRetVals.empty()) { 776 if (auto *STy = dyn_cast<StructType>(ResultOp->getType())) 777 if (MRVFunctionsTracked.count(F)) 778 for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) 779 mergeInValue(TrackedMultipleRetVals[std::make_pair(F, i)], F, 780 getStructValueState(ResultOp, i)); 781 } 782 } 783 784 void SCCPSolver::visitTerminator(Instruction &TI) { 785 SmallVector<bool, 16> SuccFeasible; 786 getFeasibleSuccessors(TI, SuccFeasible); 787 788 BasicBlock *BB = TI.getParent(); 789 790 // Mark all feasible successors executable. 791 for (unsigned i = 0, e = SuccFeasible.size(); i != e; ++i) 792 if (SuccFeasible[i]) 793 markEdgeExecutable(BB, TI.getSuccessor(i)); 794 } 795 796 void SCCPSolver::visitCastInst(CastInst &I) { 797 // ResolvedUndefsIn might mark I as overdefined. Bail out, even if we would 798 // discover a concrete value later. 799 if (ValueState[&I].isOverdefined()) 800 return; 801 802 ValueLatticeElement OpSt = getValueState(I.getOperand(0)); 803 if (Constant *OpC = getConstant(OpSt)) { 804 // Fold the constant as we build. 805 Constant *C = ConstantFoldCastOperand(I.getOpcode(), OpC, I.getType(), DL); 806 if (isa<UndefValue>(C)) 807 return; 808 // Propagate constant value 809 markConstant(&I, C); 810 } else if (OpSt.isConstantRange() && I.getDestTy()->isIntegerTy()) { 811 auto &LV = getValueState(&I); 812 ConstantRange OpRange = OpSt.getConstantRange(); 813 Type *DestTy = I.getDestTy(); 814 ConstantRange Res = 815 OpRange.castOp(I.getOpcode(), DL.getTypeSizeInBits(DestTy)); 816 mergeInValue(LV, &I, ValueLatticeElement::getRange(Res)); 817 } else if (!OpSt.isUnknownOrUndef()) 818 markOverdefined(&I); 819 } 820 821 void SCCPSolver::visitExtractValueInst(ExtractValueInst &EVI) { 822 // ResolvedUndefsIn might mark I as overdefined. Bail out, even if we would 823 // discover a concrete value later. 824 if (isOverdefined(ValueState[&EVI])) 825 return (void)markOverdefined(&EVI); 826 827 // If this returns a struct, mark all elements over defined, we don't track 828 // structs in structs. 829 if (EVI.getType()->isStructTy()) 830 return (void)markOverdefined(&EVI); 831 832 // If this is extracting from more than one level of struct, we don't know. 833 if (EVI.getNumIndices() != 1) 834 return (void)markOverdefined(&EVI); 835 836 Value *AggVal = EVI.getAggregateOperand(); 837 if (AggVal->getType()->isStructTy()) { 838 unsigned i = *EVI.idx_begin(); 839 ValueLatticeElement EltVal = getStructValueState(AggVal, i); 840 mergeInValue(getValueState(&EVI), &EVI, EltVal); 841 } else { 842 // Otherwise, must be extracting from an array. 843 return (void)markOverdefined(&EVI); 844 } 845 } 846 847 void SCCPSolver::visitInsertValueInst(InsertValueInst &IVI) { 848 auto *STy = dyn_cast<StructType>(IVI.getType()); 849 if (!STy) 850 return (void)markOverdefined(&IVI); 851 852 // ResolvedUndefsIn might mark I as overdefined. Bail out, even if we would 853 // discover a concrete value later. 854 if (isOverdefined(ValueState[&IVI])) 855 return (void)markOverdefined(&IVI); 856 857 // If this has more than one index, we can't handle it, drive all results to 858 // undef. 859 if (IVI.getNumIndices() != 1) 860 return (void)markOverdefined(&IVI); 861 862 Value *Aggr = IVI.getAggregateOperand(); 863 unsigned Idx = *IVI.idx_begin(); 864 865 // Compute the result based on what we're inserting. 866 for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) { 867 // This passes through all values that aren't the inserted element. 868 if (i != Idx) { 869 ValueLatticeElement EltVal = getStructValueState(Aggr, i); 870 mergeInValue(getStructValueState(&IVI, i), &IVI, EltVal); 871 continue; 872 } 873 874 Value *Val = IVI.getInsertedValueOperand(); 875 if (Val->getType()->isStructTy()) 876 // We don't track structs in structs. 877 markOverdefined(getStructValueState(&IVI, i), &IVI); 878 else { 879 ValueLatticeElement InVal = getValueState(Val); 880 mergeInValue(getStructValueState(&IVI, i), &IVI, InVal); 881 } 882 } 883 } 884 885 void SCCPSolver::visitSelectInst(SelectInst &I) { 886 // If this select returns a struct, just mark the result overdefined. 887 // TODO: We could do a lot better than this if code actually uses this. 888 if (I.getType()->isStructTy()) 889 return (void)markOverdefined(&I); 890 891 // ResolvedUndefsIn might mark I as overdefined. Bail out, even if we would 892 // discover a concrete value later. 893 if (ValueState[&I].isOverdefined()) 894 return (void)markOverdefined(&I); 895 896 ValueLatticeElement CondValue = getValueState(I.getCondition()); 897 if (CondValue.isUnknownOrUndef()) 898 return; 899 900 if (ConstantInt *CondCB = getConstantInt(CondValue)) { 901 Value *OpVal = CondCB->isZero() ? I.getFalseValue() : I.getTrueValue(); 902 mergeInValue(&I, getValueState(OpVal)); 903 return; 904 } 905 906 // Otherwise, the condition is overdefined or a constant we can't evaluate. 907 // See if we can produce something better than overdefined based on the T/F 908 // value. 909 ValueLatticeElement TVal = getValueState(I.getTrueValue()); 910 ValueLatticeElement FVal = getValueState(I.getFalseValue()); 911 912 bool Changed = ValueState[&I].mergeIn(TVal); 913 Changed |= ValueState[&I].mergeIn(FVal); 914 if (Changed) 915 pushToWorkListMsg(ValueState[&I], &I); 916 } 917 918 // Handle Unary Operators. 919 void SCCPSolver::visitUnaryOperator(Instruction &I) { 920 ValueLatticeElement V0State = getValueState(I.getOperand(0)); 921 922 ValueLatticeElement &IV = ValueState[&I]; 923 // ResolvedUndefsIn might mark I as overdefined. Bail out, even if we would 924 // discover a concrete value later. 925 if (isOverdefined(IV)) 926 return (void)markOverdefined(&I); 927 928 if (isConstant(V0State)) { 929 Constant *C = ConstantExpr::get(I.getOpcode(), getConstant(V0State)); 930 931 // op Y -> undef. 932 if (isa<UndefValue>(C)) 933 return; 934 return (void)markConstant(IV, &I, C); 935 } 936 937 // If something is undef, wait for it to resolve. 938 if (!isOverdefined(V0State)) 939 return; 940 941 markOverdefined(&I); 942 } 943 944 // Handle Binary Operators. 945 void SCCPSolver::visitBinaryOperator(Instruction &I) { 946 ValueLatticeElement V1State = getValueState(I.getOperand(0)); 947 ValueLatticeElement V2State = getValueState(I.getOperand(1)); 948 949 ValueLatticeElement &IV = ValueState[&I]; 950 if (IV.isOverdefined()) 951 return; 952 953 // If something is undef, wait for it to resolve. 954 if (V1State.isUnknownOrUndef() || V2State.isUnknownOrUndef()) 955 return; 956 957 if (V1State.isOverdefined() && V2State.isOverdefined()) 958 return (void)markOverdefined(&I); 959 960 // If either of the operands is a constant, try to fold it to a constant. 961 // TODO: Use information from notconstant better. 962 if ((V1State.isConstant() || V2State.isConstant())) { 963 Value *V1 = isConstant(V1State) ? getConstant(V1State) : I.getOperand(0); 964 Value *V2 = isConstant(V2State) ? getConstant(V2State) : I.getOperand(1); 965 Value *R = SimplifyBinOp(I.getOpcode(), V1, V2, SimplifyQuery(DL)); 966 auto *C = dyn_cast_or_null<Constant>(R); 967 if (C) { 968 // X op Y -> undef. 969 if (isa<UndefValue>(C)) 970 return; 971 // Conservatively assume that the result may be based on operands that may 972 // be undef. Note that we use mergeInValue to combine the constant with 973 // the existing lattice value for I, as different constants might be found 974 // after one of the operands go to overdefined, e.g. due to one operand 975 // being a special floating value. 976 ValueLatticeElement NewV; 977 NewV.markConstant(C, /*MayIncludeUndef=*/true); 978 return (void)mergeInValue(&I, NewV); 979 } 980 } 981 982 // Only use ranges for binary operators on integers. 983 if (!I.getType()->isIntegerTy()) 984 return markOverdefined(&I); 985 986 // Try to simplify to a constant range. 987 ConstantRange A = ConstantRange::getFull(I.getType()->getScalarSizeInBits()); 988 ConstantRange B = ConstantRange::getFull(I.getType()->getScalarSizeInBits()); 989 if (V1State.isConstantRange()) 990 A = V1State.getConstantRange(); 991 if (V2State.isConstantRange()) 992 B = V2State.getConstantRange(); 993 994 ConstantRange R = A.binaryOp(cast<BinaryOperator>(&I)->getOpcode(), B); 995 mergeInValue(&I, ValueLatticeElement::getRange(R)); 996 997 // TODO: Currently we do not exploit special values that produce something 998 // better than overdefined with an overdefined operand for vector or floating 999 // point types, like and <4 x i32> overdefined, zeroinitializer. 1000 } 1001 1002 // Handle ICmpInst instruction. 1003 void SCCPSolver::visitCmpInst(CmpInst &I) { 1004 // Do not cache this lookup, getValueState calls later in the function might 1005 // invalidate the reference. 1006 if (isOverdefined(ValueState[&I])) 1007 return (void)markOverdefined(&I); 1008 1009 Value *Op1 = I.getOperand(0); 1010 Value *Op2 = I.getOperand(1); 1011 1012 // For parameters, use ParamState which includes constant range info if 1013 // available. 1014 auto V1State = getValueState(Op1); 1015 auto V2State = getValueState(Op2); 1016 1017 Constant *C = V1State.getCompare(I.getPredicate(), I.getType(), V2State); 1018 if (C) { 1019 if (isa<UndefValue>(C)) 1020 return; 1021 ValueLatticeElement CV; 1022 CV.markConstant(C); 1023 mergeInValue(&I, CV); 1024 return; 1025 } 1026 1027 // If operands are still unknown, wait for it to resolve. 1028 if ((V1State.isUnknownOrUndef() || V2State.isUnknownOrUndef()) && 1029 !isConstant(ValueState[&I])) 1030 return; 1031 1032 markOverdefined(&I); 1033 } 1034 1035 // Handle getelementptr instructions. If all operands are constants then we 1036 // can turn this into a getelementptr ConstantExpr. 1037 void SCCPSolver::visitGetElementPtrInst(GetElementPtrInst &I) { 1038 if (isOverdefined(ValueState[&I])) 1039 return (void)markOverdefined(&I); 1040 1041 SmallVector<Constant*, 8> Operands; 1042 Operands.reserve(I.getNumOperands()); 1043 1044 for (unsigned i = 0, e = I.getNumOperands(); i != e; ++i) { 1045 ValueLatticeElement State = getValueState(I.getOperand(i)); 1046 if (State.isUnknownOrUndef()) 1047 return; // Operands are not resolved yet. 1048 1049 if (isOverdefined(State)) 1050 return (void)markOverdefined(&I); 1051 1052 if (Constant *C = getConstant(State)) { 1053 Operands.push_back(C); 1054 continue; 1055 } 1056 1057 return (void)markOverdefined(&I); 1058 } 1059 1060 Constant *Ptr = Operands[0]; 1061 auto Indices = makeArrayRef(Operands.begin() + 1, Operands.end()); 1062 Constant *C = 1063 ConstantExpr::getGetElementPtr(I.getSourceElementType(), Ptr, Indices); 1064 if (isa<UndefValue>(C)) 1065 return; 1066 markConstant(&I, C); 1067 } 1068 1069 void SCCPSolver::visitStoreInst(StoreInst &SI) { 1070 // If this store is of a struct, ignore it. 1071 if (SI.getOperand(0)->getType()->isStructTy()) 1072 return; 1073 1074 if (TrackedGlobals.empty() || !isa<GlobalVariable>(SI.getOperand(1))) 1075 return; 1076 1077 // ResolvedUndefsIn might mark I as overdefined. Bail out, even if we would 1078 // discover a concrete value later. 1079 if (isOverdefined(ValueState[&SI])) 1080 return (void)markOverdefined(&SI); 1081 1082 GlobalVariable *GV = cast<GlobalVariable>(SI.getOperand(1)); 1083 auto I = TrackedGlobals.find(GV); 1084 if (I == TrackedGlobals.end()) 1085 return; 1086 1087 // Get the value we are storing into the global, then merge it. 1088 mergeInValue(I->second, GV, getValueState(SI.getOperand(0))); 1089 if (isOverdefined(I->second)) 1090 TrackedGlobals.erase(I); // No need to keep tracking this! 1091 } 1092 1093 // Handle load instructions. If the operand is a constant pointer to a constant 1094 // global, we can replace the load with the loaded constant value! 1095 void SCCPSolver::visitLoadInst(LoadInst &I) { 1096 // If this load is of a struct, just mark the result overdefined. 1097 if (I.getType()->isStructTy()) 1098 return (void)markOverdefined(&I); 1099 1100 // ResolvedUndefsIn might mark I as overdefined. Bail out, even if we would 1101 // discover a concrete value later. 1102 if (isOverdefined(ValueState[&I])) 1103 return (void)markOverdefined(&I); 1104 1105 ValueLatticeElement PtrVal = getValueState(I.getOperand(0)); 1106 if (PtrVal.isUnknownOrUndef()) 1107 return; // The pointer is not resolved yet! 1108 1109 ValueLatticeElement &IV = ValueState[&I]; 1110 1111 if (!isConstant(PtrVal) || I.isVolatile()) 1112 return (void)markOverdefined(IV, &I); 1113 1114 Constant *Ptr = getConstant(PtrVal); 1115 1116 // load null is undefined. 1117 if (isa<ConstantPointerNull>(Ptr)) { 1118 if (NullPointerIsDefined(I.getFunction(), I.getPointerAddressSpace())) 1119 return (void)markOverdefined(IV, &I); 1120 else 1121 return; 1122 } 1123 1124 // Transform load (constant global) into the value loaded. 1125 if (auto *GV = dyn_cast<GlobalVariable>(Ptr)) { 1126 if (!TrackedGlobals.empty()) { 1127 // If we are tracking this global, merge in the known value for it. 1128 auto It = TrackedGlobals.find(GV); 1129 if (It != TrackedGlobals.end()) { 1130 mergeInValue(IV, &I, It->second); 1131 return; 1132 } 1133 } 1134 } 1135 1136 // Transform load from a constant into a constant if possible. 1137 if (Constant *C = ConstantFoldLoadFromConstPtr(Ptr, I.getType(), DL)) { 1138 if (isa<UndefValue>(C)) 1139 return; 1140 return (void)markConstant(IV, &I, C); 1141 } 1142 1143 // Otherwise we cannot say for certain what value this load will produce. 1144 // Bail out. 1145 markOverdefined(IV, &I); 1146 } 1147 1148 void SCCPSolver::visitCallSite(CallSite CS) { 1149 handleCallResult(CS); 1150 handleCallArguments(CS); 1151 } 1152 1153 void SCCPSolver::handleCallOverdefined(CallSite CS) { 1154 Function *F = CS.getCalledFunction(); 1155 Instruction *I = CS.getInstruction(); 1156 1157 // Void return and not tracking callee, just bail. 1158 if (I->getType()->isVoidTy()) 1159 return; 1160 1161 // Otherwise, if we have a single return value case, and if the function is 1162 // a declaration, maybe we can constant fold it. 1163 if (F && F->isDeclaration() && !I->getType()->isStructTy() && 1164 canConstantFoldCallTo(cast<CallBase>(CS.getInstruction()), F)) { 1165 SmallVector<Constant *, 8> Operands; 1166 for (CallSite::arg_iterator AI = CS.arg_begin(), E = CS.arg_end(); AI != E; 1167 ++AI) { 1168 if (AI->get()->getType()->isStructTy()) 1169 return markOverdefined(I); // Can't handle struct args. 1170 ValueLatticeElement State = getValueState(*AI); 1171 1172 if (State.isUnknownOrUndef()) 1173 return; // Operands are not resolved yet. 1174 if (isOverdefined(State)) 1175 return (void)markOverdefined(I); 1176 assert(isConstant(State) && "Unknown state!"); 1177 Operands.push_back(getConstant(State)); 1178 } 1179 1180 if (isOverdefined(getValueState(I))) 1181 return (void)markOverdefined(I); 1182 1183 // If we can constant fold this, mark the result of the call as a 1184 // constant. 1185 if (Constant *C = ConstantFoldCall(cast<CallBase>(CS.getInstruction()), F, 1186 Operands, &GetTLI(*F))) { 1187 // call -> undef. 1188 if (isa<UndefValue>(C)) 1189 return; 1190 return (void)markConstant(I, C); 1191 } 1192 } 1193 1194 // Otherwise, we don't know anything about this call, mark it overdefined. 1195 return (void)markOverdefined(I); 1196 } 1197 1198 void SCCPSolver::handleCallArguments(CallSite CS) { 1199 Function *F = CS.getCalledFunction(); 1200 // If this is a local function that doesn't have its address taken, mark its 1201 // entry block executable and merge in the actual arguments to the call into 1202 // the formal arguments of the function. 1203 if (!TrackingIncomingArguments.empty() && 1204 TrackingIncomingArguments.count(F)) { 1205 MarkBlockExecutable(&F->front()); 1206 1207 // Propagate information from this call site into the callee. 1208 CallSite::arg_iterator CAI = CS.arg_begin(); 1209 for (Function::arg_iterator AI = F->arg_begin(), E = F->arg_end(); AI != E; 1210 ++AI, ++CAI) { 1211 // If this argument is byval, and if the function is not readonly, there 1212 // will be an implicit copy formed of the input aggregate. 1213 if (AI->hasByValAttr() && !F->onlyReadsMemory()) { 1214 markOverdefined(&*AI); 1215 continue; 1216 } 1217 1218 if (auto *STy = dyn_cast<StructType>(AI->getType())) { 1219 for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) { 1220 ValueLatticeElement CallArg = getStructValueState(*CAI, i); 1221 mergeInValue(getStructValueState(&*AI, i), &*AI, CallArg); 1222 } 1223 } else 1224 mergeInValue(&*AI, getValueState(*CAI), false); 1225 } 1226 } 1227 } 1228 1229 void SCCPSolver::handleCallResult(CallSite CS) { 1230 Function *F = CS.getCalledFunction(); 1231 Instruction *I = CS.getInstruction(); 1232 1233 if (auto *II = dyn_cast<IntrinsicInst>(I)) { 1234 if (II->getIntrinsicID() == Intrinsic::ssa_copy) { 1235 if (ValueState[I].isOverdefined()) 1236 return; 1237 1238 Value *CopyOf = I->getOperand(0); 1239 auto *PI = getPredicateInfoFor(I); 1240 auto *PBranch = dyn_cast_or_null<PredicateBranch>(PI); 1241 if (!PI || !PBranch) { 1242 mergeInValue(ValueState[I], I, getValueState(CopyOf)); 1243 return; 1244 } 1245 1246 // Everything below relies on the condition being a comparison. 1247 auto *Cmp = dyn_cast<CmpInst>(PBranch->Condition); 1248 if (!Cmp) { 1249 mergeInValue(ValueState[I], I, getValueState(CopyOf)); 1250 return; 1251 } 1252 1253 Value *CmpOp0 = Cmp->getOperand(0); 1254 Value *CmpOp1 = Cmp->getOperand(1); 1255 if (CopyOf != CmpOp0 && CopyOf != CmpOp1) { 1256 mergeInValue(ValueState[I], I, getValueState(CopyOf)); 1257 return; 1258 } 1259 1260 auto Pred = Cmp->getPredicate(); 1261 if (CmpOp0 != CopyOf) { 1262 std::swap(CmpOp0, CmpOp1); 1263 Pred = Cmp->getSwappedPredicate(); 1264 } 1265 1266 // Wait until CmpOp1 is resolved. 1267 if (getValueState(CmpOp1).isUnknown()) { 1268 addAdditionalUser(CmpOp1, I); 1269 return; 1270 } 1271 1272 if (!PBranch->TrueEdge) 1273 Pred = CmpInst::getInversePredicate(Pred); 1274 1275 ValueLatticeElement CondVal = getValueState(CmpOp1); 1276 ValueLatticeElement &IV = ValueState[I]; 1277 ValueLatticeElement OriginalVal = getValueState(CopyOf); 1278 if (CondVal.isConstantRange() || OriginalVal.isConstantRange()) { 1279 auto NewCR = 1280 ConstantRange::getFull(DL.getTypeSizeInBits(CopyOf->getType())); 1281 1282 // Get the range imposed by the condition. 1283 if (CondVal.isConstantRange()) 1284 NewCR = ConstantRange::makeAllowedICmpRegion( 1285 Pred, CondVal.getConstantRange()); 1286 1287 // Combine range info for the original value with the new range from the 1288 // condition. 1289 auto OriginalCR = OriginalVal.isConstantRange() 1290 ? OriginalVal.getConstantRange() 1291 : ConstantRange::getFull( 1292 DL.getTypeSizeInBits(CopyOf->getType())); 1293 NewCR = NewCR.intersectWith(OriginalCR); 1294 1295 addAdditionalUser(CmpOp1, I); 1296 // TODO: Actually filp MayIncludeUndef for the created range to false, 1297 // once most places in the optimizer respect the branches on 1298 // undef/poison are UB rule. The reason why the new range cannot be 1299 // undef is as follows below: 1300 // The new range is based on a branch condition. That guarantees that 1301 // neither of the compare operands can be undef in the branch targets, 1302 // unless we have conditions that are always true/false (e.g. icmp ule 1303 // i32, %a, i32_max). For the latter overdefined/empty range will be 1304 // inferred, but the branch will get folded accordingly anyways. 1305 mergeInValue( 1306 IV, I, 1307 ValueLatticeElement::getRange(NewCR, /*MayIncludeUndef=*/true)); 1308 return; 1309 } else if (Pred == CmpInst::ICMP_EQ && CondVal.isConstant()) { 1310 // For non-integer values or integer constant expressions, only 1311 // propagate equal constants. 1312 addAdditionalUser(CmpOp1, I); 1313 mergeInValue(IV, I, CondVal); 1314 return; 1315 } 1316 1317 return (void)mergeInValue(IV, I, getValueState(CopyOf)); 1318 } 1319 } 1320 1321 // The common case is that we aren't tracking the callee, either because we 1322 // are not doing interprocedural analysis or the callee is indirect, or is 1323 // external. Handle these cases first. 1324 if (!F || F->isDeclaration()) 1325 return handleCallOverdefined(CS); 1326 1327 // If this is a single/zero retval case, see if we're tracking the function. 1328 if (auto *STy = dyn_cast<StructType>(F->getReturnType())) { 1329 if (!MRVFunctionsTracked.count(F)) 1330 return handleCallOverdefined(CS); // Not tracking this callee. 1331 1332 // If we are tracking this callee, propagate the result of the function 1333 // into this call site. 1334 for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) 1335 mergeInValue(getStructValueState(I, i), I, 1336 TrackedMultipleRetVals[std::make_pair(F, i)]); 1337 } else { 1338 auto TFRVI = TrackedRetVals.find(F); 1339 if (TFRVI == TrackedRetVals.end()) 1340 return handleCallOverdefined(CS); // Not tracking this callee. 1341 1342 // If so, propagate the return value of the callee into this call result. 1343 mergeInValue(I, TFRVI->second); 1344 } 1345 } 1346 1347 void SCCPSolver::Solve() { 1348 // Process the work lists until they are empty! 1349 while (!BBWorkList.empty() || !InstWorkList.empty() || 1350 !OverdefinedInstWorkList.empty()) { 1351 // Process the overdefined instruction's work list first, which drives other 1352 // things to overdefined more quickly. 1353 while (!OverdefinedInstWorkList.empty()) { 1354 Value *I = OverdefinedInstWorkList.pop_back_val(); 1355 1356 LLVM_DEBUG(dbgs() << "\nPopped off OI-WL: " << *I << '\n'); 1357 1358 // "I" got into the work list because it either made the transition from 1359 // bottom to constant, or to overdefined. 1360 // 1361 // Anything on this worklist that is overdefined need not be visited 1362 // since all of its users will have already been marked as overdefined 1363 // Update all of the users of this instruction's value. 1364 // 1365 markUsersAsChanged(I); 1366 } 1367 1368 // Process the instruction work list. 1369 while (!InstWorkList.empty()) { 1370 Value *I = InstWorkList.pop_back_val(); 1371 1372 LLVM_DEBUG(dbgs() << "\nPopped off I-WL: " << *I << '\n'); 1373 1374 // "I" got into the work list because it made the transition from undef to 1375 // constant. 1376 // 1377 // Anything on this worklist that is overdefined need not be visited 1378 // since all of its users will have already been marked as overdefined. 1379 // Update all of the users of this instruction's value. 1380 // 1381 if (I->getType()->isStructTy() || !getValueState(I).isOverdefined()) 1382 markUsersAsChanged(I); 1383 } 1384 1385 // Process the basic block work list. 1386 while (!BBWorkList.empty()) { 1387 BasicBlock *BB = BBWorkList.back(); 1388 BBWorkList.pop_back(); 1389 1390 LLVM_DEBUG(dbgs() << "\nPopped off BBWL: " << *BB << '\n'); 1391 1392 // Notify all instructions in this basic block that they are newly 1393 // executable. 1394 visit(BB); 1395 } 1396 } 1397 } 1398 1399 /// ResolvedUndefsIn - While solving the dataflow for a function, we assume 1400 /// that branches on undef values cannot reach any of their successors. 1401 /// However, this is not a safe assumption. After we solve dataflow, this 1402 /// method should be use to handle this. If this returns true, the solver 1403 /// should be rerun. 1404 /// 1405 /// This method handles this by finding an unresolved branch and marking it one 1406 /// of the edges from the block as being feasible, even though the condition 1407 /// doesn't say it would otherwise be. This allows SCCP to find the rest of the 1408 /// CFG and only slightly pessimizes the analysis results (by marking one, 1409 /// potentially infeasible, edge feasible). This cannot usefully modify the 1410 /// constraints on the condition of the branch, as that would impact other users 1411 /// of the value. 1412 /// 1413 /// This scan also checks for values that use undefs. It conservatively marks 1414 /// them as overdefined. 1415 bool SCCPSolver::ResolvedUndefsIn(Function &F) { 1416 for (BasicBlock &BB : F) { 1417 if (!BBExecutable.count(&BB)) 1418 continue; 1419 1420 for (Instruction &I : BB) { 1421 // Look for instructions which produce undef values. 1422 if (I.getType()->isVoidTy()) continue; 1423 1424 if (auto *STy = dyn_cast<StructType>(I.getType())) { 1425 // Only a few things that can be structs matter for undef. 1426 1427 // Tracked calls must never be marked overdefined in ResolvedUndefsIn. 1428 if (CallSite CS = CallSite(&I)) 1429 if (Function *F = CS.getCalledFunction()) 1430 if (MRVFunctionsTracked.count(F)) 1431 continue; 1432 1433 // extractvalue and insertvalue don't need to be marked; they are 1434 // tracked as precisely as their operands. 1435 if (isa<ExtractValueInst>(I) || isa<InsertValueInst>(I)) 1436 continue; 1437 // Send the results of everything else to overdefined. We could be 1438 // more precise than this but it isn't worth bothering. 1439 for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) { 1440 ValueLatticeElement &LV = getStructValueState(&I, i); 1441 if (LV.isUnknownOrUndef()) 1442 markOverdefined(LV, &I); 1443 } 1444 continue; 1445 } 1446 1447 ValueLatticeElement &LV = getValueState(&I); 1448 if (!LV.isUnknownOrUndef()) 1449 continue; 1450 1451 // There are two reasons a call can have an undef result 1452 // 1. It could be tracked. 1453 // 2. It could be constant-foldable. 1454 // Because of the way we solve return values, tracked calls must 1455 // never be marked overdefined in ResolvedUndefsIn. 1456 if (CallSite CS = CallSite(&I)) 1457 if (Function *F = CS.getCalledFunction()) 1458 if (TrackedRetVals.count(F)) 1459 continue; 1460 1461 if (isa<LoadInst>(I)) { 1462 // A load here means one of two things: a load of undef from a global, 1463 // a load from an unknown pointer. Either way, having it return undef 1464 // is okay. 1465 continue; 1466 } 1467 1468 markOverdefined(&I); 1469 return true; 1470 } 1471 1472 // Check to see if we have a branch or switch on an undefined value. If so 1473 // we force the branch to go one way or the other to make the successor 1474 // values live. It doesn't really matter which way we force it. 1475 Instruction *TI = BB.getTerminator(); 1476 if (auto *BI = dyn_cast<BranchInst>(TI)) { 1477 if (!BI->isConditional()) continue; 1478 if (!getValueState(BI->getCondition()).isUnknownOrUndef()) 1479 continue; 1480 1481 // If the input to SCCP is actually branch on undef, fix the undef to 1482 // false. 1483 if (isa<UndefValue>(BI->getCondition())) { 1484 BI->setCondition(ConstantInt::getFalse(BI->getContext())); 1485 markEdgeExecutable(&BB, TI->getSuccessor(1)); 1486 return true; 1487 } 1488 1489 // Otherwise, it is a branch on a symbolic value which is currently 1490 // considered to be undef. Make sure some edge is executable, so a 1491 // branch on "undef" always flows somewhere. 1492 // FIXME: Distinguish between dead code and an LLVM "undef" value. 1493 BasicBlock *DefaultSuccessor = TI->getSuccessor(1); 1494 if (markEdgeExecutable(&BB, DefaultSuccessor)) 1495 return true; 1496 1497 continue; 1498 } 1499 1500 if (auto *IBR = dyn_cast<IndirectBrInst>(TI)) { 1501 // Indirect branch with no successor ?. Its ok to assume it branches 1502 // to no target. 1503 if (IBR->getNumSuccessors() < 1) 1504 continue; 1505 1506 if (!getValueState(IBR->getAddress()).isUnknownOrUndef()) 1507 continue; 1508 1509 // If the input to SCCP is actually branch on undef, fix the undef to 1510 // the first successor of the indirect branch. 1511 if (isa<UndefValue>(IBR->getAddress())) { 1512 IBR->setAddress(BlockAddress::get(IBR->getSuccessor(0))); 1513 markEdgeExecutable(&BB, IBR->getSuccessor(0)); 1514 return true; 1515 } 1516 1517 // Otherwise, it is a branch on a symbolic value which is currently 1518 // considered to be undef. Make sure some edge is executable, so a 1519 // branch on "undef" always flows somewhere. 1520 // FIXME: IndirectBr on "undef" doesn't actually need to go anywhere: 1521 // we can assume the branch has undefined behavior instead. 1522 BasicBlock *DefaultSuccessor = IBR->getSuccessor(0); 1523 if (markEdgeExecutable(&BB, DefaultSuccessor)) 1524 return true; 1525 1526 continue; 1527 } 1528 1529 if (auto *SI = dyn_cast<SwitchInst>(TI)) { 1530 if (!SI->getNumCases() || 1531 !getValueState(SI->getCondition()).isUnknownOrUndef()) 1532 continue; 1533 1534 // If the input to SCCP is actually switch on undef, fix the undef to 1535 // the first constant. 1536 if (isa<UndefValue>(SI->getCondition())) { 1537 SI->setCondition(SI->case_begin()->getCaseValue()); 1538 markEdgeExecutable(&BB, SI->case_begin()->getCaseSuccessor()); 1539 return true; 1540 } 1541 1542 // Otherwise, it is a branch on a symbolic value which is currently 1543 // considered to be undef. Make sure some edge is executable, so a 1544 // branch on "undef" always flows somewhere. 1545 // FIXME: Distinguish between dead code and an LLVM "undef" value. 1546 BasicBlock *DefaultSuccessor = SI->case_begin()->getCaseSuccessor(); 1547 if (markEdgeExecutable(&BB, DefaultSuccessor)) 1548 return true; 1549 1550 continue; 1551 } 1552 } 1553 1554 return false; 1555 } 1556 1557 static bool tryToReplaceWithConstant(SCCPSolver &Solver, Value *V) { 1558 Constant *Const = nullptr; 1559 if (V->getType()->isStructTy()) { 1560 std::vector<ValueLatticeElement> IVs = Solver.getStructLatticeValueFor(V); 1561 if (any_of(IVs, 1562 [](const ValueLatticeElement &LV) { return isOverdefined(LV); })) 1563 return false; 1564 std::vector<Constant *> ConstVals; 1565 auto *ST = cast<StructType>(V->getType()); 1566 for (unsigned i = 0, e = ST->getNumElements(); i != e; ++i) { 1567 ValueLatticeElement V = IVs[i]; 1568 ConstVals.push_back(isConstant(V) 1569 ? Solver.getConstant(V) 1570 : UndefValue::get(ST->getElementType(i))); 1571 } 1572 Const = ConstantStruct::get(ST, ConstVals); 1573 } else { 1574 const ValueLatticeElement &IV = Solver.getLatticeValueFor(V); 1575 if (isOverdefined(IV)) 1576 return false; 1577 1578 Const = 1579 isConstant(IV) ? Solver.getConstant(IV) : UndefValue::get(V->getType()); 1580 } 1581 assert(Const && "Constant is nullptr here!"); 1582 1583 // Replacing `musttail` instructions with constant breaks `musttail` invariant 1584 // unless the call itself can be removed 1585 CallInst *CI = dyn_cast<CallInst>(V); 1586 if (CI && CI->isMustTailCall() && !CI->isSafeToRemove()) { 1587 CallSite CS(CI); 1588 Function *F = CS.getCalledFunction(); 1589 1590 // Don't zap returns of the callee 1591 if (F) 1592 Solver.AddMustTailCallee(F); 1593 1594 LLVM_DEBUG(dbgs() << " Can\'t treat the result of musttail call : " << *CI 1595 << " as a constant\n"); 1596 return false; 1597 } 1598 1599 LLVM_DEBUG(dbgs() << " Constant: " << *Const << " = " << *V << '\n'); 1600 1601 // Replaces all of the uses of a variable with uses of the constant. 1602 V->replaceAllUsesWith(Const); 1603 return true; 1604 } 1605 1606 // runSCCP() - Run the Sparse Conditional Constant Propagation algorithm, 1607 // and return true if the function was modified. 1608 static bool runSCCP(Function &F, const DataLayout &DL, 1609 const TargetLibraryInfo *TLI) { 1610 LLVM_DEBUG(dbgs() << "SCCP on function '" << F.getName() << "'\n"); 1611 SCCPSolver Solver( 1612 DL, [TLI](Function &F) -> const TargetLibraryInfo & { return *TLI; }, 1613 F.getContext()); 1614 1615 // Mark the first block of the function as being executable. 1616 Solver.MarkBlockExecutable(&F.front()); 1617 1618 // Mark all arguments to the function as being overdefined. 1619 for (Argument &AI : F.args()) 1620 Solver.markOverdefined(&AI); 1621 1622 // Solve for constants. 1623 bool ResolvedUndefs = true; 1624 while (ResolvedUndefs) { 1625 Solver.Solve(); 1626 LLVM_DEBUG(dbgs() << "RESOLVING UNDEFs\n"); 1627 ResolvedUndefs = Solver.ResolvedUndefsIn(F); 1628 } 1629 1630 bool MadeChanges = false; 1631 1632 // If we decided that there are basic blocks that are dead in this function, 1633 // delete their contents now. Note that we cannot actually delete the blocks, 1634 // as we cannot modify the CFG of the function. 1635 1636 for (BasicBlock &BB : F) { 1637 if (!Solver.isBlockExecutable(&BB)) { 1638 LLVM_DEBUG(dbgs() << " BasicBlock Dead:" << BB); 1639 1640 ++NumDeadBlocks; 1641 NumInstRemoved += removeAllNonTerminatorAndEHPadInstructions(&BB); 1642 1643 MadeChanges = true; 1644 continue; 1645 } 1646 1647 // Iterate over all of the instructions in a function, replacing them with 1648 // constants if we have found them to be of constant values. 1649 for (BasicBlock::iterator BI = BB.begin(), E = BB.end(); BI != E;) { 1650 Instruction *Inst = &*BI++; 1651 if (Inst->getType()->isVoidTy() || Inst->isTerminator()) 1652 continue; 1653 1654 if (tryToReplaceWithConstant(Solver, Inst)) { 1655 if (isInstructionTriviallyDead(Inst)) 1656 Inst->eraseFromParent(); 1657 // Hey, we just changed something! 1658 MadeChanges = true; 1659 ++NumInstRemoved; 1660 } 1661 } 1662 } 1663 1664 return MadeChanges; 1665 } 1666 1667 PreservedAnalyses SCCPPass::run(Function &F, FunctionAnalysisManager &AM) { 1668 const DataLayout &DL = F.getParent()->getDataLayout(); 1669 auto &TLI = AM.getResult<TargetLibraryAnalysis>(F); 1670 if (!runSCCP(F, DL, &TLI)) 1671 return PreservedAnalyses::all(); 1672 1673 auto PA = PreservedAnalyses(); 1674 PA.preserve<GlobalsAA>(); 1675 PA.preserveSet<CFGAnalyses>(); 1676 return PA; 1677 } 1678 1679 namespace { 1680 1681 //===--------------------------------------------------------------------===// 1682 // 1683 /// SCCP Class - This class uses the SCCPSolver to implement a per-function 1684 /// Sparse Conditional Constant Propagator. 1685 /// 1686 class SCCPLegacyPass : public FunctionPass { 1687 public: 1688 // Pass identification, replacement for typeid 1689 static char ID; 1690 1691 SCCPLegacyPass() : FunctionPass(ID) { 1692 initializeSCCPLegacyPassPass(*PassRegistry::getPassRegistry()); 1693 } 1694 1695 void getAnalysisUsage(AnalysisUsage &AU) const override { 1696 AU.addRequired<TargetLibraryInfoWrapperPass>(); 1697 AU.addPreserved<GlobalsAAWrapperPass>(); 1698 AU.setPreservesCFG(); 1699 } 1700 1701 // runOnFunction - Run the Sparse Conditional Constant Propagation 1702 // algorithm, and return true if the function was modified. 1703 bool runOnFunction(Function &F) override { 1704 if (skipFunction(F)) 1705 return false; 1706 const DataLayout &DL = F.getParent()->getDataLayout(); 1707 const TargetLibraryInfo *TLI = 1708 &getAnalysis<TargetLibraryInfoWrapperPass>().getTLI(F); 1709 return runSCCP(F, DL, TLI); 1710 } 1711 }; 1712 1713 } // end anonymous namespace 1714 1715 char SCCPLegacyPass::ID = 0; 1716 1717 INITIALIZE_PASS_BEGIN(SCCPLegacyPass, "sccp", 1718 "Sparse Conditional Constant Propagation", false, false) 1719 INITIALIZE_PASS_DEPENDENCY(TargetLibraryInfoWrapperPass) 1720 INITIALIZE_PASS_END(SCCPLegacyPass, "sccp", 1721 "Sparse Conditional Constant Propagation", false, false) 1722 1723 // createSCCPPass - This is the public interface to this file. 1724 FunctionPass *llvm::createSCCPPass() { return new SCCPLegacyPass(); } 1725 1726 static void findReturnsToZap(Function &F, 1727 SmallVector<ReturnInst *, 8> &ReturnsToZap, 1728 SCCPSolver &Solver) { 1729 // We can only do this if we know that nothing else can call the function. 1730 if (!Solver.isArgumentTrackedFunction(&F)) 1731 return; 1732 1733 // There is a non-removable musttail call site of this function. Zapping 1734 // returns is not allowed. 1735 if (Solver.isMustTailCallee(&F)) { 1736 LLVM_DEBUG(dbgs() << "Can't zap returns of the function : " << F.getName() 1737 << " due to present musttail call of it\n"); 1738 return; 1739 } 1740 1741 assert( 1742 all_of(F.users(), 1743 [&Solver](User *U) { 1744 if (isa<Instruction>(U) && 1745 !Solver.isBlockExecutable(cast<Instruction>(U)->getParent())) 1746 return true; 1747 // Non-callsite uses are not impacted by zapping. Also, constant 1748 // uses (like blockaddresses) could stuck around, without being 1749 // used in the underlying IR, meaning we do not have lattice 1750 // values for them. 1751 if (!CallSite(U)) 1752 return true; 1753 if (U->getType()->isStructTy()) { 1754 return all_of(Solver.getStructLatticeValueFor(U), 1755 [](const ValueLatticeElement &LV) { 1756 return !isOverdefined(LV); 1757 }); 1758 } 1759 return !isOverdefined(Solver.getLatticeValueFor(U)); 1760 }) && 1761 "We can only zap functions where all live users have a concrete value"); 1762 1763 for (BasicBlock &BB : F) { 1764 if (CallInst *CI = BB.getTerminatingMustTailCall()) { 1765 LLVM_DEBUG(dbgs() << "Can't zap return of the block due to present " 1766 << "musttail call : " << *CI << "\n"); 1767 (void)CI; 1768 return; 1769 } 1770 1771 if (auto *RI = dyn_cast<ReturnInst>(BB.getTerminator())) 1772 if (!isa<UndefValue>(RI->getOperand(0))) 1773 ReturnsToZap.push_back(RI); 1774 } 1775 } 1776 1777 // Update the condition for terminators that are branching on indeterminate 1778 // values, forcing them to use a specific edge. 1779 static void forceIndeterminateEdge(Instruction* I, SCCPSolver &Solver) { 1780 BasicBlock *Dest = nullptr; 1781 Constant *C = nullptr; 1782 if (SwitchInst *SI = dyn_cast<SwitchInst>(I)) { 1783 if (!isa<ConstantInt>(SI->getCondition())) { 1784 // Indeterminate switch; use first case value. 1785 Dest = SI->case_begin()->getCaseSuccessor(); 1786 C = SI->case_begin()->getCaseValue(); 1787 } 1788 } else if (BranchInst *BI = dyn_cast<BranchInst>(I)) { 1789 if (!isa<ConstantInt>(BI->getCondition())) { 1790 // Indeterminate branch; use false. 1791 Dest = BI->getSuccessor(1); 1792 C = ConstantInt::getFalse(BI->getContext()); 1793 } 1794 } else if (IndirectBrInst *IBR = dyn_cast<IndirectBrInst>(I)) { 1795 if (!isa<BlockAddress>(IBR->getAddress()->stripPointerCasts())) { 1796 // Indeterminate indirectbr; use successor 0. 1797 Dest = IBR->getSuccessor(0); 1798 C = BlockAddress::get(IBR->getSuccessor(0)); 1799 } 1800 } else { 1801 llvm_unreachable("Unexpected terminator instruction"); 1802 } 1803 if (C) { 1804 assert(Solver.isEdgeFeasible(I->getParent(), Dest) && 1805 "Didn't find feasible edge?"); 1806 (void)Dest; 1807 1808 I->setOperand(0, C); 1809 } 1810 } 1811 1812 bool llvm::runIPSCCP( 1813 Module &M, const DataLayout &DL, 1814 std::function<const TargetLibraryInfo &(Function &)> GetTLI, 1815 function_ref<AnalysisResultsForFn(Function &)> getAnalysis) { 1816 SCCPSolver Solver(DL, GetTLI, M.getContext()); 1817 1818 // Loop over all functions, marking arguments to those with their addresses 1819 // taken or that are external as overdefined. 1820 for (Function &F : M) { 1821 if (F.isDeclaration()) 1822 continue; 1823 1824 Solver.addAnalysis(F, getAnalysis(F)); 1825 1826 // Determine if we can track the function's return values. If so, add the 1827 // function to the solver's set of return-tracked functions. 1828 if (canTrackReturnsInterprocedurally(&F)) 1829 Solver.AddTrackedFunction(&F); 1830 1831 // Determine if we can track the function's arguments. If so, add the 1832 // function to the solver's set of argument-tracked functions. 1833 if (canTrackArgumentsInterprocedurally(&F)) { 1834 Solver.AddArgumentTrackedFunction(&F); 1835 continue; 1836 } 1837 1838 // Assume the function is called. 1839 Solver.MarkBlockExecutable(&F.front()); 1840 1841 // Assume nothing about the incoming arguments. 1842 for (Argument &AI : F.args()) 1843 Solver.markOverdefined(&AI); 1844 } 1845 1846 // Determine if we can track any of the module's global variables. If so, add 1847 // the global variables we can track to the solver's set of tracked global 1848 // variables. 1849 for (GlobalVariable &G : M.globals()) { 1850 G.removeDeadConstantUsers(); 1851 if (canTrackGlobalVariableInterprocedurally(&G)) 1852 Solver.TrackValueOfGlobalVariable(&G); 1853 } 1854 1855 // Solve for constants. 1856 bool ResolvedUndefs = true; 1857 Solver.Solve(); 1858 while (ResolvedUndefs) { 1859 LLVM_DEBUG(dbgs() << "RESOLVING UNDEFS\n"); 1860 ResolvedUndefs = false; 1861 for (Function &F : M) 1862 if (Solver.ResolvedUndefsIn(F)) { 1863 // We run Solve() after we resolved an undef in a function, because 1864 // we might deduce a fact that eliminates an undef in another function. 1865 Solver.Solve(); 1866 ResolvedUndefs = true; 1867 } 1868 } 1869 1870 bool MadeChanges = false; 1871 1872 // Iterate over all of the instructions in the module, replacing them with 1873 // constants if we have found them to be of constant values. 1874 1875 for (Function &F : M) { 1876 if (F.isDeclaration()) 1877 continue; 1878 1879 SmallVector<BasicBlock *, 512> BlocksToErase; 1880 1881 if (Solver.isBlockExecutable(&F.front())) 1882 for (Function::arg_iterator AI = F.arg_begin(), E = F.arg_end(); AI != E; 1883 ++AI) { 1884 if (!AI->use_empty() && tryToReplaceWithConstant(Solver, &*AI)) { 1885 ++IPNumArgsElimed; 1886 continue; 1887 } 1888 } 1889 1890 for (Function::iterator BB = F.begin(), E = F.end(); BB != E; ++BB) { 1891 if (!Solver.isBlockExecutable(&*BB)) { 1892 LLVM_DEBUG(dbgs() << " BasicBlock Dead:" << *BB); 1893 ++NumDeadBlocks; 1894 1895 MadeChanges = true; 1896 1897 if (&*BB != &F.front()) 1898 BlocksToErase.push_back(&*BB); 1899 continue; 1900 } 1901 1902 for (BasicBlock::iterator BI = BB->begin(), E = BB->end(); BI != E; ) { 1903 Instruction *Inst = &*BI++; 1904 if (Inst->getType()->isVoidTy()) 1905 continue; 1906 if (tryToReplaceWithConstant(Solver, Inst)) { 1907 if (Inst->isSafeToRemove()) 1908 Inst->eraseFromParent(); 1909 // Hey, we just changed something! 1910 MadeChanges = true; 1911 ++IPNumInstRemoved; 1912 } 1913 } 1914 } 1915 1916 DomTreeUpdater DTU = Solver.getDTU(F); 1917 // Change dead blocks to unreachable. We do it after replacing constants 1918 // in all executable blocks, because changeToUnreachable may remove PHI 1919 // nodes in executable blocks we found values for. The function's entry 1920 // block is not part of BlocksToErase, so we have to handle it separately. 1921 for (BasicBlock *BB : BlocksToErase) { 1922 NumInstRemoved += 1923 changeToUnreachable(BB->getFirstNonPHI(), /*UseLLVMTrap=*/false, 1924 /*PreserveLCSSA=*/false, &DTU); 1925 } 1926 if (!Solver.isBlockExecutable(&F.front())) 1927 NumInstRemoved += changeToUnreachable(F.front().getFirstNonPHI(), 1928 /*UseLLVMTrap=*/false, 1929 /*PreserveLCSSA=*/false, &DTU); 1930 1931 // Now that all instructions in the function are constant folded, 1932 // use ConstantFoldTerminator to get rid of in-edges, record DT updates and 1933 // delete dead BBs. 1934 for (BasicBlock *DeadBB : BlocksToErase) { 1935 // If there are any PHI nodes in this successor, drop entries for BB now. 1936 for (Value::user_iterator UI = DeadBB->user_begin(), 1937 UE = DeadBB->user_end(); 1938 UI != UE;) { 1939 // Grab the user and then increment the iterator early, as the user 1940 // will be deleted. Step past all adjacent uses from the same user. 1941 auto *I = dyn_cast<Instruction>(*UI); 1942 do { ++UI; } while (UI != UE && *UI == I); 1943 1944 // Ignore blockaddress users; BasicBlock's dtor will handle them. 1945 if (!I) continue; 1946 1947 // If we have forced an edge for an indeterminate value, then force the 1948 // terminator to fold to that edge. 1949 forceIndeterminateEdge(I, Solver); 1950 BasicBlock *InstBB = I->getParent(); 1951 bool Folded = ConstantFoldTerminator(InstBB, 1952 /*DeleteDeadConditions=*/false, 1953 /*TLI=*/nullptr, &DTU); 1954 assert(Folded && 1955 "Expect TermInst on constantint or blockaddress to be folded"); 1956 (void) Folded; 1957 // If we folded the terminator to an unconditional branch to another 1958 // dead block, replace it with Unreachable, to avoid trying to fold that 1959 // branch again. 1960 BranchInst *BI = cast<BranchInst>(InstBB->getTerminator()); 1961 if (BI && BI->isUnconditional() && 1962 !Solver.isBlockExecutable(BI->getSuccessor(0))) { 1963 InstBB->getTerminator()->eraseFromParent(); 1964 new UnreachableInst(InstBB->getContext(), InstBB); 1965 } 1966 } 1967 // Mark dead BB for deletion. 1968 DTU.deleteBB(DeadBB); 1969 } 1970 1971 for (BasicBlock &BB : F) { 1972 for (BasicBlock::iterator BI = BB.begin(), E = BB.end(); BI != E;) { 1973 Instruction *Inst = &*BI++; 1974 if (Solver.getPredicateInfoFor(Inst)) { 1975 if (auto *II = dyn_cast<IntrinsicInst>(Inst)) { 1976 if (II->getIntrinsicID() == Intrinsic::ssa_copy) { 1977 Value *Op = II->getOperand(0); 1978 Inst->replaceAllUsesWith(Op); 1979 Inst->eraseFromParent(); 1980 } 1981 } 1982 } 1983 } 1984 } 1985 } 1986 1987 // If we inferred constant or undef return values for a function, we replaced 1988 // all call uses with the inferred value. This means we don't need to bother 1989 // actually returning anything from the function. Replace all return 1990 // instructions with return undef. 1991 // 1992 // Do this in two stages: first identify the functions we should process, then 1993 // actually zap their returns. This is important because we can only do this 1994 // if the address of the function isn't taken. In cases where a return is the 1995 // last use of a function, the order of processing functions would affect 1996 // whether other functions are optimizable. 1997 SmallVector<ReturnInst*, 8> ReturnsToZap; 1998 1999 for (const auto &I : Solver.getTrackedRetVals()) { 2000 Function *F = I.first; 2001 if (isOverdefined(I.second) || F->getReturnType()->isVoidTy()) 2002 continue; 2003 findReturnsToZap(*F, ReturnsToZap, Solver); 2004 } 2005 2006 for (auto F : Solver.getMRVFunctionsTracked()) { 2007 assert(F->getReturnType()->isStructTy() && 2008 "The return type should be a struct"); 2009 StructType *STy = cast<StructType>(F->getReturnType()); 2010 if (Solver.isStructLatticeConstant(F, STy)) 2011 findReturnsToZap(*F, ReturnsToZap, Solver); 2012 } 2013 2014 // Zap all returns which we've identified as zap to change. 2015 for (unsigned i = 0, e = ReturnsToZap.size(); i != e; ++i) { 2016 Function *F = ReturnsToZap[i]->getParent()->getParent(); 2017 ReturnsToZap[i]->setOperand(0, UndefValue::get(F->getReturnType())); 2018 } 2019 2020 // If we inferred constant or undef values for globals variables, we can 2021 // delete the global and any stores that remain to it. 2022 for (auto &I : make_early_inc_range(Solver.getTrackedGlobals())) { 2023 GlobalVariable *GV = I.first; 2024 if (isOverdefined(I.second)) 2025 continue; 2026 LLVM_DEBUG(dbgs() << "Found that GV '" << GV->getName() 2027 << "' is constant!\n"); 2028 while (!GV->use_empty()) { 2029 StoreInst *SI = cast<StoreInst>(GV->user_back()); 2030 SI->eraseFromParent(); 2031 } 2032 M.getGlobalList().erase(GV); 2033 ++IPNumGlobalConst; 2034 } 2035 2036 return MadeChanges; 2037 } 2038