1 //===- SCCP.cpp - Sparse Conditional Constant Propagation -----------------===// 2 // 3 // The LLVM Compiler Infrastructure 4 // 5 // This file is distributed under the University of Illinois Open Source 6 // License. See LICENSE.TXT for details. 7 // 8 //===----------------------------------------------------------------------===// 9 // 10 // This file implements sparse conditional constant propagation and merging: 11 // 12 // Specifically, this: 13 // * Assumes values are constant unless proven otherwise 14 // * Assumes BasicBlocks are dead unless proven otherwise 15 // * Proves values to be constant, and replaces them with constants 16 // * Proves conditional branches to be unconditional 17 // 18 //===----------------------------------------------------------------------===// 19 20 #define DEBUG_TYPE "sccp" 21 #include "llvm/Transforms/Scalar.h" 22 #include "llvm/Transforms/IPO.h" 23 #include "llvm/Constants.h" 24 #include "llvm/DerivedTypes.h" 25 #include "llvm/Instructions.h" 26 #include "llvm/Pass.h" 27 #include "llvm/Analysis/ConstantFolding.h" 28 #include "llvm/Analysis/ValueTracking.h" 29 #include "llvm/Transforms/Utils/Local.h" 30 #include "llvm/Target/TargetData.h" 31 #include "llvm/Support/CallSite.h" 32 #include "llvm/Support/Debug.h" 33 #include "llvm/Support/ErrorHandling.h" 34 #include "llvm/Support/InstVisitor.h" 35 #include "llvm/Support/raw_ostream.h" 36 #include "llvm/ADT/DenseMap.h" 37 #include "llvm/ADT/DenseSet.h" 38 #include "llvm/ADT/PointerIntPair.h" 39 #include "llvm/ADT/SmallPtrSet.h" 40 #include "llvm/ADT/SmallVector.h" 41 #include "llvm/ADT/Statistic.h" 42 #include "llvm/ADT/STLExtras.h" 43 #include <algorithm> 44 #include <map> 45 using namespace llvm; 46 47 STATISTIC(NumInstRemoved, "Number of instructions removed"); 48 STATISTIC(NumDeadBlocks , "Number of basic blocks unreachable"); 49 50 STATISTIC(IPNumInstRemoved, "Number of instructions removed by IPSCCP"); 51 STATISTIC(IPNumArgsElimed ,"Number of arguments constant propagated by IPSCCP"); 52 STATISTIC(IPNumGlobalConst, "Number of globals found to be constant by IPSCCP"); 53 54 namespace { 55 /// LatticeVal class - This class represents the different lattice values that 56 /// an LLVM value may occupy. It is a simple class with value semantics. 57 /// 58 class LatticeVal { 59 enum LatticeValueTy { 60 /// undefined - This LLVM Value has no known value yet. 61 undefined, 62 63 /// constant - This LLVM Value has a specific constant value. 64 constant, 65 66 /// forcedconstant - This LLVM Value was thought to be undef until 67 /// ResolvedUndefsIn. This is treated just like 'constant', but if merged 68 /// with another (different) constant, it goes to overdefined, instead of 69 /// asserting. 70 forcedconstant, 71 72 /// overdefined - This instruction is not known to be constant, and we know 73 /// it has a value. 74 overdefined 75 }; 76 77 /// Val: This stores the current lattice value along with the Constant* for 78 /// the constant if this is a 'constant' or 'forcedconstant' value. 79 PointerIntPair<Constant *, 2, LatticeValueTy> Val; 80 81 LatticeValueTy getLatticeValue() const { 82 return Val.getInt(); 83 } 84 85 public: 86 LatticeVal() : Val(0, undefined) {} 87 88 bool isUndefined() const { return getLatticeValue() == undefined; } 89 bool isConstant() const { 90 return getLatticeValue() == constant || getLatticeValue() == forcedconstant; 91 } 92 bool isOverdefined() const { return getLatticeValue() == overdefined; } 93 94 Constant *getConstant() const { 95 assert(isConstant() && "Cannot get the constant of a non-constant!"); 96 return Val.getPointer(); 97 } 98 99 /// markOverdefined - Return true if this is a change in status. 100 bool markOverdefined() { 101 if (isOverdefined()) 102 return false; 103 104 Val.setInt(overdefined); 105 return true; 106 } 107 108 /// markConstant - Return true if this is a change in status. 109 bool markConstant(Constant *V) { 110 if (getLatticeValue() == constant) { // Constant but not forcedconstant. 111 assert(getConstant() == V && "Marking constant with different value"); 112 return false; 113 } 114 115 if (isUndefined()) { 116 Val.setInt(constant); 117 assert(V && "Marking constant with NULL"); 118 Val.setPointer(V); 119 } else { 120 assert(getLatticeValue() == forcedconstant && 121 "Cannot move from overdefined to constant!"); 122 // Stay at forcedconstant if the constant is the same. 123 if (V == getConstant()) return false; 124 125 // Otherwise, we go to overdefined. Assumptions made based on the 126 // forced value are possibly wrong. Assuming this is another constant 127 // could expose a contradiction. 128 Val.setInt(overdefined); 129 } 130 return true; 131 } 132 133 /// getConstantInt - If this is a constant with a ConstantInt value, return it 134 /// otherwise return null. 135 ConstantInt *getConstantInt() const { 136 if (isConstant()) 137 return dyn_cast<ConstantInt>(getConstant()); 138 return 0; 139 } 140 141 void markForcedConstant(Constant *V) { 142 assert(isUndefined() && "Can't force a defined value!"); 143 Val.setInt(forcedconstant); 144 Val.setPointer(V); 145 } 146 }; 147 } // end anonymous namespace. 148 149 150 namespace { 151 152 //===----------------------------------------------------------------------===// 153 // 154 /// SCCPSolver - This class is a general purpose solver for Sparse Conditional 155 /// Constant Propagation. 156 /// 157 class SCCPSolver : public InstVisitor<SCCPSolver> { 158 const TargetData *TD; 159 SmallPtrSet<BasicBlock*, 8> BBExecutable;// The BBs that are executable. 160 DenseMap<Value*, LatticeVal> ValueState; // The state each value is in. 161 162 /// StructValueState - This maintains ValueState for values that have 163 /// StructType, for example for formal arguments, calls, insertelement, etc. 164 /// 165 DenseMap<std::pair<Value*, unsigned>, LatticeVal> StructValueState; 166 167 /// GlobalValue - If we are tracking any values for the contents of a global 168 /// variable, we keep a mapping from the constant accessor to the element of 169 /// the global, to the currently known value. If the value becomes 170 /// overdefined, it's entry is simply removed from this map. 171 DenseMap<GlobalVariable*, LatticeVal> TrackedGlobals; 172 173 /// TrackedRetVals - If we are tracking arguments into and the return 174 /// value out of a function, it will have an entry in this map, indicating 175 /// what the known return value for the function is. 176 DenseMap<Function*, LatticeVal> TrackedRetVals; 177 178 /// TrackedMultipleRetVals - Same as TrackedRetVals, but used for functions 179 /// that return multiple values. 180 DenseMap<std::pair<Function*, unsigned>, LatticeVal> TrackedMultipleRetVals; 181 182 /// MRVFunctionsTracked - Each function in TrackedMultipleRetVals is 183 /// represented here for efficient lookup. 184 SmallPtrSet<Function*, 16> MRVFunctionsTracked; 185 186 /// TrackingIncomingArguments - This is the set of functions for whose 187 /// arguments we make optimistic assumptions about and try to prove as 188 /// constants. 189 SmallPtrSet<Function*, 16> TrackingIncomingArguments; 190 191 /// The reason for two worklists is that overdefined is the lowest state 192 /// on the lattice, and moving things to overdefined as fast as possible 193 /// makes SCCP converge much faster. 194 /// 195 /// By having a separate worklist, we accomplish this because everything 196 /// possibly overdefined will become overdefined at the soonest possible 197 /// point. 198 SmallVector<Value*, 64> OverdefinedInstWorkList; 199 SmallVector<Value*, 64> InstWorkList; 200 201 202 SmallVector<BasicBlock*, 64> BBWorkList; // The BasicBlock work list 203 204 /// UsersOfOverdefinedPHIs - Keep track of any users of PHI nodes that are not 205 /// overdefined, despite the fact that the PHI node is overdefined. 206 std::multimap<PHINode*, Instruction*> UsersOfOverdefinedPHIs; 207 208 /// KnownFeasibleEdges - Entries in this set are edges which have already had 209 /// PHI nodes retriggered. 210 typedef std::pair<BasicBlock*, BasicBlock*> Edge; 211 DenseSet<Edge> KnownFeasibleEdges; 212 public: 213 SCCPSolver(const TargetData *td) : TD(td) {} 214 215 /// MarkBlockExecutable - This method can be used by clients to mark all of 216 /// the blocks that are known to be intrinsically live in the processed unit. 217 /// 218 /// This returns true if the block was not considered live before. 219 bool MarkBlockExecutable(BasicBlock *BB) { 220 if (!BBExecutable.insert(BB)) return false; 221 DEBUG(dbgs() << "Marking Block Executable: " << BB->getName() << "\n"); 222 BBWorkList.push_back(BB); // Add the block to the work list! 223 return true; 224 } 225 226 /// TrackValueOfGlobalVariable - Clients can use this method to 227 /// inform the SCCPSolver that it should track loads and stores to the 228 /// specified global variable if it can. This is only legal to call if 229 /// performing Interprocedural SCCP. 230 void TrackValueOfGlobalVariable(GlobalVariable *GV) { 231 // We only track the contents of scalar globals. 232 if (GV->getType()->getElementType()->isSingleValueType()) { 233 LatticeVal &IV = TrackedGlobals[GV]; 234 if (!isa<UndefValue>(GV->getInitializer())) 235 IV.markConstant(GV->getInitializer()); 236 } 237 } 238 239 /// AddTrackedFunction - If the SCCP solver is supposed to track calls into 240 /// and out of the specified function (which cannot have its address taken), 241 /// this method must be called. 242 void AddTrackedFunction(Function *F) { 243 // Add an entry, F -> undef. 244 if (const StructType *STy = dyn_cast<StructType>(F->getReturnType())) { 245 MRVFunctionsTracked.insert(F); 246 for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) 247 TrackedMultipleRetVals.insert(std::make_pair(std::make_pair(F, i), 248 LatticeVal())); 249 } else 250 TrackedRetVals.insert(std::make_pair(F, LatticeVal())); 251 } 252 253 void AddArgumentTrackedFunction(Function *F) { 254 TrackingIncomingArguments.insert(F); 255 } 256 257 /// Solve - Solve for constants and executable blocks. 258 /// 259 void Solve(); 260 261 /// ResolvedUndefsIn - While solving the dataflow for a function, we assume 262 /// that branches on undef values cannot reach any of their successors. 263 /// However, this is not a safe assumption. After we solve dataflow, this 264 /// method should be use to handle this. If this returns true, the solver 265 /// should be rerun. 266 bool ResolvedUndefsIn(Function &F); 267 268 bool isBlockExecutable(BasicBlock *BB) const { 269 return BBExecutable.count(BB); 270 } 271 272 LatticeVal getLatticeValueFor(Value *V) const { 273 DenseMap<Value*, LatticeVal>::const_iterator I = ValueState.find(V); 274 assert(I != ValueState.end() && "V is not in valuemap!"); 275 return I->second; 276 } 277 278 /*LatticeVal getStructLatticeValueFor(Value *V, unsigned i) const { 279 DenseMap<std::pair<Value*, unsigned>, LatticeVal>::const_iterator I = 280 StructValueState.find(std::make_pair(V, i)); 281 assert(I != StructValueState.end() && "V is not in valuemap!"); 282 return I->second; 283 }*/ 284 285 /// getTrackedRetVals - Get the inferred return value map. 286 /// 287 const DenseMap<Function*, LatticeVal> &getTrackedRetVals() { 288 return TrackedRetVals; 289 } 290 291 /// getTrackedGlobals - Get and return the set of inferred initializers for 292 /// global variables. 293 const DenseMap<GlobalVariable*, LatticeVal> &getTrackedGlobals() { 294 return TrackedGlobals; 295 } 296 297 void markOverdefined(Value *V) { 298 assert(!V->getType()->isStructTy() && "Should use other method"); 299 markOverdefined(ValueState[V], V); 300 } 301 302 /// markAnythingOverdefined - Mark the specified value overdefined. This 303 /// works with both scalars and structs. 304 void markAnythingOverdefined(Value *V) { 305 if (const StructType *STy = dyn_cast<StructType>(V->getType())) 306 for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) 307 markOverdefined(getStructValueState(V, i), V); 308 else 309 markOverdefined(V); 310 } 311 312 private: 313 // markConstant - Make a value be marked as "constant". If the value 314 // is not already a constant, add it to the instruction work list so that 315 // the users of the instruction are updated later. 316 // 317 void markConstant(LatticeVal &IV, Value *V, Constant *C) { 318 if (!IV.markConstant(C)) return; 319 DEBUG(dbgs() << "markConstant: " << *C << ": " << *V << '\n'); 320 if (IV.isOverdefined()) 321 OverdefinedInstWorkList.push_back(V); 322 else 323 InstWorkList.push_back(V); 324 } 325 326 void markConstant(Value *V, Constant *C) { 327 assert(!V->getType()->isStructTy() && "Should use other method"); 328 markConstant(ValueState[V], V, C); 329 } 330 331 void markForcedConstant(Value *V, Constant *C) { 332 assert(!V->getType()->isStructTy() && "Should use other method"); 333 LatticeVal &IV = ValueState[V]; 334 IV.markForcedConstant(C); 335 DEBUG(dbgs() << "markForcedConstant: " << *C << ": " << *V << '\n'); 336 if (IV.isOverdefined()) 337 OverdefinedInstWorkList.push_back(V); 338 else 339 InstWorkList.push_back(V); 340 } 341 342 343 // markOverdefined - Make a value be marked as "overdefined". If the 344 // value is not already overdefined, add it to the overdefined instruction 345 // work list so that the users of the instruction are updated later. 346 void markOverdefined(LatticeVal &IV, Value *V) { 347 if (!IV.markOverdefined()) return; 348 349 DEBUG(dbgs() << "markOverdefined: "; 350 if (Function *F = dyn_cast<Function>(V)) 351 dbgs() << "Function '" << F->getName() << "'\n"; 352 else 353 dbgs() << *V << '\n'); 354 // Only instructions go on the work list 355 OverdefinedInstWorkList.push_back(V); 356 } 357 358 void mergeInValue(LatticeVal &IV, Value *V, LatticeVal MergeWithV) { 359 if (IV.isOverdefined() || MergeWithV.isUndefined()) 360 return; // Noop. 361 if (MergeWithV.isOverdefined()) 362 markOverdefined(IV, V); 363 else if (IV.isUndefined()) 364 markConstant(IV, V, MergeWithV.getConstant()); 365 else if (IV.getConstant() != MergeWithV.getConstant()) 366 markOverdefined(IV, V); 367 } 368 369 void mergeInValue(Value *V, LatticeVal MergeWithV) { 370 assert(!V->getType()->isStructTy() && "Should use other method"); 371 mergeInValue(ValueState[V], V, MergeWithV); 372 } 373 374 375 /// getValueState - Return the LatticeVal object that corresponds to the 376 /// value. This function handles the case when the value hasn't been seen yet 377 /// by properly seeding constants etc. 378 LatticeVal &getValueState(Value *V) { 379 assert(!V->getType()->isStructTy() && "Should use getStructValueState"); 380 381 std::pair<DenseMap<Value*, LatticeVal>::iterator, bool> I = 382 ValueState.insert(std::make_pair(V, LatticeVal())); 383 LatticeVal &LV = I.first->second; 384 385 if (!I.second) 386 return LV; // Common case, already in the map. 387 388 if (Constant *C = dyn_cast<Constant>(V)) { 389 // Undef values remain undefined. 390 if (!isa<UndefValue>(V)) 391 LV.markConstant(C); // Constants are constant 392 } 393 394 // All others are underdefined by default. 395 return LV; 396 } 397 398 /// getStructValueState - Return the LatticeVal object that corresponds to the 399 /// value/field pair. This function handles the case when the value hasn't 400 /// been seen yet by properly seeding constants etc. 401 LatticeVal &getStructValueState(Value *V, unsigned i) { 402 assert(V->getType()->isStructTy() && "Should use getValueState"); 403 assert(i < cast<StructType>(V->getType())->getNumElements() && 404 "Invalid element #"); 405 406 std::pair<DenseMap<std::pair<Value*, unsigned>, LatticeVal>::iterator, 407 bool> I = StructValueState.insert( 408 std::make_pair(std::make_pair(V, i), LatticeVal())); 409 LatticeVal &LV = I.first->second; 410 411 if (!I.second) 412 return LV; // Common case, already in the map. 413 414 if (Constant *C = dyn_cast<Constant>(V)) { 415 if (isa<UndefValue>(C)) 416 ; // Undef values remain undefined. 417 else if (ConstantStruct *CS = dyn_cast<ConstantStruct>(C)) 418 LV.markConstant(CS->getOperand(i)); // Constants are constant. 419 else if (isa<ConstantAggregateZero>(C)) { 420 const Type *FieldTy = cast<StructType>(V->getType())->getElementType(i); 421 LV.markConstant(Constant::getNullValue(FieldTy)); 422 } else 423 LV.markOverdefined(); // Unknown sort of constant. 424 } 425 426 // All others are underdefined by default. 427 return LV; 428 } 429 430 431 /// markEdgeExecutable - Mark a basic block as executable, adding it to the BB 432 /// work list if it is not already executable. 433 void markEdgeExecutable(BasicBlock *Source, BasicBlock *Dest) { 434 if (!KnownFeasibleEdges.insert(Edge(Source, Dest)).second) 435 return; // This edge is already known to be executable! 436 437 if (!MarkBlockExecutable(Dest)) { 438 // If the destination is already executable, we just made an *edge* 439 // feasible that wasn't before. Revisit the PHI nodes in the block 440 // because they have potentially new operands. 441 DEBUG(dbgs() << "Marking Edge Executable: " << Source->getName() 442 << " -> " << Dest->getName() << "\n"); 443 444 PHINode *PN; 445 for (BasicBlock::iterator I = Dest->begin(); 446 (PN = dyn_cast<PHINode>(I)); ++I) 447 visitPHINode(*PN); 448 } 449 } 450 451 // getFeasibleSuccessors - Return a vector of booleans to indicate which 452 // successors are reachable from a given terminator instruction. 453 // 454 void getFeasibleSuccessors(TerminatorInst &TI, SmallVector<bool, 16> &Succs); 455 456 // isEdgeFeasible - Return true if the control flow edge from the 'From' basic 457 // block to the 'To' basic block is currently feasible. 458 // 459 bool isEdgeFeasible(BasicBlock *From, BasicBlock *To); 460 461 // OperandChangedState - This method is invoked on all of the users of an 462 // instruction that was just changed state somehow. Based on this 463 // information, we need to update the specified user of this instruction. 464 // 465 void OperandChangedState(Instruction *I) { 466 if (BBExecutable.count(I->getParent())) // Inst is executable? 467 visit(*I); 468 } 469 470 /// RemoveFromOverdefinedPHIs - If I has any entries in the 471 /// UsersOfOverdefinedPHIs map for PN, remove them now. 472 void RemoveFromOverdefinedPHIs(Instruction *I, PHINode *PN) { 473 if (UsersOfOverdefinedPHIs.empty()) return; 474 std::multimap<PHINode*, Instruction*>::iterator It, E; 475 tie(It, E) = UsersOfOverdefinedPHIs.equal_range(PN); 476 while (It != E) { 477 if (It->second == I) 478 UsersOfOverdefinedPHIs.erase(It++); 479 else 480 ++It; 481 } 482 } 483 484 private: 485 friend class InstVisitor<SCCPSolver>; 486 487 // visit implementations - Something changed in this instruction. Either an 488 // operand made a transition, or the instruction is newly executable. Change 489 // the value type of I to reflect these changes if appropriate. 490 void visitPHINode(PHINode &I); 491 492 // Terminators 493 void visitReturnInst(ReturnInst &I); 494 void visitTerminatorInst(TerminatorInst &TI); 495 496 void visitCastInst(CastInst &I); 497 void visitSelectInst(SelectInst &I); 498 void visitBinaryOperator(Instruction &I); 499 void visitCmpInst(CmpInst &I); 500 void visitExtractElementInst(ExtractElementInst &I); 501 void visitInsertElementInst(InsertElementInst &I); 502 void visitShuffleVectorInst(ShuffleVectorInst &I); 503 void visitExtractValueInst(ExtractValueInst &EVI); 504 void visitInsertValueInst(InsertValueInst &IVI); 505 506 // Instructions that cannot be folded away. 507 void visitStoreInst (StoreInst &I); 508 void visitLoadInst (LoadInst &I); 509 void visitGetElementPtrInst(GetElementPtrInst &I); 510 void visitCallInst (CallInst &I) { 511 visitCallSite(&I); 512 } 513 void visitInvokeInst (InvokeInst &II) { 514 visitCallSite(&II); 515 visitTerminatorInst(II); 516 } 517 void visitCallSite (CallSite CS); 518 void visitUnwindInst (TerminatorInst &I) { /*returns void*/ } 519 void visitUnreachableInst(TerminatorInst &I) { /*returns void*/ } 520 void visitAllocaInst (Instruction &I) { markOverdefined(&I); } 521 void visitVAArgInst (Instruction &I) { markAnythingOverdefined(&I); } 522 523 void visitInstruction(Instruction &I) { 524 // If a new instruction is added to LLVM that we don't handle. 525 dbgs() << "SCCP: Don't know how to handle: " << I; 526 markAnythingOverdefined(&I); // Just in case 527 } 528 }; 529 530 } // end anonymous namespace 531 532 533 // getFeasibleSuccessors - Return a vector of booleans to indicate which 534 // successors are reachable from a given terminator instruction. 535 // 536 void SCCPSolver::getFeasibleSuccessors(TerminatorInst &TI, 537 SmallVector<bool, 16> &Succs) { 538 Succs.resize(TI.getNumSuccessors()); 539 if (BranchInst *BI = dyn_cast<BranchInst>(&TI)) { 540 if (BI->isUnconditional()) { 541 Succs[0] = true; 542 return; 543 } 544 545 LatticeVal BCValue = getValueState(BI->getCondition()); 546 ConstantInt *CI = BCValue.getConstantInt(); 547 if (CI == 0) { 548 // Overdefined condition variables, and branches on unfoldable constant 549 // conditions, mean the branch could go either way. 550 if (!BCValue.isUndefined()) 551 Succs[0] = Succs[1] = true; 552 return; 553 } 554 555 // Constant condition variables mean the branch can only go a single way. 556 Succs[CI->isZero()] = true; 557 return; 558 } 559 560 if (isa<InvokeInst>(TI)) { 561 // Invoke instructions successors are always executable. 562 Succs[0] = Succs[1] = true; 563 return; 564 } 565 566 if (SwitchInst *SI = dyn_cast<SwitchInst>(&TI)) { 567 LatticeVal SCValue = getValueState(SI->getCondition()); 568 ConstantInt *CI = SCValue.getConstantInt(); 569 570 if (CI == 0) { // Overdefined or undefined condition? 571 // All destinations are executable! 572 if (!SCValue.isUndefined()) 573 Succs.assign(TI.getNumSuccessors(), true); 574 return; 575 } 576 577 Succs[SI->findCaseValue(CI)] = true; 578 return; 579 } 580 581 // TODO: This could be improved if the operand is a [cast of a] BlockAddress. 582 if (isa<IndirectBrInst>(&TI)) { 583 // Just mark all destinations executable! 584 Succs.assign(TI.getNumSuccessors(), true); 585 return; 586 } 587 588 #ifndef NDEBUG 589 dbgs() << "Unknown terminator instruction: " << TI << '\n'; 590 #endif 591 llvm_unreachable("SCCP: Don't know how to handle this terminator!"); 592 } 593 594 595 // isEdgeFeasible - Return true if the control flow edge from the 'From' basic 596 // block to the 'To' basic block is currently feasible. 597 // 598 bool SCCPSolver::isEdgeFeasible(BasicBlock *From, BasicBlock *To) { 599 assert(BBExecutable.count(To) && "Dest should always be alive!"); 600 601 // Make sure the source basic block is executable!! 602 if (!BBExecutable.count(From)) return false; 603 604 // Check to make sure this edge itself is actually feasible now. 605 TerminatorInst *TI = From->getTerminator(); 606 if (BranchInst *BI = dyn_cast<BranchInst>(TI)) { 607 if (BI->isUnconditional()) 608 return true; 609 610 LatticeVal BCValue = getValueState(BI->getCondition()); 611 612 // Overdefined condition variables mean the branch could go either way, 613 // undef conditions mean that neither edge is feasible yet. 614 ConstantInt *CI = BCValue.getConstantInt(); 615 if (CI == 0) 616 return !BCValue.isUndefined(); 617 618 // Constant condition variables mean the branch can only go a single way. 619 return BI->getSuccessor(CI->isZero()) == To; 620 } 621 622 // Invoke instructions successors are always executable. 623 if (isa<InvokeInst>(TI)) 624 return true; 625 626 if (SwitchInst *SI = dyn_cast<SwitchInst>(TI)) { 627 LatticeVal SCValue = getValueState(SI->getCondition()); 628 ConstantInt *CI = SCValue.getConstantInt(); 629 630 if (CI == 0) 631 return !SCValue.isUndefined(); 632 633 // Make sure to skip the "default value" which isn't a value 634 for (unsigned i = 1, E = SI->getNumSuccessors(); i != E; ++i) 635 if (SI->getSuccessorValue(i) == CI) // Found the taken branch. 636 return SI->getSuccessor(i) == To; 637 638 // If the constant value is not equal to any of the branches, we must 639 // execute default branch. 640 return SI->getDefaultDest() == To; 641 } 642 643 // Just mark all destinations executable! 644 // TODO: This could be improved if the operand is a [cast of a] BlockAddress. 645 if (isa<IndirectBrInst>(&TI)) 646 return true; 647 648 #ifndef NDEBUG 649 dbgs() << "Unknown terminator instruction: " << *TI << '\n'; 650 #endif 651 llvm_unreachable(0); 652 } 653 654 // visit Implementations - Something changed in this instruction, either an 655 // operand made a transition, or the instruction is newly executable. Change 656 // the value type of I to reflect these changes if appropriate. This method 657 // makes sure to do the following actions: 658 // 659 // 1. If a phi node merges two constants in, and has conflicting value coming 660 // from different branches, or if the PHI node merges in an overdefined 661 // value, then the PHI node becomes overdefined. 662 // 2. If a phi node merges only constants in, and they all agree on value, the 663 // PHI node becomes a constant value equal to that. 664 // 3. If V <- x (op) y && isConstant(x) && isConstant(y) V = Constant 665 // 4. If V <- x (op) y && (isOverdefined(x) || isOverdefined(y)) V = Overdefined 666 // 5. If V <- MEM or V <- CALL or V <- (unknown) then V = Overdefined 667 // 6. If a conditional branch has a value that is constant, make the selected 668 // destination executable 669 // 7. If a conditional branch has a value that is overdefined, make all 670 // successors executable. 671 // 672 void SCCPSolver::visitPHINode(PHINode &PN) { 673 // If this PN returns a struct, just mark the result overdefined. 674 // TODO: We could do a lot better than this if code actually uses this. 675 if (PN.getType()->isStructTy()) 676 return markAnythingOverdefined(&PN); 677 678 if (getValueState(&PN).isOverdefined()) { 679 // There may be instructions using this PHI node that are not overdefined 680 // themselves. If so, make sure that they know that the PHI node operand 681 // changed. 682 std::multimap<PHINode*, Instruction*>::iterator I, E; 683 tie(I, E) = UsersOfOverdefinedPHIs.equal_range(&PN); 684 if (I == E) 685 return; 686 687 SmallVector<Instruction*, 16> Users; 688 for (; I != E; ++I) 689 Users.push_back(I->second); 690 while (!Users.empty()) 691 visit(Users.pop_back_val()); 692 return; // Quick exit 693 } 694 695 // Super-extra-high-degree PHI nodes are unlikely to ever be marked constant, 696 // and slow us down a lot. Just mark them overdefined. 697 if (PN.getNumIncomingValues() > 64) 698 return markOverdefined(&PN); 699 700 // Look at all of the executable operands of the PHI node. If any of them 701 // are overdefined, the PHI becomes overdefined as well. If they are all 702 // constant, and they agree with each other, the PHI becomes the identical 703 // constant. If they are constant and don't agree, the PHI is overdefined. 704 // If there are no executable operands, the PHI remains undefined. 705 // 706 Constant *OperandVal = 0; 707 for (unsigned i = 0, e = PN.getNumIncomingValues(); i != e; ++i) { 708 LatticeVal IV = getValueState(PN.getIncomingValue(i)); 709 if (IV.isUndefined()) continue; // Doesn't influence PHI node. 710 711 if (!isEdgeFeasible(PN.getIncomingBlock(i), PN.getParent())) 712 continue; 713 714 if (IV.isOverdefined()) // PHI node becomes overdefined! 715 return markOverdefined(&PN); 716 717 if (OperandVal == 0) { // Grab the first value. 718 OperandVal = IV.getConstant(); 719 continue; 720 } 721 722 // There is already a reachable operand. If we conflict with it, 723 // then the PHI node becomes overdefined. If we agree with it, we 724 // can continue on. 725 726 // Check to see if there are two different constants merging, if so, the PHI 727 // node is overdefined. 728 if (IV.getConstant() != OperandVal) 729 return markOverdefined(&PN); 730 } 731 732 // If we exited the loop, this means that the PHI node only has constant 733 // arguments that agree with each other(and OperandVal is the constant) or 734 // OperandVal is null because there are no defined incoming arguments. If 735 // this is the case, the PHI remains undefined. 736 // 737 if (OperandVal) 738 markConstant(&PN, OperandVal); // Acquire operand value 739 } 740 741 742 743 744 void SCCPSolver::visitReturnInst(ReturnInst &I) { 745 if (I.getNumOperands() == 0) return; // ret void 746 747 Function *F = I.getParent()->getParent(); 748 Value *ResultOp = I.getOperand(0); 749 750 // If we are tracking the return value of this function, merge it in. 751 if (!TrackedRetVals.empty() && !ResultOp->getType()->isStructTy()) { 752 DenseMap<Function*, LatticeVal>::iterator TFRVI = 753 TrackedRetVals.find(F); 754 if (TFRVI != TrackedRetVals.end()) { 755 mergeInValue(TFRVI->second, F, getValueState(ResultOp)); 756 return; 757 } 758 } 759 760 // Handle functions that return multiple values. 761 if (!TrackedMultipleRetVals.empty()) { 762 if (const StructType *STy = dyn_cast<StructType>(ResultOp->getType())) 763 if (MRVFunctionsTracked.count(F)) 764 for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) 765 mergeInValue(TrackedMultipleRetVals[std::make_pair(F, i)], F, 766 getStructValueState(ResultOp, i)); 767 768 } 769 } 770 771 void SCCPSolver::visitTerminatorInst(TerminatorInst &TI) { 772 SmallVector<bool, 16> SuccFeasible; 773 getFeasibleSuccessors(TI, SuccFeasible); 774 775 BasicBlock *BB = TI.getParent(); 776 777 // Mark all feasible successors executable. 778 for (unsigned i = 0, e = SuccFeasible.size(); i != e; ++i) 779 if (SuccFeasible[i]) 780 markEdgeExecutable(BB, TI.getSuccessor(i)); 781 } 782 783 void SCCPSolver::visitCastInst(CastInst &I) { 784 LatticeVal OpSt = getValueState(I.getOperand(0)); 785 if (OpSt.isOverdefined()) // Inherit overdefinedness of operand 786 markOverdefined(&I); 787 else if (OpSt.isConstant()) // Propagate constant value 788 markConstant(&I, ConstantExpr::getCast(I.getOpcode(), 789 OpSt.getConstant(), I.getType())); 790 } 791 792 793 void SCCPSolver::visitExtractValueInst(ExtractValueInst &EVI) { 794 // If this returns a struct, mark all elements over defined, we don't track 795 // structs in structs. 796 if (EVI.getType()->isStructTy()) 797 return markAnythingOverdefined(&EVI); 798 799 // If this is extracting from more than one level of struct, we don't know. 800 if (EVI.getNumIndices() != 1) 801 return markOverdefined(&EVI); 802 803 Value *AggVal = EVI.getAggregateOperand(); 804 if (AggVal->getType()->isStructTy()) { 805 unsigned i = *EVI.idx_begin(); 806 LatticeVal EltVal = getStructValueState(AggVal, i); 807 mergeInValue(getValueState(&EVI), &EVI, EltVal); 808 } else { 809 // Otherwise, must be extracting from an array. 810 return markOverdefined(&EVI); 811 } 812 } 813 814 void SCCPSolver::visitInsertValueInst(InsertValueInst &IVI) { 815 const StructType *STy = dyn_cast<StructType>(IVI.getType()); 816 if (STy == 0) 817 return markOverdefined(&IVI); 818 819 // If this has more than one index, we can't handle it, drive all results to 820 // undef. 821 if (IVI.getNumIndices() != 1) 822 return markAnythingOverdefined(&IVI); 823 824 Value *Aggr = IVI.getAggregateOperand(); 825 unsigned Idx = *IVI.idx_begin(); 826 827 // Compute the result based on what we're inserting. 828 for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) { 829 // This passes through all values that aren't the inserted element. 830 if (i != Idx) { 831 LatticeVal EltVal = getStructValueState(Aggr, i); 832 mergeInValue(getStructValueState(&IVI, i), &IVI, EltVal); 833 continue; 834 } 835 836 Value *Val = IVI.getInsertedValueOperand(); 837 if (Val->getType()->isStructTy()) 838 // We don't track structs in structs. 839 markOverdefined(getStructValueState(&IVI, i), &IVI); 840 else { 841 LatticeVal InVal = getValueState(Val); 842 mergeInValue(getStructValueState(&IVI, i), &IVI, InVal); 843 } 844 } 845 } 846 847 void SCCPSolver::visitSelectInst(SelectInst &I) { 848 // If this select returns a struct, just mark the result overdefined. 849 // TODO: We could do a lot better than this if code actually uses this. 850 if (I.getType()->isStructTy()) 851 return markAnythingOverdefined(&I); 852 853 LatticeVal CondValue = getValueState(I.getCondition()); 854 if (CondValue.isUndefined()) 855 return; 856 857 if (ConstantInt *CondCB = CondValue.getConstantInt()) { 858 Value *OpVal = CondCB->isZero() ? I.getFalseValue() : I.getTrueValue(); 859 mergeInValue(&I, getValueState(OpVal)); 860 return; 861 } 862 863 // Otherwise, the condition is overdefined or a constant we can't evaluate. 864 // See if we can produce something better than overdefined based on the T/F 865 // value. 866 LatticeVal TVal = getValueState(I.getTrueValue()); 867 LatticeVal FVal = getValueState(I.getFalseValue()); 868 869 // select ?, C, C -> C. 870 if (TVal.isConstant() && FVal.isConstant() && 871 TVal.getConstant() == FVal.getConstant()) 872 return markConstant(&I, FVal.getConstant()); 873 874 if (TVal.isUndefined()) // select ?, undef, X -> X. 875 return mergeInValue(&I, FVal); 876 if (FVal.isUndefined()) // select ?, X, undef -> X. 877 return mergeInValue(&I, TVal); 878 markOverdefined(&I); 879 } 880 881 // Handle Binary Operators. 882 void SCCPSolver::visitBinaryOperator(Instruction &I) { 883 LatticeVal V1State = getValueState(I.getOperand(0)); 884 LatticeVal V2State = getValueState(I.getOperand(1)); 885 886 LatticeVal &IV = ValueState[&I]; 887 if (IV.isOverdefined()) return; 888 889 if (V1State.isConstant() && V2State.isConstant()) 890 return markConstant(IV, &I, 891 ConstantExpr::get(I.getOpcode(), V1State.getConstant(), 892 V2State.getConstant())); 893 894 // If something is undef, wait for it to resolve. 895 if (!V1State.isOverdefined() && !V2State.isOverdefined()) 896 return; 897 898 // Otherwise, one of our operands is overdefined. Try to produce something 899 // better than overdefined with some tricks. 900 901 // If this is an AND or OR with 0 or -1, it doesn't matter that the other 902 // operand is overdefined. 903 if (I.getOpcode() == Instruction::And || I.getOpcode() == Instruction::Or) { 904 LatticeVal *NonOverdefVal = 0; 905 if (!V1State.isOverdefined()) 906 NonOverdefVal = &V1State; 907 else if (!V2State.isOverdefined()) 908 NonOverdefVal = &V2State; 909 910 if (NonOverdefVal) { 911 if (NonOverdefVal->isUndefined()) { 912 // Could annihilate value. 913 if (I.getOpcode() == Instruction::And) 914 markConstant(IV, &I, Constant::getNullValue(I.getType())); 915 else if (const VectorType *PT = dyn_cast<VectorType>(I.getType())) 916 markConstant(IV, &I, Constant::getAllOnesValue(PT)); 917 else 918 markConstant(IV, &I, 919 Constant::getAllOnesValue(I.getType())); 920 return; 921 } 922 923 if (I.getOpcode() == Instruction::And) { 924 // X and 0 = 0 925 if (NonOverdefVal->getConstant()->isNullValue()) 926 return markConstant(IV, &I, NonOverdefVal->getConstant()); 927 } else { 928 if (ConstantInt *CI = NonOverdefVal->getConstantInt()) 929 if (CI->isAllOnesValue()) // X or -1 = -1 930 return markConstant(IV, &I, NonOverdefVal->getConstant()); 931 } 932 } 933 } 934 935 936 // If both operands are PHI nodes, it is possible that this instruction has 937 // a constant value, despite the fact that the PHI node doesn't. Check for 938 // this condition now. 939 if (PHINode *PN1 = dyn_cast<PHINode>(I.getOperand(0))) 940 if (PHINode *PN2 = dyn_cast<PHINode>(I.getOperand(1))) 941 if (PN1->getParent() == PN2->getParent()) { 942 // Since the two PHI nodes are in the same basic block, they must have 943 // entries for the same predecessors. Walk the predecessor list, and 944 // if all of the incoming values are constants, and the result of 945 // evaluating this expression with all incoming value pairs is the 946 // same, then this expression is a constant even though the PHI node 947 // is not a constant! 948 LatticeVal Result; 949 for (unsigned i = 0, e = PN1->getNumIncomingValues(); i != e; ++i) { 950 LatticeVal In1 = getValueState(PN1->getIncomingValue(i)); 951 BasicBlock *InBlock = PN1->getIncomingBlock(i); 952 LatticeVal In2 =getValueState(PN2->getIncomingValueForBlock(InBlock)); 953 954 if (In1.isOverdefined() || In2.isOverdefined()) { 955 Result.markOverdefined(); 956 break; // Cannot fold this operation over the PHI nodes! 957 } 958 959 if (In1.isConstant() && In2.isConstant()) { 960 Constant *V = ConstantExpr::get(I.getOpcode(), In1.getConstant(), 961 In2.getConstant()); 962 if (Result.isUndefined()) 963 Result.markConstant(V); 964 else if (Result.isConstant() && Result.getConstant() != V) { 965 Result.markOverdefined(); 966 break; 967 } 968 } 969 } 970 971 // If we found a constant value here, then we know the instruction is 972 // constant despite the fact that the PHI nodes are overdefined. 973 if (Result.isConstant()) { 974 markConstant(IV, &I, Result.getConstant()); 975 // Remember that this instruction is virtually using the PHI node 976 // operands. 977 UsersOfOverdefinedPHIs.insert(std::make_pair(PN1, &I)); 978 UsersOfOverdefinedPHIs.insert(std::make_pair(PN2, &I)); 979 return; 980 } 981 982 if (Result.isUndefined()) 983 return; 984 985 // Okay, this really is overdefined now. Since we might have 986 // speculatively thought that this was not overdefined before, and 987 // added ourselves to the UsersOfOverdefinedPHIs list for the PHIs, 988 // make sure to clean out any entries that we put there, for 989 // efficiency. 990 RemoveFromOverdefinedPHIs(&I, PN1); 991 RemoveFromOverdefinedPHIs(&I, PN2); 992 } 993 994 markOverdefined(&I); 995 } 996 997 // Handle ICmpInst instruction. 998 void SCCPSolver::visitCmpInst(CmpInst &I) { 999 LatticeVal V1State = getValueState(I.getOperand(0)); 1000 LatticeVal V2State = getValueState(I.getOperand(1)); 1001 1002 LatticeVal &IV = ValueState[&I]; 1003 if (IV.isOverdefined()) return; 1004 1005 if (V1State.isConstant() && V2State.isConstant()) 1006 return markConstant(IV, &I, ConstantExpr::getCompare(I.getPredicate(), 1007 V1State.getConstant(), 1008 V2State.getConstant())); 1009 1010 // If operands are still undefined, wait for it to resolve. 1011 if (!V1State.isOverdefined() && !V2State.isOverdefined()) 1012 return; 1013 1014 // If something is overdefined, use some tricks to avoid ending up and over 1015 // defined if we can. 1016 1017 // If both operands are PHI nodes, it is possible that this instruction has 1018 // a constant value, despite the fact that the PHI node doesn't. Check for 1019 // this condition now. 1020 if (PHINode *PN1 = dyn_cast<PHINode>(I.getOperand(0))) 1021 if (PHINode *PN2 = dyn_cast<PHINode>(I.getOperand(1))) 1022 if (PN1->getParent() == PN2->getParent()) { 1023 // Since the two PHI nodes are in the same basic block, they must have 1024 // entries for the same predecessors. Walk the predecessor list, and 1025 // if all of the incoming values are constants, and the result of 1026 // evaluating this expression with all incoming value pairs is the 1027 // same, then this expression is a constant even though the PHI node 1028 // is not a constant! 1029 LatticeVal Result; 1030 for (unsigned i = 0, e = PN1->getNumIncomingValues(); i != e; ++i) { 1031 LatticeVal In1 = getValueState(PN1->getIncomingValue(i)); 1032 BasicBlock *InBlock = PN1->getIncomingBlock(i); 1033 LatticeVal In2 =getValueState(PN2->getIncomingValueForBlock(InBlock)); 1034 1035 if (In1.isOverdefined() || In2.isOverdefined()) { 1036 Result.markOverdefined(); 1037 break; // Cannot fold this operation over the PHI nodes! 1038 } 1039 1040 if (In1.isConstant() && In2.isConstant()) { 1041 Constant *V = ConstantExpr::getCompare(I.getPredicate(), 1042 In1.getConstant(), 1043 In2.getConstant()); 1044 if (Result.isUndefined()) 1045 Result.markConstant(V); 1046 else if (Result.isConstant() && Result.getConstant() != V) { 1047 Result.markOverdefined(); 1048 break; 1049 } 1050 } 1051 } 1052 1053 // If we found a constant value here, then we know the instruction is 1054 // constant despite the fact that the PHI nodes are overdefined. 1055 if (Result.isConstant()) { 1056 markConstant(&I, Result.getConstant()); 1057 // Remember that this instruction is virtually using the PHI node 1058 // operands. 1059 UsersOfOverdefinedPHIs.insert(std::make_pair(PN1, &I)); 1060 UsersOfOverdefinedPHIs.insert(std::make_pair(PN2, &I)); 1061 return; 1062 } 1063 1064 if (Result.isUndefined()) 1065 return; 1066 1067 // Okay, this really is overdefined now. Since we might have 1068 // speculatively thought that this was not overdefined before, and 1069 // added ourselves to the UsersOfOverdefinedPHIs list for the PHIs, 1070 // make sure to clean out any entries that we put there, for 1071 // efficiency. 1072 RemoveFromOverdefinedPHIs(&I, PN1); 1073 RemoveFromOverdefinedPHIs(&I, PN2); 1074 } 1075 1076 markOverdefined(&I); 1077 } 1078 1079 void SCCPSolver::visitExtractElementInst(ExtractElementInst &I) { 1080 // TODO : SCCP does not handle vectors properly. 1081 return markOverdefined(&I); 1082 1083 #if 0 1084 LatticeVal &ValState = getValueState(I.getOperand(0)); 1085 LatticeVal &IdxState = getValueState(I.getOperand(1)); 1086 1087 if (ValState.isOverdefined() || IdxState.isOverdefined()) 1088 markOverdefined(&I); 1089 else if(ValState.isConstant() && IdxState.isConstant()) 1090 markConstant(&I, ConstantExpr::getExtractElement(ValState.getConstant(), 1091 IdxState.getConstant())); 1092 #endif 1093 } 1094 1095 void SCCPSolver::visitInsertElementInst(InsertElementInst &I) { 1096 // TODO : SCCP does not handle vectors properly. 1097 return markOverdefined(&I); 1098 #if 0 1099 LatticeVal &ValState = getValueState(I.getOperand(0)); 1100 LatticeVal &EltState = getValueState(I.getOperand(1)); 1101 LatticeVal &IdxState = getValueState(I.getOperand(2)); 1102 1103 if (ValState.isOverdefined() || EltState.isOverdefined() || 1104 IdxState.isOverdefined()) 1105 markOverdefined(&I); 1106 else if(ValState.isConstant() && EltState.isConstant() && 1107 IdxState.isConstant()) 1108 markConstant(&I, ConstantExpr::getInsertElement(ValState.getConstant(), 1109 EltState.getConstant(), 1110 IdxState.getConstant())); 1111 else if (ValState.isUndefined() && EltState.isConstant() && 1112 IdxState.isConstant()) 1113 markConstant(&I,ConstantExpr::getInsertElement(UndefValue::get(I.getType()), 1114 EltState.getConstant(), 1115 IdxState.getConstant())); 1116 #endif 1117 } 1118 1119 void SCCPSolver::visitShuffleVectorInst(ShuffleVectorInst &I) { 1120 // TODO : SCCP does not handle vectors properly. 1121 return markOverdefined(&I); 1122 #if 0 1123 LatticeVal &V1State = getValueState(I.getOperand(0)); 1124 LatticeVal &V2State = getValueState(I.getOperand(1)); 1125 LatticeVal &MaskState = getValueState(I.getOperand(2)); 1126 1127 if (MaskState.isUndefined() || 1128 (V1State.isUndefined() && V2State.isUndefined())) 1129 return; // Undefined output if mask or both inputs undefined. 1130 1131 if (V1State.isOverdefined() || V2State.isOverdefined() || 1132 MaskState.isOverdefined()) { 1133 markOverdefined(&I); 1134 } else { 1135 // A mix of constant/undef inputs. 1136 Constant *V1 = V1State.isConstant() ? 1137 V1State.getConstant() : UndefValue::get(I.getType()); 1138 Constant *V2 = V2State.isConstant() ? 1139 V2State.getConstant() : UndefValue::get(I.getType()); 1140 Constant *Mask = MaskState.isConstant() ? 1141 MaskState.getConstant() : UndefValue::get(I.getOperand(2)->getType()); 1142 markConstant(&I, ConstantExpr::getShuffleVector(V1, V2, Mask)); 1143 } 1144 #endif 1145 } 1146 1147 // Handle getelementptr instructions. If all operands are constants then we 1148 // can turn this into a getelementptr ConstantExpr. 1149 // 1150 void SCCPSolver::visitGetElementPtrInst(GetElementPtrInst &I) { 1151 if (ValueState[&I].isOverdefined()) return; 1152 1153 SmallVector<Constant*, 8> Operands; 1154 Operands.reserve(I.getNumOperands()); 1155 1156 for (unsigned i = 0, e = I.getNumOperands(); i != e; ++i) { 1157 LatticeVal State = getValueState(I.getOperand(i)); 1158 if (State.isUndefined()) 1159 return; // Operands are not resolved yet. 1160 1161 if (State.isOverdefined()) 1162 return markOverdefined(&I); 1163 1164 assert(State.isConstant() && "Unknown state!"); 1165 Operands.push_back(State.getConstant()); 1166 } 1167 1168 Constant *Ptr = Operands[0]; 1169 markConstant(&I, ConstantExpr::getGetElementPtr(Ptr, &Operands[0]+1, 1170 Operands.size()-1)); 1171 } 1172 1173 void SCCPSolver::visitStoreInst(StoreInst &SI) { 1174 // If this store is of a struct, ignore it. 1175 if (SI.getOperand(0)->getType()->isStructTy()) 1176 return; 1177 1178 if (TrackedGlobals.empty() || !isa<GlobalVariable>(SI.getOperand(1))) 1179 return; 1180 1181 GlobalVariable *GV = cast<GlobalVariable>(SI.getOperand(1)); 1182 DenseMap<GlobalVariable*, LatticeVal>::iterator I = TrackedGlobals.find(GV); 1183 if (I == TrackedGlobals.end() || I->second.isOverdefined()) return; 1184 1185 // Get the value we are storing into the global, then merge it. 1186 mergeInValue(I->second, GV, getValueState(SI.getOperand(0))); 1187 if (I->second.isOverdefined()) 1188 TrackedGlobals.erase(I); // No need to keep tracking this! 1189 } 1190 1191 1192 // Handle load instructions. If the operand is a constant pointer to a constant 1193 // global, we can replace the load with the loaded constant value! 1194 void SCCPSolver::visitLoadInst(LoadInst &I) { 1195 // If this load is of a struct, just mark the result overdefined. 1196 if (I.getType()->isStructTy()) 1197 return markAnythingOverdefined(&I); 1198 1199 LatticeVal PtrVal = getValueState(I.getOperand(0)); 1200 if (PtrVal.isUndefined()) return; // The pointer is not resolved yet! 1201 1202 LatticeVal &IV = ValueState[&I]; 1203 if (IV.isOverdefined()) return; 1204 1205 if (!PtrVal.isConstant() || I.isVolatile()) 1206 return markOverdefined(IV, &I); 1207 1208 Constant *Ptr = PtrVal.getConstant(); 1209 1210 // load null -> null 1211 if (isa<ConstantPointerNull>(Ptr) && I.getPointerAddressSpace() == 0) 1212 return markConstant(IV, &I, Constant::getNullValue(I.getType())); 1213 1214 // Transform load (constant global) into the value loaded. 1215 if (GlobalVariable *GV = dyn_cast<GlobalVariable>(Ptr)) { 1216 if (!TrackedGlobals.empty()) { 1217 // If we are tracking this global, merge in the known value for it. 1218 DenseMap<GlobalVariable*, LatticeVal>::iterator It = 1219 TrackedGlobals.find(GV); 1220 if (It != TrackedGlobals.end()) { 1221 mergeInValue(IV, &I, It->second); 1222 return; 1223 } 1224 } 1225 } 1226 1227 // Transform load from a constant into a constant if possible. 1228 if (Constant *C = ConstantFoldLoadFromConstPtr(Ptr, TD)) 1229 return markConstant(IV, &I, C); 1230 1231 // Otherwise we cannot say for certain what value this load will produce. 1232 // Bail out. 1233 markOverdefined(IV, &I); 1234 } 1235 1236 void SCCPSolver::visitCallSite(CallSite CS) { 1237 Function *F = CS.getCalledFunction(); 1238 Instruction *I = CS.getInstruction(); 1239 1240 // The common case is that we aren't tracking the callee, either because we 1241 // are not doing interprocedural analysis or the callee is indirect, or is 1242 // external. Handle these cases first. 1243 if (F == 0 || F->isDeclaration()) { 1244 CallOverdefined: 1245 // Void return and not tracking callee, just bail. 1246 if (I->getType()->isVoidTy()) return; 1247 1248 // Otherwise, if we have a single return value case, and if the function is 1249 // a declaration, maybe we can constant fold it. 1250 if (F && F->isDeclaration() && !I->getType()->isStructTy() && 1251 canConstantFoldCallTo(F)) { 1252 1253 SmallVector<Constant*, 8> Operands; 1254 for (CallSite::arg_iterator AI = CS.arg_begin(), E = CS.arg_end(); 1255 AI != E; ++AI) { 1256 LatticeVal State = getValueState(*AI); 1257 1258 if (State.isUndefined()) 1259 return; // Operands are not resolved yet. 1260 if (State.isOverdefined()) 1261 return markOverdefined(I); 1262 assert(State.isConstant() && "Unknown state!"); 1263 Operands.push_back(State.getConstant()); 1264 } 1265 1266 // If we can constant fold this, mark the result of the call as a 1267 // constant. 1268 if (Constant *C = ConstantFoldCall(F, Operands.data(), Operands.size())) 1269 return markConstant(I, C); 1270 } 1271 1272 // Otherwise, we don't know anything about this call, mark it overdefined. 1273 return markAnythingOverdefined(I); 1274 } 1275 1276 // If this is a local function that doesn't have its address taken, mark its 1277 // entry block executable and merge in the actual arguments to the call into 1278 // the formal arguments of the function. 1279 if (!TrackingIncomingArguments.empty() && TrackingIncomingArguments.count(F)){ 1280 MarkBlockExecutable(F->begin()); 1281 1282 // Propagate information from this call site into the callee. 1283 CallSite::arg_iterator CAI = CS.arg_begin(); 1284 for (Function::arg_iterator AI = F->arg_begin(), E = F->arg_end(); 1285 AI != E; ++AI, ++CAI) { 1286 // If this argument is byval, and if the function is not readonly, there 1287 // will be an implicit copy formed of the input aggregate. 1288 if (AI->hasByValAttr() && !F->onlyReadsMemory()) { 1289 markOverdefined(AI); 1290 continue; 1291 } 1292 1293 if (const StructType *STy = dyn_cast<StructType>(AI->getType())) { 1294 for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) { 1295 LatticeVal CallArg = getStructValueState(*CAI, i); 1296 mergeInValue(getStructValueState(AI, i), AI, CallArg); 1297 } 1298 } else { 1299 mergeInValue(AI, getValueState(*CAI)); 1300 } 1301 } 1302 } 1303 1304 // If this is a single/zero retval case, see if we're tracking the function. 1305 if (const StructType *STy = dyn_cast<StructType>(F->getReturnType())) { 1306 if (!MRVFunctionsTracked.count(F)) 1307 goto CallOverdefined; // Not tracking this callee. 1308 1309 // If we are tracking this callee, propagate the result of the function 1310 // into this call site. 1311 for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) 1312 mergeInValue(getStructValueState(I, i), I, 1313 TrackedMultipleRetVals[std::make_pair(F, i)]); 1314 } else { 1315 DenseMap<Function*, LatticeVal>::iterator TFRVI = TrackedRetVals.find(F); 1316 if (TFRVI == TrackedRetVals.end()) 1317 goto CallOverdefined; // Not tracking this callee. 1318 1319 // If so, propagate the return value of the callee into this call result. 1320 mergeInValue(I, TFRVI->second); 1321 } 1322 } 1323 1324 void SCCPSolver::Solve() { 1325 // Process the work lists until they are empty! 1326 while (!BBWorkList.empty() || !InstWorkList.empty() || 1327 !OverdefinedInstWorkList.empty()) { 1328 // Process the overdefined instruction's work list first, which drives other 1329 // things to overdefined more quickly. 1330 while (!OverdefinedInstWorkList.empty()) { 1331 Value *I = OverdefinedInstWorkList.pop_back_val(); 1332 1333 DEBUG(dbgs() << "\nPopped off OI-WL: " << *I << '\n'); 1334 1335 // "I" got into the work list because it either made the transition from 1336 // bottom to constant 1337 // 1338 // Anything on this worklist that is overdefined need not be visited 1339 // since all of its users will have already been marked as overdefined 1340 // Update all of the users of this instruction's value. 1341 // 1342 for (Value::use_iterator UI = I->use_begin(), E = I->use_end(); 1343 UI != E; ++UI) 1344 if (Instruction *I = dyn_cast<Instruction>(*UI)) 1345 OperandChangedState(I); 1346 } 1347 1348 // Process the instruction work list. 1349 while (!InstWorkList.empty()) { 1350 Value *I = InstWorkList.pop_back_val(); 1351 1352 DEBUG(dbgs() << "\nPopped off I-WL: " << *I << '\n'); 1353 1354 // "I" got into the work list because it made the transition from undef to 1355 // constant. 1356 // 1357 // Anything on this worklist that is overdefined need not be visited 1358 // since all of its users will have already been marked as overdefined. 1359 // Update all of the users of this instruction's value. 1360 // 1361 if (I->getType()->isStructTy() || !getValueState(I).isOverdefined()) 1362 for (Value::use_iterator UI = I->use_begin(), E = I->use_end(); 1363 UI != E; ++UI) 1364 if (Instruction *I = dyn_cast<Instruction>(*UI)) 1365 OperandChangedState(I); 1366 } 1367 1368 // Process the basic block work list. 1369 while (!BBWorkList.empty()) { 1370 BasicBlock *BB = BBWorkList.back(); 1371 BBWorkList.pop_back(); 1372 1373 DEBUG(dbgs() << "\nPopped off BBWL: " << *BB << '\n'); 1374 1375 // Notify all instructions in this basic block that they are newly 1376 // executable. 1377 visit(BB); 1378 } 1379 } 1380 } 1381 1382 /// ResolvedUndefsIn - While solving the dataflow for a function, we assume 1383 /// that branches on undef values cannot reach any of their successors. 1384 /// However, this is not a safe assumption. After we solve dataflow, this 1385 /// method should be use to handle this. If this returns true, the solver 1386 /// should be rerun. 1387 /// 1388 /// This method handles this by finding an unresolved branch and marking it one 1389 /// of the edges from the block as being feasible, even though the condition 1390 /// doesn't say it would otherwise be. This allows SCCP to find the rest of the 1391 /// CFG and only slightly pessimizes the analysis results (by marking one, 1392 /// potentially infeasible, edge feasible). This cannot usefully modify the 1393 /// constraints on the condition of the branch, as that would impact other users 1394 /// of the value. 1395 /// 1396 /// This scan also checks for values that use undefs, whose results are actually 1397 /// defined. For example, 'zext i8 undef to i32' should produce all zeros 1398 /// conservatively, as "(zext i8 X -> i32) & 0xFF00" must always return zero, 1399 /// even if X isn't defined. 1400 bool SCCPSolver::ResolvedUndefsIn(Function &F) { 1401 for (Function::iterator BB = F.begin(), E = F.end(); BB != E; ++BB) { 1402 if (!BBExecutable.count(BB)) 1403 continue; 1404 1405 for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ++I) { 1406 // Look for instructions which produce undef values. 1407 if (I->getType()->isVoidTy()) continue; 1408 1409 if (const StructType *STy = dyn_cast<StructType>(I->getType())) { 1410 // Only a few things that can be structs matter for undef. Just send 1411 // all their results to overdefined. We could be more precise than this 1412 // but it isn't worth bothering. 1413 if (isa<CallInst>(I) || isa<SelectInst>(I)) { 1414 for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) { 1415 LatticeVal &LV = getStructValueState(I, i); 1416 if (LV.isUndefined()) 1417 markOverdefined(LV, I); 1418 } 1419 } 1420 continue; 1421 } 1422 1423 LatticeVal &LV = getValueState(I); 1424 if (!LV.isUndefined()) continue; 1425 1426 // No instructions using structs need disambiguation. 1427 if (I->getOperand(0)->getType()->isStructTy()) 1428 continue; 1429 1430 // Get the lattice values of the first two operands for use below. 1431 LatticeVal Op0LV = getValueState(I->getOperand(0)); 1432 LatticeVal Op1LV; 1433 if (I->getNumOperands() == 2) { 1434 // No instructions using structs need disambiguation. 1435 if (I->getOperand(1)->getType()->isStructTy()) 1436 continue; 1437 1438 // If this is a two-operand instruction, and if both operands are 1439 // undefs, the result stays undef. 1440 Op1LV = getValueState(I->getOperand(1)); 1441 if (Op0LV.isUndefined() && Op1LV.isUndefined()) 1442 continue; 1443 } 1444 1445 // If this is an instructions whose result is defined even if the input is 1446 // not fully defined, propagate the information. 1447 const Type *ITy = I->getType(); 1448 switch (I->getOpcode()) { 1449 default: break; // Leave the instruction as an undef. 1450 case Instruction::ZExt: 1451 // After a zero extend, we know the top part is zero. SExt doesn't have 1452 // to be handled here, because we don't know whether the top part is 1's 1453 // or 0's. 1454 case Instruction::SIToFP: // some FP values are not possible, just use 0. 1455 case Instruction::UIToFP: // some FP values are not possible, just use 0. 1456 markForcedConstant(I, Constant::getNullValue(ITy)); 1457 return true; 1458 case Instruction::Mul: 1459 case Instruction::And: 1460 // undef * X -> 0. X could be zero. 1461 // undef & X -> 0. X could be zero. 1462 markForcedConstant(I, Constant::getNullValue(ITy)); 1463 return true; 1464 1465 case Instruction::Or: 1466 // undef | X -> -1. X could be -1. 1467 markForcedConstant(I, Constant::getAllOnesValue(ITy)); 1468 return true; 1469 1470 case Instruction::SDiv: 1471 case Instruction::UDiv: 1472 case Instruction::SRem: 1473 case Instruction::URem: 1474 // X / undef -> undef. No change. 1475 // X % undef -> undef. No change. 1476 if (Op1LV.isUndefined()) break; 1477 1478 // undef / X -> 0. X could be maxint. 1479 // undef % X -> 0. X could be 1. 1480 markForcedConstant(I, Constant::getNullValue(ITy)); 1481 return true; 1482 1483 case Instruction::AShr: 1484 // undef >>s X -> undef. No change. 1485 if (Op0LV.isUndefined()) break; 1486 1487 // X >>s undef -> X. X could be 0, X could have the high-bit known set. 1488 if (Op0LV.isConstant()) 1489 markForcedConstant(I, Op0LV.getConstant()); 1490 else 1491 markOverdefined(I); 1492 return true; 1493 case Instruction::LShr: 1494 case Instruction::Shl: 1495 // undef >> X -> undef. No change. 1496 // undef << X -> undef. No change. 1497 if (Op0LV.isUndefined()) break; 1498 1499 // X >> undef -> 0. X could be 0. 1500 // X << undef -> 0. X could be 0. 1501 markForcedConstant(I, Constant::getNullValue(ITy)); 1502 return true; 1503 case Instruction::Select: 1504 // undef ? X : Y -> X or Y. There could be commonality between X/Y. 1505 if (Op0LV.isUndefined()) { 1506 if (!Op1LV.isConstant()) // Pick the constant one if there is any. 1507 Op1LV = getValueState(I->getOperand(2)); 1508 } else if (Op1LV.isUndefined()) { 1509 // c ? undef : undef -> undef. No change. 1510 Op1LV = getValueState(I->getOperand(2)); 1511 if (Op1LV.isUndefined()) 1512 break; 1513 // Otherwise, c ? undef : x -> x. 1514 } else { 1515 // Leave Op1LV as Operand(1)'s LatticeValue. 1516 } 1517 1518 if (Op1LV.isConstant()) 1519 markForcedConstant(I, Op1LV.getConstant()); 1520 else 1521 markOverdefined(I); 1522 return true; 1523 case Instruction::Call: 1524 // If a call has an undef result, it is because it is constant foldable 1525 // but one of the inputs was undef. Just force the result to 1526 // overdefined. 1527 markOverdefined(I); 1528 return true; 1529 } 1530 } 1531 1532 // Check to see if we have a branch or switch on an undefined value. If so 1533 // we force the branch to go one way or the other to make the successor 1534 // values live. It doesn't really matter which way we force it. 1535 TerminatorInst *TI = BB->getTerminator(); 1536 if (BranchInst *BI = dyn_cast<BranchInst>(TI)) { 1537 if (!BI->isConditional()) continue; 1538 if (!getValueState(BI->getCondition()).isUndefined()) 1539 continue; 1540 1541 // If the input to SCCP is actually branch on undef, fix the undef to 1542 // false. 1543 if (isa<UndefValue>(BI->getCondition())) { 1544 BI->setCondition(ConstantInt::getFalse(BI->getContext())); 1545 markEdgeExecutable(BB, TI->getSuccessor(1)); 1546 return true; 1547 } 1548 1549 // Otherwise, it is a branch on a symbolic value which is currently 1550 // considered to be undef. Handle this by forcing the input value to the 1551 // branch to false. 1552 markForcedConstant(BI->getCondition(), 1553 ConstantInt::getFalse(TI->getContext())); 1554 return true; 1555 } 1556 1557 if (SwitchInst *SI = dyn_cast<SwitchInst>(TI)) { 1558 if (SI->getNumSuccessors() < 2) // no cases 1559 continue; 1560 if (!getValueState(SI->getCondition()).isUndefined()) 1561 continue; 1562 1563 // If the input to SCCP is actually switch on undef, fix the undef to 1564 // the first constant. 1565 if (isa<UndefValue>(SI->getCondition())) { 1566 SI->setCondition(SI->getCaseValue(1)); 1567 markEdgeExecutable(BB, TI->getSuccessor(1)); 1568 return true; 1569 } 1570 1571 markForcedConstant(SI->getCondition(), SI->getCaseValue(1)); 1572 return true; 1573 } 1574 } 1575 1576 return false; 1577 } 1578 1579 1580 namespace { 1581 //===--------------------------------------------------------------------===// 1582 // 1583 /// SCCP Class - This class uses the SCCPSolver to implement a per-function 1584 /// Sparse Conditional Constant Propagator. 1585 /// 1586 struct SCCP : public FunctionPass { 1587 static char ID; // Pass identification, replacement for typeid 1588 SCCP() : FunctionPass(ID) {} 1589 1590 // runOnFunction - Run the Sparse Conditional Constant Propagation 1591 // algorithm, and return true if the function was modified. 1592 // 1593 bool runOnFunction(Function &F); 1594 1595 virtual void getAnalysisUsage(AnalysisUsage &AU) const { 1596 AU.setPreservesCFG(); 1597 } 1598 }; 1599 } // end anonymous namespace 1600 1601 char SCCP::ID = 0; 1602 INITIALIZE_PASS(SCCP, "sccp", 1603 "Sparse Conditional Constant Propagation", false, false); 1604 1605 // createSCCPPass - This is the public interface to this file. 1606 FunctionPass *llvm::createSCCPPass() { 1607 return new SCCP(); 1608 } 1609 1610 static void DeleteInstructionInBlock(BasicBlock *BB) { 1611 DEBUG(dbgs() << " BasicBlock Dead:" << *BB); 1612 ++NumDeadBlocks; 1613 1614 // Delete the instructions backwards, as it has a reduced likelihood of 1615 // having to update as many def-use and use-def chains. 1616 while (!isa<TerminatorInst>(BB->begin())) { 1617 Instruction *I = --BasicBlock::iterator(BB->getTerminator()); 1618 1619 if (!I->use_empty()) 1620 I->replaceAllUsesWith(UndefValue::get(I->getType())); 1621 BB->getInstList().erase(I); 1622 ++NumInstRemoved; 1623 } 1624 } 1625 1626 // runOnFunction() - Run the Sparse Conditional Constant Propagation algorithm, 1627 // and return true if the function was modified. 1628 // 1629 bool SCCP::runOnFunction(Function &F) { 1630 DEBUG(dbgs() << "SCCP on function '" << F.getName() << "'\n"); 1631 SCCPSolver Solver(getAnalysisIfAvailable<TargetData>()); 1632 1633 // Mark the first block of the function as being executable. 1634 Solver.MarkBlockExecutable(F.begin()); 1635 1636 // Mark all arguments to the function as being overdefined. 1637 for (Function::arg_iterator AI = F.arg_begin(), E = F.arg_end(); AI != E;++AI) 1638 Solver.markAnythingOverdefined(AI); 1639 1640 // Solve for constants. 1641 bool ResolvedUndefs = true; 1642 while (ResolvedUndefs) { 1643 Solver.Solve(); 1644 DEBUG(dbgs() << "RESOLVING UNDEFs\n"); 1645 ResolvedUndefs = Solver.ResolvedUndefsIn(F); 1646 } 1647 1648 bool MadeChanges = false; 1649 1650 // If we decided that there are basic blocks that are dead in this function, 1651 // delete their contents now. Note that we cannot actually delete the blocks, 1652 // as we cannot modify the CFG of the function. 1653 1654 for (Function::iterator BB = F.begin(), E = F.end(); BB != E; ++BB) { 1655 if (!Solver.isBlockExecutable(BB)) { 1656 DeleteInstructionInBlock(BB); 1657 MadeChanges = true; 1658 continue; 1659 } 1660 1661 // Iterate over all of the instructions in a function, replacing them with 1662 // constants if we have found them to be of constant values. 1663 // 1664 for (BasicBlock::iterator BI = BB->begin(), E = BB->end(); BI != E; ) { 1665 Instruction *Inst = BI++; 1666 if (Inst->getType()->isVoidTy() || isa<TerminatorInst>(Inst)) 1667 continue; 1668 1669 // TODO: Reconstruct structs from their elements. 1670 if (Inst->getType()->isStructTy()) 1671 continue; 1672 1673 LatticeVal IV = Solver.getLatticeValueFor(Inst); 1674 if (IV.isOverdefined()) 1675 continue; 1676 1677 Constant *Const = IV.isConstant() 1678 ? IV.getConstant() : UndefValue::get(Inst->getType()); 1679 DEBUG(dbgs() << " Constant: " << *Const << " = " << *Inst); 1680 1681 // Replaces all of the uses of a variable with uses of the constant. 1682 Inst->replaceAllUsesWith(Const); 1683 1684 // Delete the instruction. 1685 Inst->eraseFromParent(); 1686 1687 // Hey, we just changed something! 1688 MadeChanges = true; 1689 ++NumInstRemoved; 1690 } 1691 } 1692 1693 return MadeChanges; 1694 } 1695 1696 namespace { 1697 //===--------------------------------------------------------------------===// 1698 // 1699 /// IPSCCP Class - This class implements interprocedural Sparse Conditional 1700 /// Constant Propagation. 1701 /// 1702 struct IPSCCP : public ModulePass { 1703 static char ID; 1704 IPSCCP() : ModulePass(ID) {} 1705 bool runOnModule(Module &M); 1706 }; 1707 } // end anonymous namespace 1708 1709 char IPSCCP::ID = 0; 1710 INITIALIZE_PASS(IPSCCP, "ipsccp", 1711 "Interprocedural Sparse Conditional Constant Propagation", 1712 false, false); 1713 1714 // createIPSCCPPass - This is the public interface to this file. 1715 ModulePass *llvm::createIPSCCPPass() { 1716 return new IPSCCP(); 1717 } 1718 1719 1720 static bool AddressIsTaken(const GlobalValue *GV) { 1721 // Delete any dead constantexpr klingons. 1722 GV->removeDeadConstantUsers(); 1723 1724 for (Value::const_use_iterator UI = GV->use_begin(), E = GV->use_end(); 1725 UI != E; ++UI) { 1726 const User *U = *UI; 1727 if (const StoreInst *SI = dyn_cast<StoreInst>(U)) { 1728 if (SI->getOperand(0) == GV || SI->isVolatile()) 1729 return true; // Storing addr of GV. 1730 } else if (isa<InvokeInst>(U) || isa<CallInst>(U)) { 1731 // Make sure we are calling the function, not passing the address. 1732 ImmutableCallSite CS(cast<Instruction>(U)); 1733 if (!CS.isCallee(UI)) 1734 return true; 1735 } else if (const LoadInst *LI = dyn_cast<LoadInst>(U)) { 1736 if (LI->isVolatile()) 1737 return true; 1738 } else if (isa<BlockAddress>(U)) { 1739 // blockaddress doesn't take the address of the function, it takes addr 1740 // of label. 1741 } else { 1742 return true; 1743 } 1744 } 1745 return false; 1746 } 1747 1748 bool IPSCCP::runOnModule(Module &M) { 1749 SCCPSolver Solver(getAnalysisIfAvailable<TargetData>()); 1750 1751 // AddressTakenFunctions - This set keeps track of the address-taken functions 1752 // that are in the input. As IPSCCP runs through and simplifies code, 1753 // functions that were address taken can end up losing their 1754 // address-taken-ness. Because of this, we keep track of their addresses from 1755 // the first pass so we can use them for the later simplification pass. 1756 SmallPtrSet<Function*, 32> AddressTakenFunctions; 1757 1758 // Loop over all functions, marking arguments to those with their addresses 1759 // taken or that are external as overdefined. 1760 // 1761 for (Module::iterator F = M.begin(), E = M.end(); F != E; ++F) { 1762 if (F->isDeclaration()) 1763 continue; 1764 1765 // If this is a strong or ODR definition of this function, then we can 1766 // propagate information about its result into callsites of it. 1767 if (!F->mayBeOverridden()) 1768 Solver.AddTrackedFunction(F); 1769 1770 // If this function only has direct calls that we can see, we can track its 1771 // arguments and return value aggressively, and can assume it is not called 1772 // unless we see evidence to the contrary. 1773 if (F->hasLocalLinkage()) { 1774 if (AddressIsTaken(F)) 1775 AddressTakenFunctions.insert(F); 1776 else { 1777 Solver.AddArgumentTrackedFunction(F); 1778 continue; 1779 } 1780 } 1781 1782 // Assume the function is called. 1783 Solver.MarkBlockExecutable(F->begin()); 1784 1785 // Assume nothing about the incoming arguments. 1786 for (Function::arg_iterator AI = F->arg_begin(), E = F->arg_end(); 1787 AI != E; ++AI) 1788 Solver.markAnythingOverdefined(AI); 1789 } 1790 1791 // Loop over global variables. We inform the solver about any internal global 1792 // variables that do not have their 'addresses taken'. If they don't have 1793 // their addresses taken, we can propagate constants through them. 1794 for (Module::global_iterator G = M.global_begin(), E = M.global_end(); 1795 G != E; ++G) 1796 if (!G->isConstant() && G->hasLocalLinkage() && !AddressIsTaken(G)) 1797 Solver.TrackValueOfGlobalVariable(G); 1798 1799 // Solve for constants. 1800 bool ResolvedUndefs = true; 1801 while (ResolvedUndefs) { 1802 Solver.Solve(); 1803 1804 DEBUG(dbgs() << "RESOLVING UNDEFS\n"); 1805 ResolvedUndefs = false; 1806 for (Module::iterator F = M.begin(), E = M.end(); F != E; ++F) 1807 ResolvedUndefs |= Solver.ResolvedUndefsIn(*F); 1808 } 1809 1810 bool MadeChanges = false; 1811 1812 // Iterate over all of the instructions in the module, replacing them with 1813 // constants if we have found them to be of constant values. 1814 // 1815 SmallVector<BasicBlock*, 512> BlocksToErase; 1816 1817 for (Module::iterator F = M.begin(), E = M.end(); F != E; ++F) { 1818 if (Solver.isBlockExecutable(F->begin())) { 1819 for (Function::arg_iterator AI = F->arg_begin(), E = F->arg_end(); 1820 AI != E; ++AI) { 1821 if (AI->use_empty() || AI->getType()->isStructTy()) continue; 1822 1823 // TODO: Could use getStructLatticeValueFor to find out if the entire 1824 // result is a constant and replace it entirely if so. 1825 1826 LatticeVal IV = Solver.getLatticeValueFor(AI); 1827 if (IV.isOverdefined()) continue; 1828 1829 Constant *CST = IV.isConstant() ? 1830 IV.getConstant() : UndefValue::get(AI->getType()); 1831 DEBUG(dbgs() << "*** Arg " << *AI << " = " << *CST <<"\n"); 1832 1833 // Replaces all of the uses of a variable with uses of the 1834 // constant. 1835 AI->replaceAllUsesWith(CST); 1836 ++IPNumArgsElimed; 1837 } 1838 } 1839 1840 for (Function::iterator BB = F->begin(), E = F->end(); BB != E; ++BB) { 1841 if (!Solver.isBlockExecutable(BB)) { 1842 DeleteInstructionInBlock(BB); 1843 MadeChanges = true; 1844 1845 TerminatorInst *TI = BB->getTerminator(); 1846 for (unsigned i = 0, e = TI->getNumSuccessors(); i != e; ++i) { 1847 BasicBlock *Succ = TI->getSuccessor(i); 1848 if (!Succ->empty() && isa<PHINode>(Succ->begin())) 1849 TI->getSuccessor(i)->removePredecessor(BB); 1850 } 1851 if (!TI->use_empty()) 1852 TI->replaceAllUsesWith(UndefValue::get(TI->getType())); 1853 TI->eraseFromParent(); 1854 1855 if (&*BB != &F->front()) 1856 BlocksToErase.push_back(BB); 1857 else 1858 new UnreachableInst(M.getContext(), BB); 1859 continue; 1860 } 1861 1862 for (BasicBlock::iterator BI = BB->begin(), E = BB->end(); BI != E; ) { 1863 Instruction *Inst = BI++; 1864 if (Inst->getType()->isVoidTy() || Inst->getType()->isStructTy()) 1865 continue; 1866 1867 // TODO: Could use getStructLatticeValueFor to find out if the entire 1868 // result is a constant and replace it entirely if so. 1869 1870 LatticeVal IV = Solver.getLatticeValueFor(Inst); 1871 if (IV.isOverdefined()) 1872 continue; 1873 1874 Constant *Const = IV.isConstant() 1875 ? IV.getConstant() : UndefValue::get(Inst->getType()); 1876 DEBUG(dbgs() << " Constant: " << *Const << " = " << *Inst); 1877 1878 // Replaces all of the uses of a variable with uses of the 1879 // constant. 1880 Inst->replaceAllUsesWith(Const); 1881 1882 // Delete the instruction. 1883 if (!isa<CallInst>(Inst) && !isa<TerminatorInst>(Inst)) 1884 Inst->eraseFromParent(); 1885 1886 // Hey, we just changed something! 1887 MadeChanges = true; 1888 ++IPNumInstRemoved; 1889 } 1890 } 1891 1892 // Now that all instructions in the function are constant folded, erase dead 1893 // blocks, because we can now use ConstantFoldTerminator to get rid of 1894 // in-edges. 1895 for (unsigned i = 0, e = BlocksToErase.size(); i != e; ++i) { 1896 // If there are any PHI nodes in this successor, drop entries for BB now. 1897 BasicBlock *DeadBB = BlocksToErase[i]; 1898 for (Value::use_iterator UI = DeadBB->use_begin(), UE = DeadBB->use_end(); 1899 UI != UE; ) { 1900 // Grab the user and then increment the iterator early, as the user 1901 // will be deleted. Step past all adjacent uses from the same user. 1902 Instruction *I = dyn_cast<Instruction>(*UI); 1903 do { ++UI; } while (UI != UE && *UI == I); 1904 1905 // Ignore blockaddress users; BasicBlock's dtor will handle them. 1906 if (!I) continue; 1907 1908 bool Folded = ConstantFoldTerminator(I->getParent()); 1909 if (!Folded) { 1910 // The constant folder may not have been able to fold the terminator 1911 // if this is a branch or switch on undef. Fold it manually as a 1912 // branch to the first successor. 1913 #ifndef NDEBUG 1914 if (BranchInst *BI = dyn_cast<BranchInst>(I)) { 1915 assert(BI->isConditional() && isa<UndefValue>(BI->getCondition()) && 1916 "Branch should be foldable!"); 1917 } else if (SwitchInst *SI = dyn_cast<SwitchInst>(I)) { 1918 assert(isa<UndefValue>(SI->getCondition()) && "Switch should fold"); 1919 } else { 1920 llvm_unreachable("Didn't fold away reference to block!"); 1921 } 1922 #endif 1923 1924 // Make this an uncond branch to the first successor. 1925 TerminatorInst *TI = I->getParent()->getTerminator(); 1926 BranchInst::Create(TI->getSuccessor(0), TI); 1927 1928 // Remove entries in successor phi nodes to remove edges. 1929 for (unsigned i = 1, e = TI->getNumSuccessors(); i != e; ++i) 1930 TI->getSuccessor(i)->removePredecessor(TI->getParent()); 1931 1932 // Remove the old terminator. 1933 TI->eraseFromParent(); 1934 } 1935 } 1936 1937 // Finally, delete the basic block. 1938 F->getBasicBlockList().erase(DeadBB); 1939 } 1940 BlocksToErase.clear(); 1941 } 1942 1943 // If we inferred constant or undef return values for a function, we replaced 1944 // all call uses with the inferred value. This means we don't need to bother 1945 // actually returning anything from the function. Replace all return 1946 // instructions with return undef. 1947 // 1948 // Do this in two stages: first identify the functions we should process, then 1949 // actually zap their returns. This is important because we can only do this 1950 // if the address of the function isn't taken. In cases where a return is the 1951 // last use of a function, the order of processing functions would affect 1952 // whether other functions are optimizable. 1953 SmallVector<ReturnInst*, 8> ReturnsToZap; 1954 1955 // TODO: Process multiple value ret instructions also. 1956 const DenseMap<Function*, LatticeVal> &RV = Solver.getTrackedRetVals(); 1957 for (DenseMap<Function*, LatticeVal>::const_iterator I = RV.begin(), 1958 E = RV.end(); I != E; ++I) { 1959 Function *F = I->first; 1960 if (I->second.isOverdefined() || F->getReturnType()->isVoidTy()) 1961 continue; 1962 1963 // We can only do this if we know that nothing else can call the function. 1964 if (!F->hasLocalLinkage() || AddressTakenFunctions.count(F)) 1965 continue; 1966 1967 for (Function::iterator BB = F->begin(), E = F->end(); BB != E; ++BB) 1968 if (ReturnInst *RI = dyn_cast<ReturnInst>(BB->getTerminator())) 1969 if (!isa<UndefValue>(RI->getOperand(0))) 1970 ReturnsToZap.push_back(RI); 1971 } 1972 1973 // Zap all returns which we've identified as zap to change. 1974 for (unsigned i = 0, e = ReturnsToZap.size(); i != e; ++i) { 1975 Function *F = ReturnsToZap[i]->getParent()->getParent(); 1976 ReturnsToZap[i]->setOperand(0, UndefValue::get(F->getReturnType())); 1977 } 1978 1979 // If we infered constant or undef values for globals variables, we can delete 1980 // the global and any stores that remain to it. 1981 const DenseMap<GlobalVariable*, LatticeVal> &TG = Solver.getTrackedGlobals(); 1982 for (DenseMap<GlobalVariable*, LatticeVal>::const_iterator I = TG.begin(), 1983 E = TG.end(); I != E; ++I) { 1984 GlobalVariable *GV = I->first; 1985 assert(!I->second.isOverdefined() && 1986 "Overdefined values should have been taken out of the map!"); 1987 DEBUG(dbgs() << "Found that GV '" << GV->getName() << "' is constant!\n"); 1988 while (!GV->use_empty()) { 1989 StoreInst *SI = cast<StoreInst>(GV->use_back()); 1990 SI->eraseFromParent(); 1991 } 1992 M.getGlobalList().erase(GV); 1993 ++IPNumGlobalConst; 1994 } 1995 1996 return MadeChanges; 1997 } 1998