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/SetVector.h" 27 #include "llvm/ADT/SmallPtrSet.h" 28 #include "llvm/ADT/SmallVector.h" 29 #include "llvm/ADT/Statistic.h" 30 #include "llvm/Analysis/ConstantFolding.h" 31 #include "llvm/Analysis/DomTreeUpdater.h" 32 #include "llvm/Analysis/GlobalsModRef.h" 33 #include "llvm/Analysis/InstructionSimplify.h" 34 #include "llvm/Analysis/TargetLibraryInfo.h" 35 #include "llvm/Analysis/ValueLattice.h" 36 #include "llvm/Analysis/ValueLatticeUtils.h" 37 #include "llvm/Analysis/ValueTracking.h" 38 #include "llvm/IR/BasicBlock.h" 39 #include "llvm/IR/Constant.h" 40 #include "llvm/IR/Constants.h" 41 #include "llvm/IR/DataLayout.h" 42 #include "llvm/IR/DerivedTypes.h" 43 #include "llvm/IR/Function.h" 44 #include "llvm/IR/GlobalVariable.h" 45 #include "llvm/IR/InstVisitor.h" 46 #include "llvm/IR/InstrTypes.h" 47 #include "llvm/IR/Instruction.h" 48 #include "llvm/IR/Instructions.h" 49 #include "llvm/IR/Module.h" 50 #include "llvm/IR/PassManager.h" 51 #include "llvm/IR/Type.h" 52 #include "llvm/IR/User.h" 53 #include "llvm/IR/Value.h" 54 #include "llvm/InitializePasses.h" 55 #include "llvm/Pass.h" 56 #include "llvm/Support/Casting.h" 57 #include "llvm/Support/Debug.h" 58 #include "llvm/Support/ErrorHandling.h" 59 #include "llvm/Support/raw_ostream.h" 60 #include "llvm/Transforms/Scalar.h" 61 #include "llvm/Transforms/Utils/Local.h" 62 #include "llvm/Transforms/Utils/PredicateInfo.h" 63 #include <cassert> 64 #include <utility> 65 #include <vector> 66 67 using namespace llvm; 68 69 #define DEBUG_TYPE "sccp" 70 71 STATISTIC(NumInstRemoved, "Number of instructions removed"); 72 STATISTIC(NumDeadBlocks , "Number of basic blocks unreachable"); 73 STATISTIC(NumInstReplaced, 74 "Number of instructions replaced with (simpler) instruction"); 75 76 STATISTIC(IPNumInstRemoved, "Number of instructions removed by IPSCCP"); 77 STATISTIC(IPNumArgsElimed ,"Number of arguments constant propagated by IPSCCP"); 78 STATISTIC(IPNumGlobalConst, "Number of globals found to be constant by IPSCCP"); 79 STATISTIC( 80 IPNumInstReplaced, 81 "Number of instructions replaced with (simpler) instruction by IPSCCP"); 82 83 // The maximum number of range extensions allowed for operations requiring 84 // widening. 85 static const unsigned MaxNumRangeExtensions = 10; 86 87 /// Returns MergeOptions with MaxWidenSteps set to MaxNumRangeExtensions. 88 static ValueLatticeElement::MergeOptions getMaxWidenStepsOpts() { 89 return ValueLatticeElement::MergeOptions().setMaxWidenSteps( 90 MaxNumRangeExtensions); 91 } 92 namespace { 93 94 // Helper to check if \p LV is either a constant or a constant 95 // range with a single element. This should cover exactly the same cases as the 96 // old ValueLatticeElement::isConstant() and is intended to be used in the 97 // transition to ValueLatticeElement. 98 bool isConstant(const ValueLatticeElement &LV) { 99 return LV.isConstant() || 100 (LV.isConstantRange() && LV.getConstantRange().isSingleElement()); 101 } 102 103 // Helper to check if \p LV is either overdefined or a constant range with more 104 // than a single element. This should cover exactly the same cases as the old 105 // ValueLatticeElement::isOverdefined() and is intended to be used in the 106 // transition to ValueLatticeElement. 107 bool isOverdefined(const ValueLatticeElement &LV) { 108 return !LV.isUnknownOrUndef() && !isConstant(LV); 109 } 110 111 //===----------------------------------------------------------------------===// 112 // 113 /// SCCPSolver - This class is a general purpose solver for Sparse Conditional 114 /// Constant Propagation. 115 /// 116 class SCCPSolver : public InstVisitor<SCCPSolver> { 117 const DataLayout &DL; 118 std::function<const TargetLibraryInfo &(Function &)> GetTLI; 119 SmallPtrSet<BasicBlock *, 8> BBExecutable; // The BBs that are executable. 120 DenseMap<Value *, ValueLatticeElement> 121 ValueState; // The state each value is in. 122 123 /// StructValueState - This maintains ValueState for values that have 124 /// StructType, for example for formal arguments, calls, insertelement, etc. 125 DenseMap<std::pair<Value *, unsigned>, ValueLatticeElement> StructValueState; 126 127 /// GlobalValue - If we are tracking any values for the contents of a global 128 /// variable, we keep a mapping from the constant accessor to the element of 129 /// the global, to the currently known value. If the value becomes 130 /// overdefined, it's entry is simply removed from this map. 131 DenseMap<GlobalVariable *, ValueLatticeElement> TrackedGlobals; 132 133 /// TrackedRetVals - If we are tracking arguments into and the return 134 /// value out of a function, it will have an entry in this map, indicating 135 /// what the known return value for the function is. 136 MapVector<Function *, ValueLatticeElement> TrackedRetVals; 137 138 /// TrackedMultipleRetVals - Same as TrackedRetVals, but used for functions 139 /// that return multiple values. 140 MapVector<std::pair<Function *, unsigned>, ValueLatticeElement> 141 TrackedMultipleRetVals; 142 143 /// MRVFunctionsTracked - Each function in TrackedMultipleRetVals is 144 /// represented here for efficient lookup. 145 SmallPtrSet<Function *, 16> MRVFunctionsTracked; 146 147 /// A list of functions whose return cannot be modified. 148 SmallPtrSet<Function *, 16> MustPreserveReturnsInFunctions; 149 150 /// TrackingIncomingArguments - This is the set of functions for whose 151 /// arguments we make optimistic assumptions about and try to prove as 152 /// constants. 153 SmallPtrSet<Function *, 16> TrackingIncomingArguments; 154 155 /// The reason for two worklists is that overdefined is the lowest state 156 /// on the lattice, and moving things to overdefined as fast as possible 157 /// makes SCCP converge much faster. 158 /// 159 /// By having a separate worklist, we accomplish this because everything 160 /// possibly overdefined will become overdefined at the soonest possible 161 /// point. 162 SmallVector<Value *, 64> OverdefinedInstWorkList; 163 SmallVector<Value *, 64> InstWorkList; 164 165 // The BasicBlock work list 166 SmallVector<BasicBlock *, 64> BBWorkList; 167 168 /// KnownFeasibleEdges - Entries in this set are edges which have already had 169 /// PHI nodes retriggered. 170 using Edge = std::pair<BasicBlock *, BasicBlock *>; 171 DenseSet<Edge> KnownFeasibleEdges; 172 173 DenseMap<Function *, AnalysisResultsForFn> AnalysisResults; 174 DenseMap<Value *, SmallPtrSet<User *, 2>> AdditionalUsers; 175 176 LLVMContext &Ctx; 177 178 public: 179 void addAnalysis(Function &F, AnalysisResultsForFn A) { 180 AnalysisResults.insert({&F, std::move(A)}); 181 } 182 183 const PredicateBase *getPredicateInfoFor(Instruction *I) { 184 auto A = AnalysisResults.find(I->getParent()->getParent()); 185 if (A == AnalysisResults.end()) 186 return nullptr; 187 return A->second.PredInfo->getPredicateInfoFor(I); 188 } 189 190 DomTreeUpdater getDTU(Function &F) { 191 auto A = AnalysisResults.find(&F); 192 assert(A != AnalysisResults.end() && "Need analysis results for function."); 193 return {A->second.DT, A->second.PDT, DomTreeUpdater::UpdateStrategy::Lazy}; 194 } 195 196 SCCPSolver(const DataLayout &DL, 197 std::function<const TargetLibraryInfo &(Function &)> GetTLI, 198 LLVMContext &Ctx) 199 : DL(DL), GetTLI(std::move(GetTLI)), Ctx(Ctx) {} 200 201 /// MarkBlockExecutable - This method can be used by clients to mark all of 202 /// the blocks that are known to be intrinsically live in the processed unit. 203 /// 204 /// This returns true if the block was not considered live before. 205 bool MarkBlockExecutable(BasicBlock *BB) { 206 if (!BBExecutable.insert(BB).second) 207 return false; 208 LLVM_DEBUG(dbgs() << "Marking Block Executable: " << BB->getName() << '\n'); 209 BBWorkList.push_back(BB); // Add the block to the work list! 210 return true; 211 } 212 213 /// TrackValueOfGlobalVariable - Clients can use this method to 214 /// inform the SCCPSolver that it should track loads and stores to the 215 /// specified global variable if it can. This is only legal to call if 216 /// performing Interprocedural SCCP. 217 void TrackValueOfGlobalVariable(GlobalVariable *GV) { 218 // We only track the contents of scalar globals. 219 if (GV->getValueType()->isSingleValueType()) { 220 ValueLatticeElement &IV = TrackedGlobals[GV]; 221 if (!isa<UndefValue>(GV->getInitializer())) 222 IV.markConstant(GV->getInitializer()); 223 } 224 } 225 226 /// AddTrackedFunction - If the SCCP solver is supposed to track calls into 227 /// and out of the specified function (which cannot have its address taken), 228 /// this method must be called. 229 void AddTrackedFunction(Function *F) { 230 // Add an entry, F -> undef. 231 if (auto *STy = dyn_cast<StructType>(F->getReturnType())) { 232 MRVFunctionsTracked.insert(F); 233 for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) 234 TrackedMultipleRetVals.insert( 235 std::make_pair(std::make_pair(F, i), ValueLatticeElement())); 236 } else if (!F->getReturnType()->isVoidTy()) 237 TrackedRetVals.insert(std::make_pair(F, ValueLatticeElement())); 238 } 239 240 /// Add function to the list of functions whose return cannot be modified. 241 void addToMustPreserveReturnsInFunctions(Function *F) { 242 MustPreserveReturnsInFunctions.insert(F); 243 } 244 245 /// Returns true if the return of the given function cannot be modified. 246 bool mustPreserveReturn(Function *F) { 247 return MustPreserveReturnsInFunctions.count(F); 248 } 249 250 void AddArgumentTrackedFunction(Function *F) { 251 TrackingIncomingArguments.insert(F); 252 } 253 254 /// Returns true if the given function is in the solver's set of 255 /// argument-tracked functions. 256 bool isArgumentTrackedFunction(Function *F) { 257 return TrackingIncomingArguments.count(F); 258 } 259 260 /// Solve - Solve for constants and executable blocks. 261 void Solve(); 262 263 /// ResolvedUndefsIn - While solving the dataflow for a function, we assume 264 /// that branches on undef values cannot reach any of their successors. 265 /// However, this is not a safe assumption. After we solve dataflow, this 266 /// method should be use to handle this. If this returns true, the solver 267 /// should be rerun. 268 bool ResolvedUndefsIn(Function &F); 269 270 bool isBlockExecutable(BasicBlock *BB) const { 271 return BBExecutable.count(BB); 272 } 273 274 // isEdgeFeasible - Return true if the control flow edge from the 'From' basic 275 // block to the 'To' basic block is currently feasible. 276 bool isEdgeFeasible(BasicBlock *From, BasicBlock *To) const; 277 278 std::vector<ValueLatticeElement> getStructLatticeValueFor(Value *V) const { 279 std::vector<ValueLatticeElement> StructValues; 280 auto *STy = dyn_cast<StructType>(V->getType()); 281 assert(STy && "getStructLatticeValueFor() can be called only on structs"); 282 for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) { 283 auto I = StructValueState.find(std::make_pair(V, i)); 284 assert(I != StructValueState.end() && "Value not in valuemap!"); 285 StructValues.push_back(I->second); 286 } 287 return StructValues; 288 } 289 290 void removeLatticeValueFor(Value *V) { ValueState.erase(V); } 291 292 const ValueLatticeElement &getLatticeValueFor(Value *V) const { 293 assert(!V->getType()->isStructTy() && 294 "Should use getStructLatticeValueFor"); 295 DenseMap<Value *, ValueLatticeElement>::const_iterator I = 296 ValueState.find(V); 297 assert(I != ValueState.end() && 298 "V not found in ValueState nor Paramstate map!"); 299 return I->second; 300 } 301 302 /// getTrackedRetVals - Get the inferred return value map. 303 const MapVector<Function *, ValueLatticeElement> &getTrackedRetVals() { 304 return TrackedRetVals; 305 } 306 307 /// getTrackedGlobals - Get and return the set of inferred initializers for 308 /// global variables. 309 const DenseMap<GlobalVariable *, ValueLatticeElement> &getTrackedGlobals() { 310 return TrackedGlobals; 311 } 312 313 /// getMRVFunctionsTracked - Get the set of functions which return multiple 314 /// values tracked by the pass. 315 const SmallPtrSet<Function *, 16> getMRVFunctionsTracked() { 316 return MRVFunctionsTracked; 317 } 318 319 /// markOverdefined - Mark the specified value overdefined. This 320 /// works with both scalars and structs. 321 void markOverdefined(Value *V) { 322 if (auto *STy = dyn_cast<StructType>(V->getType())) 323 for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) 324 markOverdefined(getStructValueState(V, i), V); 325 else 326 markOverdefined(ValueState[V], V); 327 } 328 329 // isStructLatticeConstant - Return true if all the lattice values 330 // corresponding to elements of the structure are constants, 331 // false otherwise. 332 bool isStructLatticeConstant(Function *F, StructType *STy) { 333 for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) { 334 const auto &It = TrackedMultipleRetVals.find(std::make_pair(F, i)); 335 assert(It != TrackedMultipleRetVals.end()); 336 ValueLatticeElement LV = It->second; 337 if (!isConstant(LV)) 338 return false; 339 } 340 return true; 341 } 342 343 /// Helper to return a Constant if \p LV is either a constant or a constant 344 /// range with a single element. 345 Constant *getConstant(const ValueLatticeElement &LV) const { 346 if (LV.isConstant()) 347 return LV.getConstant(); 348 349 if (LV.isConstantRange()) { 350 auto &CR = LV.getConstantRange(); 351 if (CR.getSingleElement()) 352 return ConstantInt::get(Ctx, *CR.getSingleElement()); 353 } 354 return nullptr; 355 } 356 357 private: 358 ConstantInt *getConstantInt(const ValueLatticeElement &IV) const { 359 return dyn_cast_or_null<ConstantInt>(getConstant(IV)); 360 } 361 362 // pushToWorkList - Helper for markConstant/markOverdefined 363 void pushToWorkList(ValueLatticeElement &IV, Value *V) { 364 if (IV.isOverdefined()) 365 return OverdefinedInstWorkList.push_back(V); 366 InstWorkList.push_back(V); 367 } 368 369 // Helper to push \p V to the worklist, after updating it to \p IV. Also 370 // prints a debug message with the updated value. 371 void pushToWorkListMsg(ValueLatticeElement &IV, Value *V) { 372 LLVM_DEBUG(dbgs() << "updated " << IV << ": " << *V << '\n'); 373 pushToWorkList(IV, V); 374 } 375 376 // markConstant - Make a value be marked as "constant". If the value 377 // is not already a constant, add it to the instruction work list so that 378 // the users of the instruction are updated later. 379 bool markConstant(ValueLatticeElement &IV, Value *V, Constant *C, 380 bool MayIncludeUndef = false) { 381 if (!IV.markConstant(C, MayIncludeUndef)) 382 return false; 383 LLVM_DEBUG(dbgs() << "markConstant: " << *C << ": " << *V << '\n'); 384 pushToWorkList(IV, V); 385 return true; 386 } 387 388 bool markConstant(Value *V, Constant *C) { 389 assert(!V->getType()->isStructTy() && "structs should use mergeInValue"); 390 return markConstant(ValueState[V], V, C); 391 } 392 393 // markOverdefined - Make a value be marked as "overdefined". If the 394 // value is not already overdefined, add it to the overdefined instruction 395 // work list so that the users of the instruction are updated later. 396 bool markOverdefined(ValueLatticeElement &IV, Value *V) { 397 if (!IV.markOverdefined()) return false; 398 399 LLVM_DEBUG(dbgs() << "markOverdefined: "; 400 if (auto *F = dyn_cast<Function>(V)) dbgs() 401 << "Function '" << F->getName() << "'\n"; 402 else dbgs() << *V << '\n'); 403 // Only instructions go on the work list 404 pushToWorkList(IV, V); 405 return true; 406 } 407 408 /// Merge \p MergeWithV into \p IV and push \p V to the worklist, if \p IV 409 /// changes. 410 bool mergeInValue(ValueLatticeElement &IV, Value *V, 411 ValueLatticeElement MergeWithV, 412 ValueLatticeElement::MergeOptions Opts = { 413 /*MayIncludeUndef=*/false, /*CheckWiden=*/false}) { 414 if (IV.mergeIn(MergeWithV, Opts)) { 415 pushToWorkList(IV, V); 416 LLVM_DEBUG(dbgs() << "Merged " << MergeWithV << " into " << *V << " : " 417 << IV << "\n"); 418 return true; 419 } 420 return false; 421 } 422 423 bool mergeInValue(Value *V, ValueLatticeElement MergeWithV, 424 ValueLatticeElement::MergeOptions Opts = { 425 /*MayIncludeUndef=*/false, /*CheckWiden=*/false}) { 426 assert(!V->getType()->isStructTy() && 427 "non-structs should use markConstant"); 428 return mergeInValue(ValueState[V], V, MergeWithV, Opts); 429 } 430 431 /// getValueState - Return the ValueLatticeElement object that corresponds to 432 /// the value. This function handles the case when the value hasn't been seen 433 /// yet by properly seeding constants etc. 434 ValueLatticeElement &getValueState(Value *V) { 435 assert(!V->getType()->isStructTy() && "Should use getStructValueState"); 436 437 auto I = ValueState.insert(std::make_pair(V, ValueLatticeElement())); 438 ValueLatticeElement &LV = I.first->second; 439 440 if (!I.second) 441 return LV; // Common case, already in the map. 442 443 if (auto *C = dyn_cast<Constant>(V)) 444 LV.markConstant(C); // Constants are constant 445 446 // All others are unknown by default. 447 return LV; 448 } 449 450 /// getStructValueState - Return the ValueLatticeElement object that 451 /// corresponds to the value/field pair. This function handles the case when 452 /// the value hasn't been seen yet by properly seeding constants etc. 453 ValueLatticeElement &getStructValueState(Value *V, unsigned i) { 454 assert(V->getType()->isStructTy() && "Should use getValueState"); 455 assert(i < cast<StructType>(V->getType())->getNumElements() && 456 "Invalid element #"); 457 458 auto I = StructValueState.insert( 459 std::make_pair(std::make_pair(V, i), ValueLatticeElement())); 460 ValueLatticeElement &LV = I.first->second; 461 462 if (!I.second) 463 return LV; // Common case, already in the map. 464 465 if (auto *C = dyn_cast<Constant>(V)) { 466 Constant *Elt = C->getAggregateElement(i); 467 468 if (!Elt) 469 LV.markOverdefined(); // Unknown sort of constant. 470 else if (isa<UndefValue>(Elt)) 471 ; // Undef values remain unknown. 472 else 473 LV.markConstant(Elt); // Constants are constant. 474 } 475 476 // All others are underdefined by default. 477 return LV; 478 } 479 480 /// markEdgeExecutable - Mark a basic block as executable, adding it to the BB 481 /// work list if it is not already executable. 482 bool markEdgeExecutable(BasicBlock *Source, BasicBlock *Dest) { 483 if (!KnownFeasibleEdges.insert(Edge(Source, Dest)).second) 484 return false; // This edge is already known to be executable! 485 486 if (!MarkBlockExecutable(Dest)) { 487 // If the destination is already executable, we just made an *edge* 488 // feasible that wasn't before. Revisit the PHI nodes in the block 489 // because they have potentially new operands. 490 LLVM_DEBUG(dbgs() << "Marking Edge Executable: " << Source->getName() 491 << " -> " << Dest->getName() << '\n'); 492 493 for (PHINode &PN : Dest->phis()) 494 visitPHINode(PN); 495 } 496 return true; 497 } 498 499 // getFeasibleSuccessors - Return a vector of booleans to indicate which 500 // successors are reachable from a given terminator instruction. 501 void getFeasibleSuccessors(Instruction &TI, SmallVectorImpl<bool> &Succs); 502 503 // OperandChangedState - This method is invoked on all of the users of an 504 // instruction that was just changed state somehow. Based on this 505 // information, we need to update the specified user of this instruction. 506 void OperandChangedState(Instruction *I) { 507 if (BBExecutable.count(I->getParent())) // Inst is executable? 508 visit(*I); 509 } 510 511 // Add U as additional user of V. 512 void addAdditionalUser(Value *V, User *U) { 513 auto Iter = AdditionalUsers.insert({V, {}}); 514 Iter.first->second.insert(U); 515 } 516 517 // Mark I's users as changed, including AdditionalUsers. 518 void markUsersAsChanged(Value *I) { 519 // Functions include their arguments in the use-list. Changed function 520 // values mean that the result of the function changed. We only need to 521 // update the call sites with the new function result and do not have to 522 // propagate the call arguments. 523 if (isa<Function>(I)) { 524 for (User *U : I->users()) { 525 if (auto *CB = dyn_cast<CallBase>(U)) 526 handleCallResult(*CB); 527 } 528 } else { 529 for (User *U : I->users()) 530 if (auto *UI = dyn_cast<Instruction>(U)) 531 OperandChangedState(UI); 532 } 533 534 auto Iter = AdditionalUsers.find(I); 535 if (Iter != AdditionalUsers.end()) { 536 // Copy additional users before notifying them of changes, because new 537 // users may be added, potentially invalidating the iterator. 538 SmallVector<Instruction *, 2> ToNotify; 539 for (User *U : Iter->second) 540 if (auto *UI = dyn_cast<Instruction>(U)) 541 ToNotify.push_back(UI); 542 for (Instruction *UI : ToNotify) 543 OperandChangedState(UI); 544 } 545 } 546 void handleCallOverdefined(CallBase &CB); 547 void handleCallResult(CallBase &CB); 548 void handleCallArguments(CallBase &CB); 549 550 private: 551 friend class InstVisitor<SCCPSolver>; 552 553 // visit implementations - Something changed in this instruction. Either an 554 // operand made a transition, or the instruction is newly executable. Change 555 // the value type of I to reflect these changes if appropriate. 556 void visitPHINode(PHINode &I); 557 558 // Terminators 559 560 void visitReturnInst(ReturnInst &I); 561 void visitTerminator(Instruction &TI); 562 563 void visitCastInst(CastInst &I); 564 void visitSelectInst(SelectInst &I); 565 void visitUnaryOperator(Instruction &I); 566 void visitBinaryOperator(Instruction &I); 567 void visitCmpInst(CmpInst &I); 568 void visitExtractValueInst(ExtractValueInst &EVI); 569 void visitInsertValueInst(InsertValueInst &IVI); 570 571 void visitCatchSwitchInst(CatchSwitchInst &CPI) { 572 markOverdefined(&CPI); 573 visitTerminator(CPI); 574 } 575 576 // Instructions that cannot be folded away. 577 578 void visitStoreInst (StoreInst &I); 579 void visitLoadInst (LoadInst &I); 580 void visitGetElementPtrInst(GetElementPtrInst &I); 581 582 void visitCallInst (CallInst &I) { 583 visitCallBase(I); 584 } 585 586 void visitInvokeInst (InvokeInst &II) { 587 visitCallBase(II); 588 visitTerminator(II); 589 } 590 591 void visitCallBrInst (CallBrInst &CBI) { 592 visitCallBase(CBI); 593 visitTerminator(CBI); 594 } 595 596 void visitCallBase (CallBase &CB); 597 void visitResumeInst (ResumeInst &I) { /*returns void*/ } 598 void visitUnreachableInst(UnreachableInst &I) { /*returns void*/ } 599 void visitFenceInst (FenceInst &I) { /*returns void*/ } 600 601 void visitInstruction(Instruction &I) { 602 // All the instructions we don't do any special handling for just 603 // go to overdefined. 604 LLVM_DEBUG(dbgs() << "SCCP: Don't know how to handle: " << I << '\n'); 605 markOverdefined(&I); 606 } 607 }; 608 609 } // end anonymous namespace 610 611 // getFeasibleSuccessors - Return a vector of booleans to indicate which 612 // successors are reachable from a given terminator instruction. 613 void SCCPSolver::getFeasibleSuccessors(Instruction &TI, 614 SmallVectorImpl<bool> &Succs) { 615 Succs.resize(TI.getNumSuccessors()); 616 if (auto *BI = dyn_cast<BranchInst>(&TI)) { 617 if (BI->isUnconditional()) { 618 Succs[0] = true; 619 return; 620 } 621 622 ValueLatticeElement BCValue = getValueState(BI->getCondition()); 623 ConstantInt *CI = getConstantInt(BCValue); 624 if (!CI) { 625 // Overdefined condition variables, and branches on unfoldable constant 626 // conditions, mean the branch could go either way. 627 if (!BCValue.isUnknownOrUndef()) 628 Succs[0] = Succs[1] = true; 629 return; 630 } 631 632 // Constant condition variables mean the branch can only go a single way. 633 Succs[CI->isZero()] = true; 634 return; 635 } 636 637 // Unwinding instructions successors are always executable. 638 if (TI.isExceptionalTerminator()) { 639 Succs.assign(TI.getNumSuccessors(), true); 640 return; 641 } 642 643 if (auto *SI = dyn_cast<SwitchInst>(&TI)) { 644 if (!SI->getNumCases()) { 645 Succs[0] = true; 646 return; 647 } 648 const ValueLatticeElement &SCValue = getValueState(SI->getCondition()); 649 if (ConstantInt *CI = getConstantInt(SCValue)) { 650 Succs[SI->findCaseValue(CI)->getSuccessorIndex()] = true; 651 return; 652 } 653 654 // TODO: Switch on undef is UB. Stop passing false once the rest of LLVM 655 // is ready. 656 if (SCValue.isConstantRange(/*UndefAllowed=*/false)) { 657 const ConstantRange &Range = SCValue.getConstantRange(); 658 for (const auto &Case : SI->cases()) { 659 const APInt &CaseValue = Case.getCaseValue()->getValue(); 660 if (Range.contains(CaseValue)) 661 Succs[Case.getSuccessorIndex()] = true; 662 } 663 664 // TODO: Determine whether default case is reachable. 665 Succs[SI->case_default()->getSuccessorIndex()] = true; 666 return; 667 } 668 669 // Overdefined or unknown condition? All destinations are executable! 670 if (!SCValue.isUnknownOrUndef()) 671 Succs.assign(TI.getNumSuccessors(), true); 672 return; 673 } 674 675 // In case of indirect branch and its address is a blockaddress, we mark 676 // the target as executable. 677 if (auto *IBR = dyn_cast<IndirectBrInst>(&TI)) { 678 // Casts are folded by visitCastInst. 679 ValueLatticeElement IBRValue = getValueState(IBR->getAddress()); 680 BlockAddress *Addr = dyn_cast_or_null<BlockAddress>(getConstant(IBRValue)); 681 if (!Addr) { // Overdefined or unknown condition? 682 // All destinations are executable! 683 if (!IBRValue.isUnknownOrUndef()) 684 Succs.assign(TI.getNumSuccessors(), true); 685 return; 686 } 687 688 BasicBlock* T = Addr->getBasicBlock(); 689 assert(Addr->getFunction() == T->getParent() && 690 "Block address of a different function ?"); 691 for (unsigned i = 0; i < IBR->getNumSuccessors(); ++i) { 692 // This is the target. 693 if (IBR->getDestination(i) == T) { 694 Succs[i] = true; 695 return; 696 } 697 } 698 699 // If we didn't find our destination in the IBR successor list, then we 700 // have undefined behavior. Its ok to assume no successor is executable. 701 return; 702 } 703 704 // In case of callbr, we pessimistically assume that all successors are 705 // feasible. 706 if (isa<CallBrInst>(&TI)) { 707 Succs.assign(TI.getNumSuccessors(), true); 708 return; 709 } 710 711 LLVM_DEBUG(dbgs() << "Unknown terminator instruction: " << TI << '\n'); 712 llvm_unreachable("SCCP: Don't know how to handle this terminator!"); 713 } 714 715 // isEdgeFeasible - Return true if the control flow edge from the 'From' basic 716 // block to the 'To' basic block is currently feasible. 717 bool SCCPSolver::isEdgeFeasible(BasicBlock *From, BasicBlock *To) const { 718 // Check if we've called markEdgeExecutable on the edge yet. (We could 719 // be more aggressive and try to consider edges which haven't been marked 720 // yet, but there isn't any need.) 721 return KnownFeasibleEdges.count(Edge(From, To)); 722 } 723 724 // visit Implementations - Something changed in this instruction, either an 725 // operand made a transition, or the instruction is newly executable. Change 726 // the value type of I to reflect these changes if appropriate. This method 727 // makes sure to do the following actions: 728 // 729 // 1. If a phi node merges two constants in, and has conflicting value coming 730 // from different branches, or if the PHI node merges in an overdefined 731 // value, then the PHI node becomes overdefined. 732 // 2. If a phi node merges only constants in, and they all agree on value, the 733 // PHI node becomes a constant value equal to that. 734 // 3. If V <- x (op) y && isConstant(x) && isConstant(y) V = Constant 735 // 4. If V <- x (op) y && (isOverdefined(x) || isOverdefined(y)) V = Overdefined 736 // 5. If V <- MEM or V <- CALL or V <- (unknown) then V = Overdefined 737 // 6. If a conditional branch has a value that is constant, make the selected 738 // destination executable 739 // 7. If a conditional branch has a value that is overdefined, make all 740 // successors executable. 741 void SCCPSolver::visitPHINode(PHINode &PN) { 742 // If this PN returns a struct, just mark the result overdefined. 743 // TODO: We could do a lot better than this if code actually uses this. 744 if (PN.getType()->isStructTy()) 745 return (void)markOverdefined(&PN); 746 747 if (getValueState(&PN).isOverdefined()) 748 return; // Quick exit 749 750 // Super-extra-high-degree PHI nodes are unlikely to ever be marked constant, 751 // and slow us down a lot. Just mark them overdefined. 752 if (PN.getNumIncomingValues() > 64) 753 return (void)markOverdefined(&PN); 754 755 unsigned NumActiveIncoming = 0; 756 757 // Look at all of the executable operands of the PHI node. If any of them 758 // are overdefined, the PHI becomes overdefined as well. If they are all 759 // constant, and they agree with each other, the PHI becomes the identical 760 // constant. If they are constant and don't agree, the PHI is a constant 761 // range. If there are no executable operands, the PHI remains unknown. 762 ValueLatticeElement PhiState = getValueState(&PN); 763 for (unsigned i = 0, e = PN.getNumIncomingValues(); i != e; ++i) { 764 if (!isEdgeFeasible(PN.getIncomingBlock(i), PN.getParent())) 765 continue; 766 767 ValueLatticeElement IV = getValueState(PN.getIncomingValue(i)); 768 PhiState.mergeIn(IV); 769 NumActiveIncoming++; 770 if (PhiState.isOverdefined()) 771 break; 772 } 773 774 // We allow up to 1 range extension per active incoming value and one 775 // additional extension. Note that we manually adjust the number of range 776 // extensions to match the number of active incoming values. This helps to 777 // limit multiple extensions caused by the same incoming value, if other 778 // incoming values are equal. 779 mergeInValue(&PN, PhiState, 780 ValueLatticeElement::MergeOptions().setMaxWidenSteps( 781 NumActiveIncoming + 1)); 782 ValueLatticeElement &PhiStateRef = getValueState(&PN); 783 PhiStateRef.setNumRangeExtensions( 784 std::max(NumActiveIncoming, PhiStateRef.getNumRangeExtensions())); 785 } 786 787 void SCCPSolver::visitReturnInst(ReturnInst &I) { 788 if (I.getNumOperands() == 0) return; // ret void 789 790 Function *F = I.getParent()->getParent(); 791 Value *ResultOp = I.getOperand(0); 792 793 // If we are tracking the return value of this function, merge it in. 794 if (!TrackedRetVals.empty() && !ResultOp->getType()->isStructTy()) { 795 auto TFRVI = TrackedRetVals.find(F); 796 if (TFRVI != TrackedRetVals.end()) { 797 mergeInValue(TFRVI->second, F, getValueState(ResultOp)); 798 return; 799 } 800 } 801 802 // Handle functions that return multiple values. 803 if (!TrackedMultipleRetVals.empty()) { 804 if (auto *STy = dyn_cast<StructType>(ResultOp->getType())) 805 if (MRVFunctionsTracked.count(F)) 806 for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) 807 mergeInValue(TrackedMultipleRetVals[std::make_pair(F, i)], F, 808 getStructValueState(ResultOp, i)); 809 } 810 } 811 812 void SCCPSolver::visitTerminator(Instruction &TI) { 813 SmallVector<bool, 16> SuccFeasible; 814 getFeasibleSuccessors(TI, SuccFeasible); 815 816 BasicBlock *BB = TI.getParent(); 817 818 // Mark all feasible successors executable. 819 for (unsigned i = 0, e = SuccFeasible.size(); i != e; ++i) 820 if (SuccFeasible[i]) 821 markEdgeExecutable(BB, TI.getSuccessor(i)); 822 } 823 824 void SCCPSolver::visitCastInst(CastInst &I) { 825 // ResolvedUndefsIn might mark I as overdefined. Bail out, even if we would 826 // discover a concrete value later. 827 if (ValueState[&I].isOverdefined()) 828 return; 829 830 ValueLatticeElement OpSt = getValueState(I.getOperand(0)); 831 if (Constant *OpC = getConstant(OpSt)) { 832 // Fold the constant as we build. 833 Constant *C = ConstantFoldCastOperand(I.getOpcode(), OpC, I.getType(), DL); 834 if (isa<UndefValue>(C)) 835 return; 836 // Propagate constant value 837 markConstant(&I, C); 838 } else if (OpSt.isConstantRange() && I.getDestTy()->isIntegerTy()) { 839 auto &LV = getValueState(&I); 840 ConstantRange OpRange = OpSt.getConstantRange(); 841 Type *DestTy = I.getDestTy(); 842 // Vectors where all elements have the same known constant range are treated 843 // as a single constant range in the lattice. When bitcasting such vectors, 844 // there is a mis-match between the width of the lattice value (single 845 // constant range) and the original operands (vector). Go to overdefined in 846 // that case. 847 if (I.getOpcode() == Instruction::BitCast && 848 I.getOperand(0)->getType()->isVectorTy() && 849 OpRange.getBitWidth() < DL.getTypeSizeInBits(DestTy)) 850 return (void)markOverdefined(&I); 851 852 ConstantRange Res = 853 OpRange.castOp(I.getOpcode(), DL.getTypeSizeInBits(DestTy)); 854 mergeInValue(LV, &I, ValueLatticeElement::getRange(Res)); 855 } else if (!OpSt.isUnknownOrUndef()) 856 markOverdefined(&I); 857 } 858 859 void SCCPSolver::visitExtractValueInst(ExtractValueInst &EVI) { 860 // If this returns a struct, mark all elements over defined, we don't track 861 // structs in structs. 862 if (EVI.getType()->isStructTy()) 863 return (void)markOverdefined(&EVI); 864 865 // ResolvedUndefsIn might mark I as overdefined. Bail out, even if we would 866 // discover a concrete value later. 867 if (ValueState[&EVI].isOverdefined()) 868 return (void)markOverdefined(&EVI); 869 870 // If this is extracting from more than one level of struct, we don't know. 871 if (EVI.getNumIndices() != 1) 872 return (void)markOverdefined(&EVI); 873 874 Value *AggVal = EVI.getAggregateOperand(); 875 if (AggVal->getType()->isStructTy()) { 876 unsigned i = *EVI.idx_begin(); 877 ValueLatticeElement EltVal = getStructValueState(AggVal, i); 878 mergeInValue(getValueState(&EVI), &EVI, EltVal); 879 } else { 880 // Otherwise, must be extracting from an array. 881 return (void)markOverdefined(&EVI); 882 } 883 } 884 885 void SCCPSolver::visitInsertValueInst(InsertValueInst &IVI) { 886 auto *STy = dyn_cast<StructType>(IVI.getType()); 887 if (!STy) 888 return (void)markOverdefined(&IVI); 889 890 // ResolvedUndefsIn might mark I as overdefined. Bail out, even if we would 891 // discover a concrete value later. 892 if (isOverdefined(ValueState[&IVI])) 893 return (void)markOverdefined(&IVI); 894 895 // If this has more than one index, we can't handle it, drive all results to 896 // undef. 897 if (IVI.getNumIndices() != 1) 898 return (void)markOverdefined(&IVI); 899 900 Value *Aggr = IVI.getAggregateOperand(); 901 unsigned Idx = *IVI.idx_begin(); 902 903 // Compute the result based on what we're inserting. 904 for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) { 905 // This passes through all values that aren't the inserted element. 906 if (i != Idx) { 907 ValueLatticeElement EltVal = getStructValueState(Aggr, i); 908 mergeInValue(getStructValueState(&IVI, i), &IVI, EltVal); 909 continue; 910 } 911 912 Value *Val = IVI.getInsertedValueOperand(); 913 if (Val->getType()->isStructTy()) 914 // We don't track structs in structs. 915 markOverdefined(getStructValueState(&IVI, i), &IVI); 916 else { 917 ValueLatticeElement InVal = getValueState(Val); 918 mergeInValue(getStructValueState(&IVI, i), &IVI, InVal); 919 } 920 } 921 } 922 923 void SCCPSolver::visitSelectInst(SelectInst &I) { 924 // If this select returns a struct, just mark the result overdefined. 925 // TODO: We could do a lot better than this if code actually uses this. 926 if (I.getType()->isStructTy()) 927 return (void)markOverdefined(&I); 928 929 // ResolvedUndefsIn might mark I as overdefined. Bail out, even if we would 930 // discover a concrete value later. 931 if (ValueState[&I].isOverdefined()) 932 return (void)markOverdefined(&I); 933 934 ValueLatticeElement CondValue = getValueState(I.getCondition()); 935 if (CondValue.isUnknownOrUndef()) 936 return; 937 938 if (ConstantInt *CondCB = getConstantInt(CondValue)) { 939 Value *OpVal = CondCB->isZero() ? I.getFalseValue() : I.getTrueValue(); 940 mergeInValue(&I, getValueState(OpVal)); 941 return; 942 } 943 944 // Otherwise, the condition is overdefined or a constant we can't evaluate. 945 // See if we can produce something better than overdefined based on the T/F 946 // value. 947 ValueLatticeElement TVal = getValueState(I.getTrueValue()); 948 ValueLatticeElement FVal = getValueState(I.getFalseValue()); 949 950 bool Changed = ValueState[&I].mergeIn(TVal); 951 Changed |= ValueState[&I].mergeIn(FVal); 952 if (Changed) 953 pushToWorkListMsg(ValueState[&I], &I); 954 } 955 956 // Handle Unary Operators. 957 void SCCPSolver::visitUnaryOperator(Instruction &I) { 958 ValueLatticeElement V0State = getValueState(I.getOperand(0)); 959 960 ValueLatticeElement &IV = ValueState[&I]; 961 // ResolvedUndefsIn might mark I as overdefined. Bail out, even if we would 962 // discover a concrete value later. 963 if (isOverdefined(IV)) 964 return (void)markOverdefined(&I); 965 966 if (isConstant(V0State)) { 967 Constant *C = ConstantExpr::get(I.getOpcode(), getConstant(V0State)); 968 969 // op Y -> undef. 970 if (isa<UndefValue>(C)) 971 return; 972 return (void)markConstant(IV, &I, C); 973 } 974 975 // If something is undef, wait for it to resolve. 976 if (!isOverdefined(V0State)) 977 return; 978 979 markOverdefined(&I); 980 } 981 982 // Handle Binary Operators. 983 void SCCPSolver::visitBinaryOperator(Instruction &I) { 984 ValueLatticeElement V1State = getValueState(I.getOperand(0)); 985 ValueLatticeElement V2State = getValueState(I.getOperand(1)); 986 987 ValueLatticeElement &IV = ValueState[&I]; 988 if (IV.isOverdefined()) 989 return; 990 991 // If something is undef, wait for it to resolve. 992 if (V1State.isUnknownOrUndef() || V2State.isUnknownOrUndef()) 993 return; 994 995 if (V1State.isOverdefined() && V2State.isOverdefined()) 996 return (void)markOverdefined(&I); 997 998 // If either of the operands is a constant, try to fold it to a constant. 999 // TODO: Use information from notconstant better. 1000 if ((V1State.isConstant() || V2State.isConstant())) { 1001 Value *V1 = isConstant(V1State) ? getConstant(V1State) : I.getOperand(0); 1002 Value *V2 = isConstant(V2State) ? getConstant(V2State) : I.getOperand(1); 1003 Value *R = SimplifyBinOp(I.getOpcode(), V1, V2, SimplifyQuery(DL)); 1004 auto *C = dyn_cast_or_null<Constant>(R); 1005 if (C) { 1006 // X op Y -> undef. 1007 if (isa<UndefValue>(C)) 1008 return; 1009 // Conservatively assume that the result may be based on operands that may 1010 // be undef. Note that we use mergeInValue to combine the constant with 1011 // the existing lattice value for I, as different constants might be found 1012 // after one of the operands go to overdefined, e.g. due to one operand 1013 // being a special floating value. 1014 ValueLatticeElement NewV; 1015 NewV.markConstant(C, /*MayIncludeUndef=*/true); 1016 return (void)mergeInValue(&I, NewV); 1017 } 1018 } 1019 1020 // Only use ranges for binary operators on integers. 1021 if (!I.getType()->isIntegerTy()) 1022 return markOverdefined(&I); 1023 1024 // Try to simplify to a constant range. 1025 ConstantRange A = ConstantRange::getFull(I.getType()->getScalarSizeInBits()); 1026 ConstantRange B = ConstantRange::getFull(I.getType()->getScalarSizeInBits()); 1027 if (V1State.isConstantRange()) 1028 A = V1State.getConstantRange(); 1029 if (V2State.isConstantRange()) 1030 B = V2State.getConstantRange(); 1031 1032 ConstantRange R = A.binaryOp(cast<BinaryOperator>(&I)->getOpcode(), B); 1033 mergeInValue(&I, ValueLatticeElement::getRange(R)); 1034 1035 // TODO: Currently we do not exploit special values that produce something 1036 // better than overdefined with an overdefined operand for vector or floating 1037 // point types, like and <4 x i32> overdefined, zeroinitializer. 1038 } 1039 1040 // Handle ICmpInst instruction. 1041 void SCCPSolver::visitCmpInst(CmpInst &I) { 1042 // Do not cache this lookup, getValueState calls later in the function might 1043 // invalidate the reference. 1044 if (isOverdefined(ValueState[&I])) 1045 return (void)markOverdefined(&I); 1046 1047 Value *Op1 = I.getOperand(0); 1048 Value *Op2 = I.getOperand(1); 1049 1050 // For parameters, use ParamState which includes constant range info if 1051 // available. 1052 auto V1State = getValueState(Op1); 1053 auto V2State = getValueState(Op2); 1054 1055 Constant *C = V1State.getCompare(I.getPredicate(), I.getType(), V2State); 1056 if (C) { 1057 if (isa<UndefValue>(C)) 1058 return; 1059 ValueLatticeElement CV; 1060 CV.markConstant(C); 1061 mergeInValue(&I, CV); 1062 return; 1063 } 1064 1065 // If operands are still unknown, wait for it to resolve. 1066 if ((V1State.isUnknownOrUndef() || V2State.isUnknownOrUndef()) && 1067 !isConstant(ValueState[&I])) 1068 return; 1069 1070 markOverdefined(&I); 1071 } 1072 1073 // Handle getelementptr instructions. If all operands are constants then we 1074 // can turn this into a getelementptr ConstantExpr. 1075 void SCCPSolver::visitGetElementPtrInst(GetElementPtrInst &I) { 1076 if (isOverdefined(ValueState[&I])) 1077 return (void)markOverdefined(&I); 1078 1079 SmallVector<Constant*, 8> Operands; 1080 Operands.reserve(I.getNumOperands()); 1081 1082 for (unsigned i = 0, e = I.getNumOperands(); i != e; ++i) { 1083 ValueLatticeElement State = getValueState(I.getOperand(i)); 1084 if (State.isUnknownOrUndef()) 1085 return; // Operands are not resolved yet. 1086 1087 if (isOverdefined(State)) 1088 return (void)markOverdefined(&I); 1089 1090 if (Constant *C = getConstant(State)) { 1091 Operands.push_back(C); 1092 continue; 1093 } 1094 1095 return (void)markOverdefined(&I); 1096 } 1097 1098 Constant *Ptr = Operands[0]; 1099 auto Indices = makeArrayRef(Operands.begin() + 1, Operands.end()); 1100 Constant *C = 1101 ConstantExpr::getGetElementPtr(I.getSourceElementType(), Ptr, Indices); 1102 if (isa<UndefValue>(C)) 1103 return; 1104 markConstant(&I, C); 1105 } 1106 1107 void SCCPSolver::visitStoreInst(StoreInst &SI) { 1108 // If this store is of a struct, ignore it. 1109 if (SI.getOperand(0)->getType()->isStructTy()) 1110 return; 1111 1112 if (TrackedGlobals.empty() || !isa<GlobalVariable>(SI.getOperand(1))) 1113 return; 1114 1115 GlobalVariable *GV = cast<GlobalVariable>(SI.getOperand(1)); 1116 auto I = TrackedGlobals.find(GV); 1117 if (I == TrackedGlobals.end()) 1118 return; 1119 1120 // Get the value we are storing into the global, then merge it. 1121 mergeInValue(I->second, GV, getValueState(SI.getOperand(0)), 1122 ValueLatticeElement::MergeOptions().setCheckWiden(false)); 1123 if (I->second.isOverdefined()) 1124 TrackedGlobals.erase(I); // No need to keep tracking this! 1125 } 1126 1127 static ValueLatticeElement getValueFromMetadata(const Instruction *I) { 1128 if (MDNode *Ranges = I->getMetadata(LLVMContext::MD_range)) 1129 if (I->getType()->isIntegerTy()) 1130 return ValueLatticeElement::getRange( 1131 getConstantRangeFromMetadata(*Ranges)); 1132 if (I->hasMetadata(LLVMContext::MD_nonnull)) 1133 return ValueLatticeElement::getNot( 1134 ConstantPointerNull::get(cast<PointerType>(I->getType()))); 1135 return ValueLatticeElement::getOverdefined(); 1136 } 1137 1138 // Handle load instructions. If the operand is a constant pointer to a constant 1139 // global, we can replace the load with the loaded constant value! 1140 void SCCPSolver::visitLoadInst(LoadInst &I) { 1141 // If this load is of a struct or the load is volatile, just mark the result 1142 // as overdefined. 1143 if (I.getType()->isStructTy() || I.isVolatile()) 1144 return (void)markOverdefined(&I); 1145 1146 // ResolvedUndefsIn might mark I as overdefined. Bail out, even if we would 1147 // discover a concrete value later. 1148 if (ValueState[&I].isOverdefined()) 1149 return (void)markOverdefined(&I); 1150 1151 ValueLatticeElement PtrVal = getValueState(I.getOperand(0)); 1152 if (PtrVal.isUnknownOrUndef()) 1153 return; // The pointer is not resolved yet! 1154 1155 ValueLatticeElement &IV = ValueState[&I]; 1156 1157 if (isConstant(PtrVal)) { 1158 Constant *Ptr = getConstant(PtrVal); 1159 1160 // load null is undefined. 1161 if (isa<ConstantPointerNull>(Ptr)) { 1162 if (NullPointerIsDefined(I.getFunction(), I.getPointerAddressSpace())) 1163 return (void)markOverdefined(IV, &I); 1164 else 1165 return; 1166 } 1167 1168 // Transform load (constant global) into the value loaded. 1169 if (auto *GV = dyn_cast<GlobalVariable>(Ptr)) { 1170 if (!TrackedGlobals.empty()) { 1171 // If we are tracking this global, merge in the known value for it. 1172 auto It = TrackedGlobals.find(GV); 1173 if (It != TrackedGlobals.end()) { 1174 mergeInValue(IV, &I, It->second, getMaxWidenStepsOpts()); 1175 return; 1176 } 1177 } 1178 } 1179 1180 // Transform load from a constant into a constant if possible. 1181 if (Constant *C = ConstantFoldLoadFromConstPtr(Ptr, I.getType(), DL)) { 1182 if (isa<UndefValue>(C)) 1183 return; 1184 return (void)markConstant(IV, &I, C); 1185 } 1186 } 1187 1188 // Fall back to metadata. 1189 mergeInValue(&I, getValueFromMetadata(&I)); 1190 } 1191 1192 void SCCPSolver::visitCallBase(CallBase &CB) { 1193 handleCallResult(CB); 1194 handleCallArguments(CB); 1195 } 1196 1197 void SCCPSolver::handleCallOverdefined(CallBase &CB) { 1198 Function *F = CB.getCalledFunction(); 1199 1200 // Void return and not tracking callee, just bail. 1201 if (CB.getType()->isVoidTy()) 1202 return; 1203 1204 // Always mark struct return as overdefined. 1205 if (CB.getType()->isStructTy()) 1206 return (void)markOverdefined(&CB); 1207 1208 // Otherwise, if we have a single return value case, and if the function is 1209 // a declaration, maybe we can constant fold it. 1210 if (F && F->isDeclaration() && canConstantFoldCallTo(&CB, F)) { 1211 SmallVector<Constant *, 8> Operands; 1212 for (auto AI = CB.arg_begin(), E = CB.arg_end(); AI != E; ++AI) { 1213 if (AI->get()->getType()->isStructTy()) 1214 return markOverdefined(&CB); // Can't handle struct args. 1215 ValueLatticeElement State = getValueState(*AI); 1216 1217 if (State.isUnknownOrUndef()) 1218 return; // Operands are not resolved yet. 1219 if (isOverdefined(State)) 1220 return (void)markOverdefined(&CB); 1221 assert(isConstant(State) && "Unknown state!"); 1222 Operands.push_back(getConstant(State)); 1223 } 1224 1225 if (isOverdefined(getValueState(&CB))) 1226 return (void)markOverdefined(&CB); 1227 1228 // If we can constant fold this, mark the result of the call as a 1229 // constant. 1230 if (Constant *C = ConstantFoldCall(&CB, F, Operands, &GetTLI(*F))) { 1231 // call -> undef. 1232 if (isa<UndefValue>(C)) 1233 return; 1234 return (void)markConstant(&CB, C); 1235 } 1236 } 1237 1238 // Fall back to metadata. 1239 mergeInValue(&CB, getValueFromMetadata(&CB)); 1240 } 1241 1242 void SCCPSolver::handleCallArguments(CallBase &CB) { 1243 Function *F = CB.getCalledFunction(); 1244 // If this is a local function that doesn't have its address taken, mark its 1245 // entry block executable and merge in the actual arguments to the call into 1246 // the formal arguments of the function. 1247 if (!TrackingIncomingArguments.empty() && 1248 TrackingIncomingArguments.count(F)) { 1249 MarkBlockExecutable(&F->front()); 1250 1251 // Propagate information from this call site into the callee. 1252 auto CAI = CB.arg_begin(); 1253 for (Function::arg_iterator AI = F->arg_begin(), E = F->arg_end(); AI != E; 1254 ++AI, ++CAI) { 1255 // If this argument is byval, and if the function is not readonly, there 1256 // will be an implicit copy formed of the input aggregate. 1257 if (AI->hasByValAttr() && !F->onlyReadsMemory()) { 1258 markOverdefined(&*AI); 1259 continue; 1260 } 1261 1262 if (auto *STy = dyn_cast<StructType>(AI->getType())) { 1263 for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) { 1264 ValueLatticeElement CallArg = getStructValueState(*CAI, i); 1265 mergeInValue(getStructValueState(&*AI, i), &*AI, CallArg, 1266 getMaxWidenStepsOpts()); 1267 } 1268 } else 1269 mergeInValue(&*AI, getValueState(*CAI), getMaxWidenStepsOpts()); 1270 } 1271 } 1272 } 1273 1274 void SCCPSolver::handleCallResult(CallBase &CB) { 1275 Function *F = CB.getCalledFunction(); 1276 1277 if (auto *II = dyn_cast<IntrinsicInst>(&CB)) { 1278 if (II->getIntrinsicID() == Intrinsic::ssa_copy) { 1279 if (ValueState[&CB].isOverdefined()) 1280 return; 1281 1282 Value *CopyOf = CB.getOperand(0); 1283 ValueLatticeElement CopyOfVal = getValueState(CopyOf); 1284 auto *PI = getPredicateInfoFor(&CB); 1285 assert(PI && "Missing predicate info for ssa.copy"); 1286 1287 const Optional<PredicateConstraint> &Constraint = PI->getConstraint(); 1288 if (!Constraint) { 1289 mergeInValue(ValueState[&CB], &CB, CopyOfVal); 1290 return; 1291 } 1292 1293 CmpInst::Predicate Pred = Constraint->Predicate; 1294 Value *OtherOp = Constraint->OtherOp; 1295 1296 // Wait until OtherOp is resolved. 1297 if (getValueState(OtherOp).isUnknown()) { 1298 addAdditionalUser(OtherOp, &CB); 1299 return; 1300 } 1301 1302 // TODO: Actually filp MayIncludeUndef for the created range to false, 1303 // once most places in the optimizer respect the branches on 1304 // undef/poison are UB rule. The reason why the new range cannot be 1305 // undef is as follows below: 1306 // The new range is based on a branch condition. That guarantees that 1307 // neither of the compare operands can be undef in the branch targets, 1308 // unless we have conditions that are always true/false (e.g. icmp ule 1309 // i32, %a, i32_max). For the latter overdefined/empty range will be 1310 // inferred, but the branch will get folded accordingly anyways. 1311 bool MayIncludeUndef = !isa<PredicateAssume>(PI); 1312 1313 ValueLatticeElement CondVal = getValueState(OtherOp); 1314 ValueLatticeElement &IV = ValueState[&CB]; 1315 if (CondVal.isConstantRange() || CopyOfVal.isConstantRange()) { 1316 auto ImposedCR = 1317 ConstantRange::getFull(DL.getTypeSizeInBits(CopyOf->getType())); 1318 1319 // Get the range imposed by the condition. 1320 if (CondVal.isConstantRange()) 1321 ImposedCR = ConstantRange::makeAllowedICmpRegion( 1322 Pred, CondVal.getConstantRange()); 1323 1324 // Combine range info for the original value with the new range from the 1325 // condition. 1326 auto CopyOfCR = CopyOfVal.isConstantRange() 1327 ? CopyOfVal.getConstantRange() 1328 : ConstantRange::getFull( 1329 DL.getTypeSizeInBits(CopyOf->getType())); 1330 auto NewCR = ImposedCR.intersectWith(CopyOfCR); 1331 // If the existing information is != x, do not use the information from 1332 // a chained predicate, as the != x information is more likely to be 1333 // helpful in practice. 1334 if (!CopyOfCR.contains(NewCR) && CopyOfCR.getSingleMissingElement()) 1335 NewCR = CopyOfCR; 1336 1337 addAdditionalUser(OtherOp, &CB); 1338 mergeInValue( 1339 IV, &CB, 1340 ValueLatticeElement::getRange(NewCR, MayIncludeUndef)); 1341 return; 1342 } else if (Pred == CmpInst::ICMP_EQ && CondVal.isConstant()) { 1343 // For non-integer values or integer constant expressions, only 1344 // propagate equal constants. 1345 addAdditionalUser(OtherOp, &CB); 1346 mergeInValue(IV, &CB, CondVal); 1347 return; 1348 } else if (Pred == CmpInst::ICMP_NE && CondVal.isConstant() && 1349 !MayIncludeUndef) { 1350 // Propagate inequalities. 1351 addAdditionalUser(OtherOp, &CB); 1352 mergeInValue(IV, &CB, 1353 ValueLatticeElement::getNot(CondVal.getConstant())); 1354 return; 1355 } 1356 1357 return (void)mergeInValue(IV, &CB, CopyOfVal); 1358 } 1359 1360 if (ConstantRange::isIntrinsicSupported(II->getIntrinsicID())) { 1361 // Compute result range for intrinsics supported by ConstantRange. 1362 // Do this even if we don't know a range for all operands, as we may 1363 // still know something about the result range, e.g. of abs(x). 1364 SmallVector<ConstantRange, 2> OpRanges; 1365 for (Value *Op : II->args()) { 1366 const ValueLatticeElement &State = getValueState(Op); 1367 if (State.isConstantRange()) 1368 OpRanges.push_back(State.getConstantRange()); 1369 else 1370 OpRanges.push_back( 1371 ConstantRange::getFull(Op->getType()->getScalarSizeInBits())); 1372 } 1373 1374 ConstantRange Result = 1375 ConstantRange::intrinsic(II->getIntrinsicID(), OpRanges); 1376 return (void)mergeInValue(II, ValueLatticeElement::getRange(Result)); 1377 } 1378 } 1379 1380 // The common case is that we aren't tracking the callee, either because we 1381 // are not doing interprocedural analysis or the callee is indirect, or is 1382 // external. Handle these cases first. 1383 if (!F || F->isDeclaration()) 1384 return handleCallOverdefined(CB); 1385 1386 // If this is a single/zero retval case, see if we're tracking the function. 1387 if (auto *STy = dyn_cast<StructType>(F->getReturnType())) { 1388 if (!MRVFunctionsTracked.count(F)) 1389 return handleCallOverdefined(CB); // Not tracking this callee. 1390 1391 // If we are tracking this callee, propagate the result of the function 1392 // into this call site. 1393 for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) 1394 mergeInValue(getStructValueState(&CB, i), &CB, 1395 TrackedMultipleRetVals[std::make_pair(F, i)], 1396 getMaxWidenStepsOpts()); 1397 } else { 1398 auto TFRVI = TrackedRetVals.find(F); 1399 if (TFRVI == TrackedRetVals.end()) 1400 return handleCallOverdefined(CB); // Not tracking this callee. 1401 1402 // If so, propagate the return value of the callee into this call result. 1403 mergeInValue(&CB, TFRVI->second, getMaxWidenStepsOpts()); 1404 } 1405 } 1406 1407 void SCCPSolver::Solve() { 1408 // Process the work lists until they are empty! 1409 while (!BBWorkList.empty() || !InstWorkList.empty() || 1410 !OverdefinedInstWorkList.empty()) { 1411 // Process the overdefined instruction's work list first, which drives other 1412 // things to overdefined more quickly. 1413 while (!OverdefinedInstWorkList.empty()) { 1414 Value *I = OverdefinedInstWorkList.pop_back_val(); 1415 1416 LLVM_DEBUG(dbgs() << "\nPopped off OI-WL: " << *I << '\n'); 1417 1418 // "I" got into the work list because it either made the transition from 1419 // bottom to constant, or to overdefined. 1420 // 1421 // Anything on this worklist that is overdefined need not be visited 1422 // since all of its users will have already been marked as overdefined 1423 // Update all of the users of this instruction's value. 1424 // 1425 markUsersAsChanged(I); 1426 } 1427 1428 // Process the instruction work list. 1429 while (!InstWorkList.empty()) { 1430 Value *I = InstWorkList.pop_back_val(); 1431 1432 LLVM_DEBUG(dbgs() << "\nPopped off I-WL: " << *I << '\n'); 1433 1434 // "I" got into the work list because it made the transition from undef to 1435 // constant. 1436 // 1437 // Anything on this worklist that is overdefined need not be visited 1438 // since all of its users will have already been marked as overdefined. 1439 // Update all of the users of this instruction's value. 1440 // 1441 if (I->getType()->isStructTy() || !getValueState(I).isOverdefined()) 1442 markUsersAsChanged(I); 1443 } 1444 1445 // Process the basic block work list. 1446 while (!BBWorkList.empty()) { 1447 BasicBlock *BB = BBWorkList.pop_back_val(); 1448 1449 LLVM_DEBUG(dbgs() << "\nPopped off BBWL: " << *BB << '\n'); 1450 1451 // Notify all instructions in this basic block that they are newly 1452 // executable. 1453 visit(BB); 1454 } 1455 } 1456 } 1457 1458 /// ResolvedUndefsIn - While solving the dataflow for a function, we assume 1459 /// that branches on undef values cannot reach any of their successors. 1460 /// However, this is not a safe assumption. After we solve dataflow, this 1461 /// method should be use to handle this. If this returns true, the solver 1462 /// should be rerun. 1463 /// 1464 /// This method handles this by finding an unresolved branch and marking it one 1465 /// of the edges from the block as being feasible, even though the condition 1466 /// doesn't say it would otherwise be. This allows SCCP to find the rest of the 1467 /// CFG and only slightly pessimizes the analysis results (by marking one, 1468 /// potentially infeasible, edge feasible). This cannot usefully modify the 1469 /// constraints on the condition of the branch, as that would impact other users 1470 /// of the value. 1471 /// 1472 /// This scan also checks for values that use undefs. It conservatively marks 1473 /// them as overdefined. 1474 bool SCCPSolver::ResolvedUndefsIn(Function &F) { 1475 bool MadeChange = false; 1476 for (BasicBlock &BB : F) { 1477 if (!BBExecutable.count(&BB)) 1478 continue; 1479 1480 for (Instruction &I : BB) { 1481 // Look for instructions which produce undef values. 1482 if (I.getType()->isVoidTy()) continue; 1483 1484 if (auto *STy = dyn_cast<StructType>(I.getType())) { 1485 // Only a few things that can be structs matter for undef. 1486 1487 // Tracked calls must never be marked overdefined in ResolvedUndefsIn. 1488 if (auto *CB = dyn_cast<CallBase>(&I)) 1489 if (Function *F = CB->getCalledFunction()) 1490 if (MRVFunctionsTracked.count(F)) 1491 continue; 1492 1493 // extractvalue and insertvalue don't need to be marked; they are 1494 // tracked as precisely as their operands. 1495 if (isa<ExtractValueInst>(I) || isa<InsertValueInst>(I)) 1496 continue; 1497 // Send the results of everything else to overdefined. We could be 1498 // more precise than this but it isn't worth bothering. 1499 for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) { 1500 ValueLatticeElement &LV = getStructValueState(&I, i); 1501 if (LV.isUnknownOrUndef()) { 1502 markOverdefined(LV, &I); 1503 MadeChange = true; 1504 } 1505 } 1506 continue; 1507 } 1508 1509 ValueLatticeElement &LV = getValueState(&I); 1510 if (!LV.isUnknownOrUndef()) 1511 continue; 1512 1513 // There are two reasons a call can have an undef result 1514 // 1. It could be tracked. 1515 // 2. It could be constant-foldable. 1516 // Because of the way we solve return values, tracked calls must 1517 // never be marked overdefined in ResolvedUndefsIn. 1518 if (auto *CB = dyn_cast<CallBase>(&I)) 1519 if (Function *F = CB->getCalledFunction()) 1520 if (TrackedRetVals.count(F)) 1521 continue; 1522 1523 if (isa<LoadInst>(I)) { 1524 // A load here means one of two things: a load of undef from a global, 1525 // a load from an unknown pointer. Either way, having it return undef 1526 // is okay. 1527 continue; 1528 } 1529 1530 markOverdefined(&I); 1531 MadeChange = true; 1532 } 1533 1534 // Check to see if we have a branch or switch on an undefined value. If so 1535 // we force the branch to go one way or the other to make the successor 1536 // values live. It doesn't really matter which way we force it. 1537 Instruction *TI = BB.getTerminator(); 1538 if (auto *BI = dyn_cast<BranchInst>(TI)) { 1539 if (!BI->isConditional()) continue; 1540 if (!getValueState(BI->getCondition()).isUnknownOrUndef()) 1541 continue; 1542 1543 // If the input to SCCP is actually branch on undef, fix the undef to 1544 // false. 1545 if (isa<UndefValue>(BI->getCondition())) { 1546 BI->setCondition(ConstantInt::getFalse(BI->getContext())); 1547 markEdgeExecutable(&BB, TI->getSuccessor(1)); 1548 MadeChange = true; 1549 continue; 1550 } 1551 1552 // Otherwise, it is a branch on a symbolic value which is currently 1553 // considered to be undef. Make sure some edge is executable, so a 1554 // branch on "undef" always flows somewhere. 1555 // FIXME: Distinguish between dead code and an LLVM "undef" value. 1556 BasicBlock *DefaultSuccessor = TI->getSuccessor(1); 1557 if (markEdgeExecutable(&BB, DefaultSuccessor)) 1558 MadeChange = true; 1559 1560 continue; 1561 } 1562 1563 if (auto *IBR = dyn_cast<IndirectBrInst>(TI)) { 1564 // Indirect branch with no successor ?. Its ok to assume it branches 1565 // to no target. 1566 if (IBR->getNumSuccessors() < 1) 1567 continue; 1568 1569 if (!getValueState(IBR->getAddress()).isUnknownOrUndef()) 1570 continue; 1571 1572 // If the input to SCCP is actually branch on undef, fix the undef to 1573 // the first successor of the indirect branch. 1574 if (isa<UndefValue>(IBR->getAddress())) { 1575 IBR->setAddress(BlockAddress::get(IBR->getSuccessor(0))); 1576 markEdgeExecutable(&BB, IBR->getSuccessor(0)); 1577 MadeChange = true; 1578 continue; 1579 } 1580 1581 // Otherwise, it is a branch on a symbolic value which is currently 1582 // considered to be undef. Make sure some edge is executable, so a 1583 // branch on "undef" always flows somewhere. 1584 // FIXME: IndirectBr on "undef" doesn't actually need to go anywhere: 1585 // we can assume the branch has undefined behavior instead. 1586 BasicBlock *DefaultSuccessor = IBR->getSuccessor(0); 1587 if (markEdgeExecutable(&BB, DefaultSuccessor)) 1588 MadeChange = true; 1589 1590 continue; 1591 } 1592 1593 if (auto *SI = dyn_cast<SwitchInst>(TI)) { 1594 if (!SI->getNumCases() || 1595 !getValueState(SI->getCondition()).isUnknownOrUndef()) 1596 continue; 1597 1598 // If the input to SCCP is actually switch on undef, fix the undef to 1599 // the first constant. 1600 if (isa<UndefValue>(SI->getCondition())) { 1601 SI->setCondition(SI->case_begin()->getCaseValue()); 1602 markEdgeExecutable(&BB, SI->case_begin()->getCaseSuccessor()); 1603 MadeChange = true; 1604 continue; 1605 } 1606 1607 // Otherwise, it is a branch on a symbolic value which is currently 1608 // considered to be undef. Make sure some edge is executable, so a 1609 // branch on "undef" always flows somewhere. 1610 // FIXME: Distinguish between dead code and an LLVM "undef" value. 1611 BasicBlock *DefaultSuccessor = SI->case_begin()->getCaseSuccessor(); 1612 if (markEdgeExecutable(&BB, DefaultSuccessor)) 1613 MadeChange = true; 1614 1615 continue; 1616 } 1617 } 1618 1619 return MadeChange; 1620 } 1621 1622 static bool tryToReplaceWithConstant(SCCPSolver &Solver, Value *V) { 1623 Constant *Const = nullptr; 1624 if (V->getType()->isStructTy()) { 1625 std::vector<ValueLatticeElement> IVs = Solver.getStructLatticeValueFor(V); 1626 if (any_of(IVs, 1627 [](const ValueLatticeElement &LV) { return isOverdefined(LV); })) 1628 return false; 1629 std::vector<Constant *> ConstVals; 1630 auto *ST = cast<StructType>(V->getType()); 1631 for (unsigned i = 0, e = ST->getNumElements(); i != e; ++i) { 1632 ValueLatticeElement V = IVs[i]; 1633 ConstVals.push_back(isConstant(V) 1634 ? Solver.getConstant(V) 1635 : UndefValue::get(ST->getElementType(i))); 1636 } 1637 Const = ConstantStruct::get(ST, ConstVals); 1638 } else { 1639 const ValueLatticeElement &IV = Solver.getLatticeValueFor(V); 1640 if (isOverdefined(IV)) 1641 return false; 1642 1643 Const = 1644 isConstant(IV) ? Solver.getConstant(IV) : UndefValue::get(V->getType()); 1645 } 1646 assert(Const && "Constant is nullptr here!"); 1647 1648 // Replacing `musttail` instructions with constant breaks `musttail` invariant 1649 // unless the call itself can be removed. 1650 // Calls with "clang.arc.attachedcall" implicitly use the return value and 1651 // those uses cannot be updated with a constant. 1652 CallBase *CB = dyn_cast<CallBase>(V); 1653 if (CB && ((CB->isMustTailCall() && !CB->isSafeToRemove()) || 1654 CB->getOperandBundle(LLVMContext::OB_clang_arc_attachedcall))) { 1655 Function *F = CB->getCalledFunction(); 1656 1657 // Don't zap returns of the callee 1658 if (F) 1659 Solver.addToMustPreserveReturnsInFunctions(F); 1660 1661 LLVM_DEBUG(dbgs() << " Can\'t treat the result of call " << *CB 1662 << " as a constant\n"); 1663 return false; 1664 } 1665 1666 LLVM_DEBUG(dbgs() << " Constant: " << *Const << " = " << *V << '\n'); 1667 1668 // Replaces all of the uses of a variable with uses of the constant. 1669 V->replaceAllUsesWith(Const); 1670 return true; 1671 } 1672 1673 static bool simplifyInstsInBlock(SCCPSolver &Solver, BasicBlock &BB, 1674 SmallPtrSetImpl<Value *> &InsertedValues, 1675 Statistic &InstRemovedStat, 1676 Statistic &InstReplacedStat) { 1677 bool MadeChanges = false; 1678 for (Instruction &Inst : make_early_inc_range(BB)) { 1679 if (Inst.getType()->isVoidTy()) 1680 continue; 1681 if (tryToReplaceWithConstant(Solver, &Inst)) { 1682 if (Inst.isSafeToRemove()) 1683 Inst.eraseFromParent(); 1684 // Hey, we just changed something! 1685 MadeChanges = true; 1686 ++InstRemovedStat; 1687 } else if (isa<SExtInst>(&Inst)) { 1688 Value *ExtOp = Inst.getOperand(0); 1689 if (isa<Constant>(ExtOp) || InsertedValues.count(ExtOp)) 1690 continue; 1691 const ValueLatticeElement &IV = Solver.getLatticeValueFor(ExtOp); 1692 if (!IV.isConstantRange(/*UndefAllowed=*/false)) 1693 continue; 1694 if (IV.getConstantRange().isAllNonNegative()) { 1695 auto *ZExt = new ZExtInst(ExtOp, Inst.getType(), "", &Inst); 1696 InsertedValues.insert(ZExt); 1697 Inst.replaceAllUsesWith(ZExt); 1698 Solver.removeLatticeValueFor(&Inst); 1699 Inst.eraseFromParent(); 1700 InstReplacedStat++; 1701 MadeChanges = true; 1702 } 1703 } 1704 } 1705 return MadeChanges; 1706 } 1707 1708 // runSCCP() - Run the Sparse Conditional Constant Propagation algorithm, 1709 // and return true if the function was modified. 1710 static bool runSCCP(Function &F, const DataLayout &DL, 1711 const TargetLibraryInfo *TLI) { 1712 LLVM_DEBUG(dbgs() << "SCCP on function '" << F.getName() << "'\n"); 1713 SCCPSolver Solver( 1714 DL, [TLI](Function &F) -> const TargetLibraryInfo & { return *TLI; }, 1715 F.getContext()); 1716 1717 // Mark the first block of the function as being executable. 1718 Solver.MarkBlockExecutable(&F.front()); 1719 1720 // Mark all arguments to the function as being overdefined. 1721 for (Argument &AI : F.args()) 1722 Solver.markOverdefined(&AI); 1723 1724 // Solve for constants. 1725 bool ResolvedUndefs = true; 1726 while (ResolvedUndefs) { 1727 Solver.Solve(); 1728 LLVM_DEBUG(dbgs() << "RESOLVING UNDEFs\n"); 1729 ResolvedUndefs = Solver.ResolvedUndefsIn(F); 1730 } 1731 1732 bool MadeChanges = false; 1733 1734 // If we decided that there are basic blocks that are dead in this function, 1735 // delete their contents now. Note that we cannot actually delete the blocks, 1736 // as we cannot modify the CFG of the function. 1737 1738 SmallPtrSet<Value *, 32> InsertedValues; 1739 for (BasicBlock &BB : F) { 1740 if (!Solver.isBlockExecutable(&BB)) { 1741 LLVM_DEBUG(dbgs() << " BasicBlock Dead:" << BB); 1742 1743 ++NumDeadBlocks; 1744 NumInstRemoved += removeAllNonTerminatorAndEHPadInstructions(&BB).first; 1745 1746 MadeChanges = true; 1747 continue; 1748 } 1749 1750 MadeChanges |= simplifyInstsInBlock(Solver, BB, InsertedValues, 1751 NumInstRemoved, NumInstReplaced); 1752 } 1753 1754 return MadeChanges; 1755 } 1756 1757 PreservedAnalyses SCCPPass::run(Function &F, FunctionAnalysisManager &AM) { 1758 const DataLayout &DL = F.getParent()->getDataLayout(); 1759 auto &TLI = AM.getResult<TargetLibraryAnalysis>(F); 1760 if (!runSCCP(F, DL, &TLI)) 1761 return PreservedAnalyses::all(); 1762 1763 auto PA = PreservedAnalyses(); 1764 PA.preserve<GlobalsAA>(); 1765 PA.preserveSet<CFGAnalyses>(); 1766 return PA; 1767 } 1768 1769 namespace { 1770 1771 //===--------------------------------------------------------------------===// 1772 // 1773 /// SCCP Class - This class uses the SCCPSolver to implement a per-function 1774 /// Sparse Conditional Constant Propagator. 1775 /// 1776 class SCCPLegacyPass : public FunctionPass { 1777 public: 1778 // Pass identification, replacement for typeid 1779 static char ID; 1780 1781 SCCPLegacyPass() : FunctionPass(ID) { 1782 initializeSCCPLegacyPassPass(*PassRegistry::getPassRegistry()); 1783 } 1784 1785 void getAnalysisUsage(AnalysisUsage &AU) const override { 1786 AU.addRequired<TargetLibraryInfoWrapperPass>(); 1787 AU.addPreserved<GlobalsAAWrapperPass>(); 1788 AU.setPreservesCFG(); 1789 } 1790 1791 // runOnFunction - Run the Sparse Conditional Constant Propagation 1792 // algorithm, and return true if the function was modified. 1793 bool runOnFunction(Function &F) override { 1794 if (skipFunction(F)) 1795 return false; 1796 const DataLayout &DL = F.getParent()->getDataLayout(); 1797 const TargetLibraryInfo *TLI = 1798 &getAnalysis<TargetLibraryInfoWrapperPass>().getTLI(F); 1799 return runSCCP(F, DL, TLI); 1800 } 1801 }; 1802 1803 } // end anonymous namespace 1804 1805 char SCCPLegacyPass::ID = 0; 1806 1807 INITIALIZE_PASS_BEGIN(SCCPLegacyPass, "sccp", 1808 "Sparse Conditional Constant Propagation", false, false) 1809 INITIALIZE_PASS_DEPENDENCY(TargetLibraryInfoWrapperPass) 1810 INITIALIZE_PASS_END(SCCPLegacyPass, "sccp", 1811 "Sparse Conditional Constant Propagation", false, false) 1812 1813 // createSCCPPass - This is the public interface to this file. 1814 FunctionPass *llvm::createSCCPPass() { return new SCCPLegacyPass(); } 1815 1816 static void findReturnsToZap(Function &F, 1817 SmallVector<ReturnInst *, 8> &ReturnsToZap, 1818 SCCPSolver &Solver) { 1819 // We can only do this if we know that nothing else can call the function. 1820 if (!Solver.isArgumentTrackedFunction(&F)) 1821 return; 1822 1823 if (Solver.mustPreserveReturn(&F)) { 1824 LLVM_DEBUG( 1825 dbgs() 1826 << "Can't zap returns of the function : " << F.getName() 1827 << " due to present musttail or \"clang.arc.attachedcall\" call of " 1828 "it\n"); 1829 return; 1830 } 1831 1832 assert( 1833 all_of(F.users(), 1834 [&Solver](User *U) { 1835 if (isa<Instruction>(U) && 1836 !Solver.isBlockExecutable(cast<Instruction>(U)->getParent())) 1837 return true; 1838 // Non-callsite uses are not impacted by zapping. Also, constant 1839 // uses (like blockaddresses) could stuck around, without being 1840 // used in the underlying IR, meaning we do not have lattice 1841 // values for them. 1842 if (!isa<CallBase>(U)) 1843 return true; 1844 if (U->getType()->isStructTy()) { 1845 return all_of(Solver.getStructLatticeValueFor(U), 1846 [](const ValueLatticeElement &LV) { 1847 return !isOverdefined(LV); 1848 }); 1849 } 1850 return !isOverdefined(Solver.getLatticeValueFor(U)); 1851 }) && 1852 "We can only zap functions where all live users have a concrete value"); 1853 1854 for (BasicBlock &BB : F) { 1855 if (CallInst *CI = BB.getTerminatingMustTailCall()) { 1856 LLVM_DEBUG(dbgs() << "Can't zap return of the block due to present " 1857 << "musttail call : " << *CI << "\n"); 1858 (void)CI; 1859 return; 1860 } 1861 1862 if (auto *RI = dyn_cast<ReturnInst>(BB.getTerminator())) 1863 if (!isa<UndefValue>(RI->getOperand(0))) 1864 ReturnsToZap.push_back(RI); 1865 } 1866 } 1867 1868 static bool removeNonFeasibleEdges(const SCCPSolver &Solver, BasicBlock *BB, 1869 DomTreeUpdater &DTU) { 1870 SmallPtrSet<BasicBlock *, 8> FeasibleSuccessors; 1871 bool HasNonFeasibleEdges = false; 1872 for (BasicBlock *Succ : successors(BB)) { 1873 if (Solver.isEdgeFeasible(BB, Succ)) 1874 FeasibleSuccessors.insert(Succ); 1875 else 1876 HasNonFeasibleEdges = true; 1877 } 1878 1879 // All edges feasible, nothing to do. 1880 if (!HasNonFeasibleEdges) 1881 return false; 1882 1883 // SCCP can only determine non-feasible edges for br, switch and indirectbr. 1884 Instruction *TI = BB->getTerminator(); 1885 assert((isa<BranchInst>(TI) || isa<SwitchInst>(TI) || 1886 isa<IndirectBrInst>(TI)) && 1887 "Terminator must be a br, switch or indirectbr"); 1888 1889 if (FeasibleSuccessors.size() == 1) { 1890 // Replace with an unconditional branch to the only feasible successor. 1891 BasicBlock *OnlyFeasibleSuccessor = *FeasibleSuccessors.begin(); 1892 SmallVector<DominatorTree::UpdateType, 8> Updates; 1893 bool HaveSeenOnlyFeasibleSuccessor = false; 1894 for (BasicBlock *Succ : successors(BB)) { 1895 if (Succ == OnlyFeasibleSuccessor && !HaveSeenOnlyFeasibleSuccessor) { 1896 // Don't remove the edge to the only feasible successor the first time 1897 // we see it. We still do need to remove any multi-edges to it though. 1898 HaveSeenOnlyFeasibleSuccessor = true; 1899 continue; 1900 } 1901 1902 Succ->removePredecessor(BB); 1903 Updates.push_back({DominatorTree::Delete, BB, Succ}); 1904 } 1905 1906 BranchInst::Create(OnlyFeasibleSuccessor, BB); 1907 TI->eraseFromParent(); 1908 DTU.applyUpdatesPermissive(Updates); 1909 } else if (FeasibleSuccessors.size() > 1) { 1910 SwitchInstProfUpdateWrapper SI(*cast<SwitchInst>(TI)); 1911 SmallVector<DominatorTree::UpdateType, 8> Updates; 1912 for (auto CI = SI->case_begin(); CI != SI->case_end();) { 1913 if (FeasibleSuccessors.contains(CI->getCaseSuccessor())) { 1914 ++CI; 1915 continue; 1916 } 1917 1918 BasicBlock *Succ = CI->getCaseSuccessor(); 1919 Succ->removePredecessor(BB); 1920 Updates.push_back({DominatorTree::Delete, BB, Succ}); 1921 SI.removeCase(CI); 1922 // Don't increment CI, as we removed a case. 1923 } 1924 1925 DTU.applyUpdatesPermissive(Updates); 1926 } else { 1927 llvm_unreachable("Must have at least one feasible successor"); 1928 } 1929 return true; 1930 } 1931 1932 bool llvm::runIPSCCP( 1933 Module &M, const DataLayout &DL, 1934 std::function<const TargetLibraryInfo &(Function &)> GetTLI, 1935 function_ref<AnalysisResultsForFn(Function &)> getAnalysis) { 1936 SCCPSolver Solver(DL, GetTLI, M.getContext()); 1937 1938 // Loop over all functions, marking arguments to those with their addresses 1939 // taken or that are external as overdefined. 1940 for (Function &F : M) { 1941 if (F.isDeclaration()) 1942 continue; 1943 1944 Solver.addAnalysis(F, getAnalysis(F)); 1945 1946 // Determine if we can track the function's return values. If so, add the 1947 // function to the solver's set of return-tracked functions. 1948 if (canTrackReturnsInterprocedurally(&F)) 1949 Solver.AddTrackedFunction(&F); 1950 1951 // Determine if we can track the function's arguments. If so, add the 1952 // function to the solver's set of argument-tracked functions. 1953 if (canTrackArgumentsInterprocedurally(&F)) { 1954 Solver.AddArgumentTrackedFunction(&F); 1955 continue; 1956 } 1957 1958 // Assume the function is called. 1959 Solver.MarkBlockExecutable(&F.front()); 1960 1961 // Assume nothing about the incoming arguments. 1962 for (Argument &AI : F.args()) 1963 Solver.markOverdefined(&AI); 1964 } 1965 1966 // Determine if we can track any of the module's global variables. If so, add 1967 // the global variables we can track to the solver's set of tracked global 1968 // variables. 1969 for (GlobalVariable &G : M.globals()) { 1970 G.removeDeadConstantUsers(); 1971 if (canTrackGlobalVariableInterprocedurally(&G)) 1972 Solver.TrackValueOfGlobalVariable(&G); 1973 } 1974 1975 // Solve for constants. 1976 bool ResolvedUndefs = true; 1977 Solver.Solve(); 1978 while (ResolvedUndefs) { 1979 LLVM_DEBUG(dbgs() << "RESOLVING UNDEFS\n"); 1980 ResolvedUndefs = false; 1981 for (Function &F : M) { 1982 if (Solver.ResolvedUndefsIn(F)) 1983 ResolvedUndefs = true; 1984 } 1985 if (ResolvedUndefs) 1986 Solver.Solve(); 1987 } 1988 1989 bool MadeChanges = false; 1990 1991 // Iterate over all of the instructions in the module, replacing them with 1992 // constants if we have found them to be of constant values. 1993 1994 for (Function &F : M) { 1995 if (F.isDeclaration()) 1996 continue; 1997 1998 SmallVector<BasicBlock *, 512> BlocksToErase; 1999 2000 if (Solver.isBlockExecutable(&F.front())) { 2001 bool ReplacedPointerArg = false; 2002 for (Argument &Arg : F.args()) { 2003 if (!Arg.use_empty() && tryToReplaceWithConstant(Solver, &Arg)) { 2004 ReplacedPointerArg |= Arg.getType()->isPointerTy(); 2005 ++IPNumArgsElimed; 2006 } 2007 } 2008 2009 // If we replaced an argument, the argmemonly and 2010 // inaccessiblemem_or_argmemonly attributes do not hold any longer. Remove 2011 // them from both the function and callsites. 2012 if (ReplacedPointerArg) { 2013 AttrBuilder AttributesToRemove; 2014 AttributesToRemove.addAttribute(Attribute::ArgMemOnly); 2015 AttributesToRemove.addAttribute(Attribute::InaccessibleMemOrArgMemOnly); 2016 F.removeAttributes(AttributeList::FunctionIndex, AttributesToRemove); 2017 2018 for (User *U : F.users()) { 2019 auto *CB = dyn_cast<CallBase>(U); 2020 if (!CB || CB->getCalledFunction() != &F) 2021 continue; 2022 2023 CB->removeAttributes(AttributeList::FunctionIndex, 2024 AttributesToRemove); 2025 } 2026 } 2027 } 2028 2029 SmallPtrSet<Value *, 32> InsertedValues; 2030 for (BasicBlock &BB : F) { 2031 if (!Solver.isBlockExecutable(&BB)) { 2032 LLVM_DEBUG(dbgs() << " BasicBlock Dead:" << BB); 2033 ++NumDeadBlocks; 2034 2035 MadeChanges = true; 2036 2037 if (&BB != &F.front()) 2038 BlocksToErase.push_back(&BB); 2039 continue; 2040 } 2041 2042 MadeChanges |= simplifyInstsInBlock(Solver, BB, InsertedValues, 2043 IPNumInstRemoved, IPNumInstReplaced); 2044 } 2045 2046 DomTreeUpdater DTU = Solver.getDTU(F); 2047 // Change dead blocks to unreachable. We do it after replacing constants 2048 // in all executable blocks, because changeToUnreachable may remove PHI 2049 // nodes in executable blocks we found values for. The function's entry 2050 // block is not part of BlocksToErase, so we have to handle it separately. 2051 for (BasicBlock *BB : BlocksToErase) { 2052 NumInstRemoved += 2053 changeToUnreachable(BB->getFirstNonPHI(), /*UseLLVMTrap=*/false, 2054 /*PreserveLCSSA=*/false, &DTU); 2055 } 2056 if (!Solver.isBlockExecutable(&F.front())) 2057 NumInstRemoved += changeToUnreachable(F.front().getFirstNonPHI(), 2058 /*UseLLVMTrap=*/false, 2059 /*PreserveLCSSA=*/false, &DTU); 2060 2061 for (BasicBlock &BB : F) 2062 MadeChanges |= removeNonFeasibleEdges(Solver, &BB, DTU); 2063 2064 for (BasicBlock *DeadBB : BlocksToErase) 2065 DTU.deleteBB(DeadBB); 2066 2067 for (BasicBlock &BB : F) { 2068 for (BasicBlock::iterator BI = BB.begin(), E = BB.end(); BI != E;) { 2069 Instruction *Inst = &*BI++; 2070 if (Solver.getPredicateInfoFor(Inst)) { 2071 if (auto *II = dyn_cast<IntrinsicInst>(Inst)) { 2072 if (II->getIntrinsicID() == Intrinsic::ssa_copy) { 2073 Value *Op = II->getOperand(0); 2074 Inst->replaceAllUsesWith(Op); 2075 Inst->eraseFromParent(); 2076 } 2077 } 2078 } 2079 } 2080 } 2081 } 2082 2083 // If we inferred constant or undef return values for a function, we replaced 2084 // all call uses with the inferred value. This means we don't need to bother 2085 // actually returning anything from the function. Replace all return 2086 // instructions with return undef. 2087 // 2088 // Do this in two stages: first identify the functions we should process, then 2089 // actually zap their returns. This is important because we can only do this 2090 // if the address of the function isn't taken. In cases where a return is the 2091 // last use of a function, the order of processing functions would affect 2092 // whether other functions are optimizable. 2093 SmallVector<ReturnInst*, 8> ReturnsToZap; 2094 2095 for (const auto &I : Solver.getTrackedRetVals()) { 2096 Function *F = I.first; 2097 const ValueLatticeElement &ReturnValue = I.second; 2098 2099 // If there is a known constant range for the return value, add !range 2100 // metadata to the function's call sites. 2101 if (ReturnValue.isConstantRange() && 2102 !ReturnValue.getConstantRange().isSingleElement()) { 2103 // Do not add range metadata if the return value may include undef. 2104 if (ReturnValue.isConstantRangeIncludingUndef()) 2105 continue; 2106 2107 auto &CR = ReturnValue.getConstantRange(); 2108 for (User *User : F->users()) { 2109 auto *CB = dyn_cast<CallBase>(User); 2110 if (!CB || CB->getCalledFunction() != F) 2111 continue; 2112 2113 // Limit to cases where the return value is guaranteed to be neither 2114 // poison nor undef. Poison will be outside any range and currently 2115 // values outside of the specified range cause immediate undefined 2116 // behavior. 2117 if (!isGuaranteedNotToBeUndefOrPoison(CB, nullptr, CB)) 2118 continue; 2119 2120 // Do not touch existing metadata for now. 2121 // TODO: We should be able to take the intersection of the existing 2122 // metadata and the inferred range. 2123 if (CB->getMetadata(LLVMContext::MD_range)) 2124 continue; 2125 2126 LLVMContext &Context = CB->getParent()->getContext(); 2127 Metadata *RangeMD[] = { 2128 ConstantAsMetadata::get(ConstantInt::get(Context, CR.getLower())), 2129 ConstantAsMetadata::get(ConstantInt::get(Context, CR.getUpper()))}; 2130 CB->setMetadata(LLVMContext::MD_range, MDNode::get(Context, RangeMD)); 2131 } 2132 continue; 2133 } 2134 if (F->getReturnType()->isVoidTy()) 2135 continue; 2136 if (isConstant(ReturnValue) || ReturnValue.isUnknownOrUndef()) 2137 findReturnsToZap(*F, ReturnsToZap, Solver); 2138 } 2139 2140 for (auto F : Solver.getMRVFunctionsTracked()) { 2141 assert(F->getReturnType()->isStructTy() && 2142 "The return type should be a struct"); 2143 StructType *STy = cast<StructType>(F->getReturnType()); 2144 if (Solver.isStructLatticeConstant(F, STy)) 2145 findReturnsToZap(*F, ReturnsToZap, Solver); 2146 } 2147 2148 // Zap all returns which we've identified as zap to change. 2149 SmallSetVector<Function *, 8> FuncZappedReturn; 2150 for (unsigned i = 0, e = ReturnsToZap.size(); i != e; ++i) { 2151 Function *F = ReturnsToZap[i]->getParent()->getParent(); 2152 ReturnsToZap[i]->setOperand(0, UndefValue::get(F->getReturnType())); 2153 // Record all functions that are zapped. 2154 FuncZappedReturn.insert(F); 2155 } 2156 2157 // Remove the returned attribute for zapped functions and the 2158 // corresponding call sites. 2159 for (Function *F : FuncZappedReturn) { 2160 for (Argument &A : F->args()) 2161 F->removeParamAttr(A.getArgNo(), Attribute::Returned); 2162 for (Use &U : F->uses()) { 2163 // Skip over blockaddr users. 2164 if (isa<BlockAddress>(U.getUser())) 2165 continue; 2166 CallBase *CB = cast<CallBase>(U.getUser()); 2167 for (Use &Arg : CB->args()) 2168 CB->removeParamAttr(CB->getArgOperandNo(&Arg), Attribute::Returned); 2169 } 2170 } 2171 2172 // If we inferred constant or undef values for globals variables, we can 2173 // delete the global and any stores that remain to it. 2174 for (auto &I : make_early_inc_range(Solver.getTrackedGlobals())) { 2175 GlobalVariable *GV = I.first; 2176 if (isOverdefined(I.second)) 2177 continue; 2178 LLVM_DEBUG(dbgs() << "Found that GV '" << GV->getName() 2179 << "' is constant!\n"); 2180 while (!GV->use_empty()) { 2181 StoreInst *SI = cast<StoreInst>(GV->user_back()); 2182 SI->eraseFromParent(); 2183 MadeChanges = true; 2184 } 2185 M.getGlobalList().erase(GV); 2186 ++IPNumGlobalConst; 2187 } 2188 2189 return MadeChanges; 2190 } 2191