GVN.cpp revision 3eb4f7e2dd98038f94e45d1c45ccff49e6659c87
1//===- GVN.cpp - Eliminate redundant values and loads ---------------------===// 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 pass performs global value numbering to eliminate fully redundant 11// instructions. It also performs simple dead load elimination. 12// 13// Note that this pass does the value numbering itself; it does not use the 14// ValueNumbering analysis passes. 15// 16//===----------------------------------------------------------------------===// 17 18#define DEBUG_TYPE "gvn" 19#include "llvm/Transforms/Scalar.h" 20#include "llvm/BasicBlock.h" 21#include "llvm/Constants.h" 22#include "llvm/DerivedTypes.h" 23#include "llvm/GlobalVariable.h" 24#include "llvm/Function.h" 25#include "llvm/IntrinsicInst.h" 26#include "llvm/LLVMContext.h" 27#include "llvm/Operator.h" 28#include "llvm/Value.h" 29#include "llvm/ADT/DenseMap.h" 30#include "llvm/ADT/DepthFirstIterator.h" 31#include "llvm/ADT/PostOrderIterator.h" 32#include "llvm/ADT/SmallPtrSet.h" 33#include "llvm/ADT/SmallVector.h" 34#include "llvm/ADT/Statistic.h" 35#include "llvm/Analysis/AliasAnalysis.h" 36#include "llvm/Analysis/ConstantFolding.h" 37#include "llvm/Analysis/Dominators.h" 38#include "llvm/Analysis/MemoryBuiltins.h" 39#include "llvm/Analysis/MemoryDependenceAnalysis.h" 40#include "llvm/Analysis/PHITransAddr.h" 41#include "llvm/Support/CFG.h" 42#include "llvm/Support/CommandLine.h" 43#include "llvm/Support/Debug.h" 44#include "llvm/Support/ErrorHandling.h" 45#include "llvm/Support/GetElementPtrTypeIterator.h" 46#include "llvm/Support/IRBuilder.h" 47#include "llvm/Support/raw_ostream.h" 48#include "llvm/Target/TargetData.h" 49#include "llvm/Transforms/Utils/BasicBlockUtils.h" 50#include "llvm/Transforms/Utils/Local.h" 51#include "llvm/Transforms/Utils/SSAUpdater.h" 52using namespace llvm; 53 54STATISTIC(NumGVNInstr, "Number of instructions deleted"); 55STATISTIC(NumGVNLoad, "Number of loads deleted"); 56STATISTIC(NumGVNPRE, "Number of instructions PRE'd"); 57STATISTIC(NumGVNBlocks, "Number of blocks merged"); 58STATISTIC(NumPRELoad, "Number of loads PRE'd"); 59 60static cl::opt<bool> EnablePRE("enable-pre", 61 cl::init(true), cl::Hidden); 62static cl::opt<bool> EnableLoadPRE("enable-load-pre", cl::init(true)); 63 64//===----------------------------------------------------------------------===// 65// ValueTable Class 66//===----------------------------------------------------------------------===// 67 68/// This class holds the mapping between values and value numbers. It is used 69/// as an efficient mechanism to determine the expression-wise equivalence of 70/// two values. 71namespace { 72 struct Expression { 73 enum ExpressionOpcode { 74 ADD = Instruction::Add, 75 FADD = Instruction::FAdd, 76 SUB = Instruction::Sub, 77 FSUB = Instruction::FSub, 78 MUL = Instruction::Mul, 79 FMUL = Instruction::FMul, 80 UDIV = Instruction::UDiv, 81 SDIV = Instruction::SDiv, 82 FDIV = Instruction::FDiv, 83 UREM = Instruction::URem, 84 SREM = Instruction::SRem, 85 FREM = Instruction::FRem, 86 SHL = Instruction::Shl, 87 LSHR = Instruction::LShr, 88 ASHR = Instruction::AShr, 89 AND = Instruction::And, 90 OR = Instruction::Or, 91 XOR = Instruction::Xor, 92 TRUNC = Instruction::Trunc, 93 ZEXT = Instruction::ZExt, 94 SEXT = Instruction::SExt, 95 FPTOUI = Instruction::FPToUI, 96 FPTOSI = Instruction::FPToSI, 97 UITOFP = Instruction::UIToFP, 98 SITOFP = Instruction::SIToFP, 99 FPTRUNC = Instruction::FPTrunc, 100 FPEXT = Instruction::FPExt, 101 PTRTOINT = Instruction::PtrToInt, 102 INTTOPTR = Instruction::IntToPtr, 103 BITCAST = Instruction::BitCast, 104 ICMPEQ, ICMPNE, ICMPUGT, ICMPUGE, ICMPULT, ICMPULE, 105 ICMPSGT, ICMPSGE, ICMPSLT, ICMPSLE, FCMPOEQ, 106 FCMPOGT, FCMPOGE, FCMPOLT, FCMPOLE, FCMPONE, 107 FCMPORD, FCMPUNO, FCMPUEQ, FCMPUGT, FCMPUGE, 108 FCMPULT, FCMPULE, FCMPUNE, EXTRACT, INSERT, 109 SHUFFLE, SELECT, GEP, CALL, CONSTANT, 110 INSERTVALUE, EXTRACTVALUE, EMPTY, TOMBSTONE }; 111 112 ExpressionOpcode opcode; 113 const Type* type; 114 SmallVector<uint32_t, 4> varargs; 115 Value *function; 116 117 Expression() { } 118 Expression(ExpressionOpcode o) : opcode(o) { } 119 120 bool operator==(const Expression &other) const { 121 if (opcode != other.opcode) 122 return false; 123 else if (opcode == EMPTY || opcode == TOMBSTONE) 124 return true; 125 else if (type != other.type) 126 return false; 127 else if (function != other.function) 128 return false; 129 else { 130 if (varargs.size() != other.varargs.size()) 131 return false; 132 133 for (size_t i = 0; i < varargs.size(); ++i) 134 if (varargs[i] != other.varargs[i]) 135 return false; 136 137 return true; 138 } 139 } 140 141 bool operator!=(const Expression &other) const { 142 return !(*this == other); 143 } 144 }; 145 146 class ValueTable { 147 private: 148 DenseMap<Value*, uint32_t> valueNumbering; 149 DenseMap<Expression, uint32_t> expressionNumbering; 150 AliasAnalysis* AA; 151 MemoryDependenceAnalysis* MD; 152 DominatorTree* DT; 153 154 uint32_t nextValueNumber; 155 156 Expression::ExpressionOpcode getOpcode(CmpInst* C); 157 Expression create_expression(BinaryOperator* BO); 158 Expression create_expression(CmpInst* C); 159 Expression create_expression(ShuffleVectorInst* V); 160 Expression create_expression(ExtractElementInst* C); 161 Expression create_expression(InsertElementInst* V); 162 Expression create_expression(SelectInst* V); 163 Expression create_expression(CastInst* C); 164 Expression create_expression(GetElementPtrInst* G); 165 Expression create_expression(CallInst* C); 166 Expression create_expression(Constant* C); 167 Expression create_expression(ExtractValueInst* C); 168 Expression create_expression(InsertValueInst* C); 169 170 uint32_t lookup_or_add_call(CallInst* C); 171 public: 172 ValueTable() : nextValueNumber(1) { } 173 uint32_t lookup_or_add(Value *V); 174 uint32_t lookup(Value *V) const; 175 void add(Value *V, uint32_t num); 176 void clear(); 177 void erase(Value *v); 178 unsigned size(); 179 void setAliasAnalysis(AliasAnalysis* A) { AA = A; } 180 AliasAnalysis *getAliasAnalysis() const { return AA; } 181 void setMemDep(MemoryDependenceAnalysis* M) { MD = M; } 182 void setDomTree(DominatorTree* D) { DT = D; } 183 uint32_t getNextUnusedValueNumber() { return nextValueNumber; } 184 void verifyRemoved(const Value *) const; 185 }; 186} 187 188namespace llvm { 189template <> struct DenseMapInfo<Expression> { 190 static inline Expression getEmptyKey() { 191 return Expression(Expression::EMPTY); 192 } 193 194 static inline Expression getTombstoneKey() { 195 return Expression(Expression::TOMBSTONE); 196 } 197 198 static unsigned getHashValue(const Expression e) { 199 unsigned hash = e.opcode; 200 201 hash = ((unsigned)((uintptr_t)e.type >> 4) ^ 202 (unsigned)((uintptr_t)e.type >> 9)); 203 204 for (SmallVector<uint32_t, 4>::const_iterator I = e.varargs.begin(), 205 E = e.varargs.end(); I != E; ++I) 206 hash = *I + hash * 37; 207 208 hash = ((unsigned)((uintptr_t)e.function >> 4) ^ 209 (unsigned)((uintptr_t)e.function >> 9)) + 210 hash * 37; 211 212 return hash; 213 } 214 static bool isEqual(const Expression &LHS, const Expression &RHS) { 215 return LHS == RHS; 216 } 217}; 218 219template <> 220struct isPodLike<Expression> { static const bool value = true; }; 221 222} 223 224//===----------------------------------------------------------------------===// 225// ValueTable Internal Functions 226//===----------------------------------------------------------------------===// 227 228Expression::ExpressionOpcode ValueTable::getOpcode(CmpInst* C) { 229 if (isa<ICmpInst>(C)) { 230 switch (C->getPredicate()) { 231 default: // THIS SHOULD NEVER HAPPEN 232 llvm_unreachable("Comparison with unknown predicate?"); 233 case ICmpInst::ICMP_EQ: return Expression::ICMPEQ; 234 case ICmpInst::ICMP_NE: return Expression::ICMPNE; 235 case ICmpInst::ICMP_UGT: return Expression::ICMPUGT; 236 case ICmpInst::ICMP_UGE: return Expression::ICMPUGE; 237 case ICmpInst::ICMP_ULT: return Expression::ICMPULT; 238 case ICmpInst::ICMP_ULE: return Expression::ICMPULE; 239 case ICmpInst::ICMP_SGT: return Expression::ICMPSGT; 240 case ICmpInst::ICMP_SGE: return Expression::ICMPSGE; 241 case ICmpInst::ICMP_SLT: return Expression::ICMPSLT; 242 case ICmpInst::ICMP_SLE: return Expression::ICMPSLE; 243 } 244 } else { 245 switch (C->getPredicate()) { 246 default: // THIS SHOULD NEVER HAPPEN 247 llvm_unreachable("Comparison with unknown predicate?"); 248 case FCmpInst::FCMP_OEQ: return Expression::FCMPOEQ; 249 case FCmpInst::FCMP_OGT: return Expression::FCMPOGT; 250 case FCmpInst::FCMP_OGE: return Expression::FCMPOGE; 251 case FCmpInst::FCMP_OLT: return Expression::FCMPOLT; 252 case FCmpInst::FCMP_OLE: return Expression::FCMPOLE; 253 case FCmpInst::FCMP_ONE: return Expression::FCMPONE; 254 case FCmpInst::FCMP_ORD: return Expression::FCMPORD; 255 case FCmpInst::FCMP_UNO: return Expression::FCMPUNO; 256 case FCmpInst::FCMP_UEQ: return Expression::FCMPUEQ; 257 case FCmpInst::FCMP_UGT: return Expression::FCMPUGT; 258 case FCmpInst::FCMP_UGE: return Expression::FCMPUGE; 259 case FCmpInst::FCMP_ULT: return Expression::FCMPULT; 260 case FCmpInst::FCMP_ULE: return Expression::FCMPULE; 261 case FCmpInst::FCMP_UNE: return Expression::FCMPUNE; 262 } 263 } 264} 265 266Expression ValueTable::create_expression(CallInst* C) { 267 Expression e; 268 269 e.type = C->getType(); 270 e.function = C->getCalledFunction(); 271 e.opcode = Expression::CALL; 272 273 for (CallInst::op_iterator I = C->op_begin()+1, E = C->op_end(); 274 I != E; ++I) 275 e.varargs.push_back(lookup_or_add(*I)); 276 277 return e; 278} 279 280Expression ValueTable::create_expression(BinaryOperator* BO) { 281 Expression e; 282 e.varargs.push_back(lookup_or_add(BO->getOperand(0))); 283 e.varargs.push_back(lookup_or_add(BO->getOperand(1))); 284 e.function = 0; 285 e.type = BO->getType(); 286 e.opcode = static_cast<Expression::ExpressionOpcode>(BO->getOpcode()); 287 288 return e; 289} 290 291Expression ValueTable::create_expression(CmpInst* C) { 292 Expression e; 293 294 e.varargs.push_back(lookup_or_add(C->getOperand(0))); 295 e.varargs.push_back(lookup_or_add(C->getOperand(1))); 296 e.function = 0; 297 e.type = C->getType(); 298 e.opcode = getOpcode(C); 299 300 return e; 301} 302 303Expression ValueTable::create_expression(CastInst* C) { 304 Expression e; 305 306 e.varargs.push_back(lookup_or_add(C->getOperand(0))); 307 e.function = 0; 308 e.type = C->getType(); 309 e.opcode = static_cast<Expression::ExpressionOpcode>(C->getOpcode()); 310 311 return e; 312} 313 314Expression ValueTable::create_expression(ShuffleVectorInst* S) { 315 Expression e; 316 317 e.varargs.push_back(lookup_or_add(S->getOperand(0))); 318 e.varargs.push_back(lookup_or_add(S->getOperand(1))); 319 e.varargs.push_back(lookup_or_add(S->getOperand(2))); 320 e.function = 0; 321 e.type = S->getType(); 322 e.opcode = Expression::SHUFFLE; 323 324 return e; 325} 326 327Expression ValueTable::create_expression(ExtractElementInst* E) { 328 Expression e; 329 330 e.varargs.push_back(lookup_or_add(E->getOperand(0))); 331 e.varargs.push_back(lookup_or_add(E->getOperand(1))); 332 e.function = 0; 333 e.type = E->getType(); 334 e.opcode = Expression::EXTRACT; 335 336 return e; 337} 338 339Expression ValueTable::create_expression(InsertElementInst* I) { 340 Expression e; 341 342 e.varargs.push_back(lookup_or_add(I->getOperand(0))); 343 e.varargs.push_back(lookup_or_add(I->getOperand(1))); 344 e.varargs.push_back(lookup_or_add(I->getOperand(2))); 345 e.function = 0; 346 e.type = I->getType(); 347 e.opcode = Expression::INSERT; 348 349 return e; 350} 351 352Expression ValueTable::create_expression(SelectInst* I) { 353 Expression e; 354 355 e.varargs.push_back(lookup_or_add(I->getCondition())); 356 e.varargs.push_back(lookup_or_add(I->getTrueValue())); 357 e.varargs.push_back(lookup_or_add(I->getFalseValue())); 358 e.function = 0; 359 e.type = I->getType(); 360 e.opcode = Expression::SELECT; 361 362 return e; 363} 364 365Expression ValueTable::create_expression(GetElementPtrInst* G) { 366 Expression e; 367 368 e.varargs.push_back(lookup_or_add(G->getPointerOperand())); 369 e.function = 0; 370 e.type = G->getType(); 371 e.opcode = Expression::GEP; 372 373 for (GetElementPtrInst::op_iterator I = G->idx_begin(), E = G->idx_end(); 374 I != E; ++I) 375 e.varargs.push_back(lookup_or_add(*I)); 376 377 return e; 378} 379 380Expression ValueTable::create_expression(ExtractValueInst* E) { 381 Expression e; 382 383 e.varargs.push_back(lookup_or_add(E->getAggregateOperand())); 384 for (ExtractValueInst::idx_iterator II = E->idx_begin(), IE = E->idx_end(); 385 II != IE; ++II) 386 e.varargs.push_back(*II); 387 e.function = 0; 388 e.type = E->getType(); 389 e.opcode = Expression::EXTRACTVALUE; 390 391 return e; 392} 393 394Expression ValueTable::create_expression(InsertValueInst* E) { 395 Expression e; 396 397 e.varargs.push_back(lookup_or_add(E->getAggregateOperand())); 398 e.varargs.push_back(lookup_or_add(E->getInsertedValueOperand())); 399 for (InsertValueInst::idx_iterator II = E->idx_begin(), IE = E->idx_end(); 400 II != IE; ++II) 401 e.varargs.push_back(*II); 402 e.function = 0; 403 e.type = E->getType(); 404 e.opcode = Expression::INSERTVALUE; 405 406 return e; 407} 408 409//===----------------------------------------------------------------------===// 410// ValueTable External Functions 411//===----------------------------------------------------------------------===// 412 413/// add - Insert a value into the table with a specified value number. 414void ValueTable::add(Value *V, uint32_t num) { 415 valueNumbering.insert(std::make_pair(V, num)); 416} 417 418uint32_t ValueTable::lookup_or_add_call(CallInst* C) { 419 if (AA->doesNotAccessMemory(C)) { 420 Expression exp = create_expression(C); 421 uint32_t& e = expressionNumbering[exp]; 422 if (!e) e = nextValueNumber++; 423 valueNumbering[C] = e; 424 return e; 425 } else if (AA->onlyReadsMemory(C)) { 426 Expression exp = create_expression(C); 427 uint32_t& e = expressionNumbering[exp]; 428 if (!e) { 429 e = nextValueNumber++; 430 valueNumbering[C] = e; 431 return e; 432 } 433 if (!MD) { 434 e = nextValueNumber++; 435 valueNumbering[C] = e; 436 return e; 437 } 438 439 MemDepResult local_dep = MD->getDependency(C); 440 441 if (!local_dep.isDef() && !local_dep.isNonLocal()) { 442 valueNumbering[C] = nextValueNumber; 443 return nextValueNumber++; 444 } 445 446 if (local_dep.isDef()) { 447 CallInst* local_cdep = cast<CallInst>(local_dep.getInst()); 448 449 if (local_cdep->getNumOperands() != C->getNumOperands()) { 450 valueNumbering[C] = nextValueNumber; 451 return nextValueNumber++; 452 } 453 454 for (unsigned i = 1; i < C->getNumOperands(); ++i) { 455 uint32_t c_vn = lookup_or_add(C->getOperand(i)); 456 uint32_t cd_vn = lookup_or_add(local_cdep->getOperand(i)); 457 if (c_vn != cd_vn) { 458 valueNumbering[C] = nextValueNumber; 459 return nextValueNumber++; 460 } 461 } 462 463 uint32_t v = lookup_or_add(local_cdep); 464 valueNumbering[C] = v; 465 return v; 466 } 467 468 // Non-local case. 469 const MemoryDependenceAnalysis::NonLocalDepInfo &deps = 470 MD->getNonLocalCallDependency(CallSite(C)); 471 // FIXME: call/call dependencies for readonly calls should return def, not 472 // clobber! Move the checking logic to MemDep! 473 CallInst* cdep = 0; 474 475 // Check to see if we have a single dominating call instruction that is 476 // identical to C. 477 for (unsigned i = 0, e = deps.size(); i != e; ++i) { 478 const NonLocalDepEntry *I = &deps[i]; 479 // Ignore non-local dependencies. 480 if (I->getResult().isNonLocal()) 481 continue; 482 483 // We don't handle non-depedencies. If we already have a call, reject 484 // instruction dependencies. 485 if (I->getResult().isClobber() || cdep != 0) { 486 cdep = 0; 487 break; 488 } 489 490 CallInst *NonLocalDepCall = dyn_cast<CallInst>(I->getResult().getInst()); 491 // FIXME: All duplicated with non-local case. 492 if (NonLocalDepCall && DT->properlyDominates(I->getBB(), C->getParent())){ 493 cdep = NonLocalDepCall; 494 continue; 495 } 496 497 cdep = 0; 498 break; 499 } 500 501 if (!cdep) { 502 valueNumbering[C] = nextValueNumber; 503 return nextValueNumber++; 504 } 505 506 if (cdep->getNumOperands() != C->getNumOperands()) { 507 valueNumbering[C] = nextValueNumber; 508 return nextValueNumber++; 509 } 510 for (unsigned i = 1; i < C->getNumOperands(); ++i) { 511 uint32_t c_vn = lookup_or_add(C->getOperand(i)); 512 uint32_t cd_vn = lookup_or_add(cdep->getOperand(i)); 513 if (c_vn != cd_vn) { 514 valueNumbering[C] = nextValueNumber; 515 return nextValueNumber++; 516 } 517 } 518 519 uint32_t v = lookup_or_add(cdep); 520 valueNumbering[C] = v; 521 return v; 522 523 } else { 524 valueNumbering[C] = nextValueNumber; 525 return nextValueNumber++; 526 } 527} 528 529/// lookup_or_add - Returns the value number for the specified value, assigning 530/// it a new number if it did not have one before. 531uint32_t ValueTable::lookup_or_add(Value *V) { 532 DenseMap<Value*, uint32_t>::iterator VI = valueNumbering.find(V); 533 if (VI != valueNumbering.end()) 534 return VI->second; 535 536 if (!isa<Instruction>(V)) { 537 valueNumbering[V] = nextValueNumber; 538 return nextValueNumber++; 539 } 540 541 Instruction* I = cast<Instruction>(V); 542 Expression exp; 543 switch (I->getOpcode()) { 544 case Instruction::Call: 545 return lookup_or_add_call(cast<CallInst>(I)); 546 case Instruction::Add: 547 case Instruction::FAdd: 548 case Instruction::Sub: 549 case Instruction::FSub: 550 case Instruction::Mul: 551 case Instruction::FMul: 552 case Instruction::UDiv: 553 case Instruction::SDiv: 554 case Instruction::FDiv: 555 case Instruction::URem: 556 case Instruction::SRem: 557 case Instruction::FRem: 558 case Instruction::Shl: 559 case Instruction::LShr: 560 case Instruction::AShr: 561 case Instruction::And: 562 case Instruction::Or : 563 case Instruction::Xor: 564 exp = create_expression(cast<BinaryOperator>(I)); 565 break; 566 case Instruction::ICmp: 567 case Instruction::FCmp: 568 exp = create_expression(cast<CmpInst>(I)); 569 break; 570 case Instruction::Trunc: 571 case Instruction::ZExt: 572 case Instruction::SExt: 573 case Instruction::FPToUI: 574 case Instruction::FPToSI: 575 case Instruction::UIToFP: 576 case Instruction::SIToFP: 577 case Instruction::FPTrunc: 578 case Instruction::FPExt: 579 case Instruction::PtrToInt: 580 case Instruction::IntToPtr: 581 case Instruction::BitCast: 582 exp = create_expression(cast<CastInst>(I)); 583 break; 584 case Instruction::Select: 585 exp = create_expression(cast<SelectInst>(I)); 586 break; 587 case Instruction::ExtractElement: 588 exp = create_expression(cast<ExtractElementInst>(I)); 589 break; 590 case Instruction::InsertElement: 591 exp = create_expression(cast<InsertElementInst>(I)); 592 break; 593 case Instruction::ShuffleVector: 594 exp = create_expression(cast<ShuffleVectorInst>(I)); 595 break; 596 case Instruction::ExtractValue: 597 exp = create_expression(cast<ExtractValueInst>(I)); 598 break; 599 case Instruction::InsertValue: 600 exp = create_expression(cast<InsertValueInst>(I)); 601 break; 602 case Instruction::GetElementPtr: 603 exp = create_expression(cast<GetElementPtrInst>(I)); 604 break; 605 default: 606 valueNumbering[V] = nextValueNumber; 607 return nextValueNumber++; 608 } 609 610 uint32_t& e = expressionNumbering[exp]; 611 if (!e) e = nextValueNumber++; 612 valueNumbering[V] = e; 613 return e; 614} 615 616/// lookup - Returns the value number of the specified value. Fails if 617/// the value has not yet been numbered. 618uint32_t ValueTable::lookup(Value *V) const { 619 DenseMap<Value*, uint32_t>::const_iterator VI = valueNumbering.find(V); 620 assert(VI != valueNumbering.end() && "Value not numbered?"); 621 return VI->second; 622} 623 624/// clear - Remove all entries from the ValueTable 625void ValueTable::clear() { 626 valueNumbering.clear(); 627 expressionNumbering.clear(); 628 nextValueNumber = 1; 629} 630 631/// erase - Remove a value from the value numbering 632void ValueTable::erase(Value *V) { 633 valueNumbering.erase(V); 634} 635 636/// verifyRemoved - Verify that the value is removed from all internal data 637/// structures. 638void ValueTable::verifyRemoved(const Value *V) const { 639 for (DenseMap<Value*, uint32_t>::const_iterator 640 I = valueNumbering.begin(), E = valueNumbering.end(); I != E; ++I) { 641 assert(I->first != V && "Inst still occurs in value numbering map!"); 642 } 643} 644 645//===----------------------------------------------------------------------===// 646// GVN Pass 647//===----------------------------------------------------------------------===// 648 649namespace { 650 struct ValueNumberScope { 651 ValueNumberScope* parent; 652 DenseMap<uint32_t, Value*> table; 653 654 ValueNumberScope(ValueNumberScope* p) : parent(p) { } 655 }; 656} 657 658namespace { 659 660 class GVN : public FunctionPass { 661 bool runOnFunction(Function &F); 662 public: 663 static char ID; // Pass identification, replacement for typeid 664 explicit GVN(bool nopre = false, bool noloads = false) 665 : FunctionPass(&ID), NoPRE(nopre), NoLoads(noloads), MD(0) { } 666 667 private: 668 bool NoPRE; 669 bool NoLoads; 670 MemoryDependenceAnalysis *MD; 671 DominatorTree *DT; 672 673 ValueTable VN; 674 DenseMap<BasicBlock*, ValueNumberScope*> localAvail; 675 676 // This transformation requires dominator postdominator info 677 virtual void getAnalysisUsage(AnalysisUsage &AU) const { 678 AU.addRequired<DominatorTree>(); 679 if (!NoLoads) 680 AU.addRequired<MemoryDependenceAnalysis>(); 681 AU.addRequired<AliasAnalysis>(); 682 683 AU.addPreserved<DominatorTree>(); 684 AU.addPreserved<AliasAnalysis>(); 685 } 686 687 // Helper fuctions 688 // FIXME: eliminate or document these better 689 bool processLoad(LoadInst* L, 690 SmallVectorImpl<Instruction*> &toErase); 691 bool processInstruction(Instruction *I, 692 SmallVectorImpl<Instruction*> &toErase); 693 bool processNonLocalLoad(LoadInst* L, 694 SmallVectorImpl<Instruction*> &toErase); 695 bool processBlock(BasicBlock *BB); 696 void dump(DenseMap<uint32_t, Value*>& d); 697 bool iterateOnFunction(Function &F); 698 Value *CollapsePhi(PHINode* p); 699 bool performPRE(Function& F); 700 Value *lookupNumber(BasicBlock *BB, uint32_t num); 701 void cleanupGlobalSets(); 702 void verifyRemoved(const Instruction *I) const; 703 }; 704 705 char GVN::ID = 0; 706} 707 708// createGVNPass - The public interface to this file... 709FunctionPass *llvm::createGVNPass(bool NoPRE, bool NoLoads) { 710 return new GVN(NoPRE, NoLoads); 711} 712 713static RegisterPass<GVN> X("gvn", 714 "Global Value Numbering"); 715 716void GVN::dump(DenseMap<uint32_t, Value*>& d) { 717 errs() << "{\n"; 718 for (DenseMap<uint32_t, Value*>::iterator I = d.begin(), 719 E = d.end(); I != E; ++I) { 720 errs() << I->first << "\n"; 721 I->second->dump(); 722 } 723 errs() << "}\n"; 724} 725 726static bool isSafeReplacement(PHINode* p, Instruction *inst) { 727 if (!isa<PHINode>(inst)) 728 return true; 729 730 for (Instruction::use_iterator UI = p->use_begin(), E = p->use_end(); 731 UI != E; ++UI) 732 if (PHINode* use_phi = dyn_cast<PHINode>(UI)) 733 if (use_phi->getParent() == inst->getParent()) 734 return false; 735 736 return true; 737} 738 739Value *GVN::CollapsePhi(PHINode *PN) { 740 Value *ConstVal = PN->hasConstantValue(DT); 741 if (!ConstVal) return 0; 742 743 Instruction *Inst = dyn_cast<Instruction>(ConstVal); 744 if (!Inst) 745 return ConstVal; 746 747 if (DT->dominates(Inst, PN)) 748 if (isSafeReplacement(PN, Inst)) 749 return Inst; 750 return 0; 751} 752 753/// IsValueFullyAvailableInBlock - Return true if we can prove that the value 754/// we're analyzing is fully available in the specified block. As we go, keep 755/// track of which blocks we know are fully alive in FullyAvailableBlocks. This 756/// map is actually a tri-state map with the following values: 757/// 0) we know the block *is not* fully available. 758/// 1) we know the block *is* fully available. 759/// 2) we do not know whether the block is fully available or not, but we are 760/// currently speculating that it will be. 761/// 3) we are speculating for this block and have used that to speculate for 762/// other blocks. 763static bool IsValueFullyAvailableInBlock(BasicBlock *BB, 764 DenseMap<BasicBlock*, char> &FullyAvailableBlocks) { 765 // Optimistically assume that the block is fully available and check to see 766 // if we already know about this block in one lookup. 767 std::pair<DenseMap<BasicBlock*, char>::iterator, char> IV = 768 FullyAvailableBlocks.insert(std::make_pair(BB, 2)); 769 770 // If the entry already existed for this block, return the precomputed value. 771 if (!IV.second) { 772 // If this is a speculative "available" value, mark it as being used for 773 // speculation of other blocks. 774 if (IV.first->second == 2) 775 IV.first->second = 3; 776 return IV.first->second != 0; 777 } 778 779 // Otherwise, see if it is fully available in all predecessors. 780 pred_iterator PI = pred_begin(BB), PE = pred_end(BB); 781 782 // If this block has no predecessors, it isn't live-in here. 783 if (PI == PE) 784 goto SpeculationFailure; 785 786 for (; PI != PE; ++PI) 787 // If the value isn't fully available in one of our predecessors, then it 788 // isn't fully available in this block either. Undo our previous 789 // optimistic assumption and bail out. 790 if (!IsValueFullyAvailableInBlock(*PI, FullyAvailableBlocks)) 791 goto SpeculationFailure; 792 793 return true; 794 795// SpeculationFailure - If we get here, we found out that this is not, after 796// all, a fully-available block. We have a problem if we speculated on this and 797// used the speculation to mark other blocks as available. 798SpeculationFailure: 799 char &BBVal = FullyAvailableBlocks[BB]; 800 801 // If we didn't speculate on this, just return with it set to false. 802 if (BBVal == 2) { 803 BBVal = 0; 804 return false; 805 } 806 807 // If we did speculate on this value, we could have blocks set to 1 that are 808 // incorrect. Walk the (transitive) successors of this block and mark them as 809 // 0 if set to one. 810 SmallVector<BasicBlock*, 32> BBWorklist; 811 BBWorklist.push_back(BB); 812 813 do { 814 BasicBlock *Entry = BBWorklist.pop_back_val(); 815 // Note that this sets blocks to 0 (unavailable) if they happen to not 816 // already be in FullyAvailableBlocks. This is safe. 817 char &EntryVal = FullyAvailableBlocks[Entry]; 818 if (EntryVal == 0) continue; // Already unavailable. 819 820 // Mark as unavailable. 821 EntryVal = 0; 822 823 for (succ_iterator I = succ_begin(Entry), E = succ_end(Entry); I != E; ++I) 824 BBWorklist.push_back(*I); 825 } while (!BBWorklist.empty()); 826 827 return false; 828} 829 830 831/// CanCoerceMustAliasedValueToLoad - Return true if 832/// CoerceAvailableValueToLoadType will succeed. 833static bool CanCoerceMustAliasedValueToLoad(Value *StoredVal, 834 const Type *LoadTy, 835 const TargetData &TD) { 836 // If the loaded or stored value is an first class array or struct, don't try 837 // to transform them. We need to be able to bitcast to integer. 838 if (isa<StructType>(LoadTy) || isa<ArrayType>(LoadTy) || 839 isa<StructType>(StoredVal->getType()) || 840 isa<ArrayType>(StoredVal->getType())) 841 return false; 842 843 // The store has to be at least as big as the load. 844 if (TD.getTypeSizeInBits(StoredVal->getType()) < 845 TD.getTypeSizeInBits(LoadTy)) 846 return false; 847 848 return true; 849} 850 851 852/// CoerceAvailableValueToLoadType - If we saw a store of a value to memory, and 853/// then a load from a must-aliased pointer of a different type, try to coerce 854/// the stored value. LoadedTy is the type of the load we want to replace and 855/// InsertPt is the place to insert new instructions. 856/// 857/// If we can't do it, return null. 858static Value *CoerceAvailableValueToLoadType(Value *StoredVal, 859 const Type *LoadedTy, 860 Instruction *InsertPt, 861 const TargetData &TD) { 862 if (!CanCoerceMustAliasedValueToLoad(StoredVal, LoadedTy, TD)) 863 return 0; 864 865 const Type *StoredValTy = StoredVal->getType(); 866 867 uint64_t StoreSize = TD.getTypeSizeInBits(StoredValTy); 868 uint64_t LoadSize = TD.getTypeSizeInBits(LoadedTy); 869 870 // If the store and reload are the same size, we can always reuse it. 871 if (StoreSize == LoadSize) { 872 if (isa<PointerType>(StoredValTy) && isa<PointerType>(LoadedTy)) { 873 // Pointer to Pointer -> use bitcast. 874 return new BitCastInst(StoredVal, LoadedTy, "", InsertPt); 875 } 876 877 // Convert source pointers to integers, which can be bitcast. 878 if (isa<PointerType>(StoredValTy)) { 879 StoredValTy = TD.getIntPtrType(StoredValTy->getContext()); 880 StoredVal = new PtrToIntInst(StoredVal, StoredValTy, "", InsertPt); 881 } 882 883 const Type *TypeToCastTo = LoadedTy; 884 if (isa<PointerType>(TypeToCastTo)) 885 TypeToCastTo = TD.getIntPtrType(StoredValTy->getContext()); 886 887 if (StoredValTy != TypeToCastTo) 888 StoredVal = new BitCastInst(StoredVal, TypeToCastTo, "", InsertPt); 889 890 // Cast to pointer if the load needs a pointer type. 891 if (isa<PointerType>(LoadedTy)) 892 StoredVal = new IntToPtrInst(StoredVal, LoadedTy, "", InsertPt); 893 894 return StoredVal; 895 } 896 897 // If the loaded value is smaller than the available value, then we can 898 // extract out a piece from it. If the available value is too small, then we 899 // can't do anything. 900 assert(StoreSize >= LoadSize && "CanCoerceMustAliasedValueToLoad fail"); 901 902 // Convert source pointers to integers, which can be manipulated. 903 if (isa<PointerType>(StoredValTy)) { 904 StoredValTy = TD.getIntPtrType(StoredValTy->getContext()); 905 StoredVal = new PtrToIntInst(StoredVal, StoredValTy, "", InsertPt); 906 } 907 908 // Convert vectors and fp to integer, which can be manipulated. 909 if (!isa<IntegerType>(StoredValTy)) { 910 StoredValTy = IntegerType::get(StoredValTy->getContext(), StoreSize); 911 StoredVal = new BitCastInst(StoredVal, StoredValTy, "", InsertPt); 912 } 913 914 // If this is a big-endian system, we need to shift the value down to the low 915 // bits so that a truncate will work. 916 if (TD.isBigEndian()) { 917 Constant *Val = ConstantInt::get(StoredVal->getType(), StoreSize-LoadSize); 918 StoredVal = BinaryOperator::CreateLShr(StoredVal, Val, "tmp", InsertPt); 919 } 920 921 // Truncate the integer to the right size now. 922 const Type *NewIntTy = IntegerType::get(StoredValTy->getContext(), LoadSize); 923 StoredVal = new TruncInst(StoredVal, NewIntTy, "trunc", InsertPt); 924 925 if (LoadedTy == NewIntTy) 926 return StoredVal; 927 928 // If the result is a pointer, inttoptr. 929 if (isa<PointerType>(LoadedTy)) 930 return new IntToPtrInst(StoredVal, LoadedTy, "inttoptr", InsertPt); 931 932 // Otherwise, bitcast. 933 return new BitCastInst(StoredVal, LoadedTy, "bitcast", InsertPt); 934} 935 936/// GetBaseWithConstantOffset - Analyze the specified pointer to see if it can 937/// be expressed as a base pointer plus a constant offset. Return the base and 938/// offset to the caller. 939static Value *GetBaseWithConstantOffset(Value *Ptr, int64_t &Offset, 940 const TargetData &TD) { 941 Operator *PtrOp = dyn_cast<Operator>(Ptr); 942 if (PtrOp == 0) return Ptr; 943 944 // Just look through bitcasts. 945 if (PtrOp->getOpcode() == Instruction::BitCast) 946 return GetBaseWithConstantOffset(PtrOp->getOperand(0), Offset, TD); 947 948 // If this is a GEP with constant indices, we can look through it. 949 GEPOperator *GEP = dyn_cast<GEPOperator>(PtrOp); 950 if (GEP == 0 || !GEP->hasAllConstantIndices()) return Ptr; 951 952 gep_type_iterator GTI = gep_type_begin(GEP); 953 for (User::op_iterator I = GEP->idx_begin(), E = GEP->idx_end(); I != E; 954 ++I, ++GTI) { 955 ConstantInt *OpC = cast<ConstantInt>(*I); 956 if (OpC->isZero()) continue; 957 958 // Handle a struct and array indices which add their offset to the pointer. 959 if (const StructType *STy = dyn_cast<StructType>(*GTI)) { 960 Offset += TD.getStructLayout(STy)->getElementOffset(OpC->getZExtValue()); 961 } else { 962 uint64_t Size = TD.getTypeAllocSize(GTI.getIndexedType()); 963 Offset += OpC->getSExtValue()*Size; 964 } 965 } 966 967 // Re-sign extend from the pointer size if needed to get overflow edge cases 968 // right. 969 unsigned PtrSize = TD.getPointerSizeInBits(); 970 if (PtrSize < 64) 971 Offset = (Offset << (64-PtrSize)) >> (64-PtrSize); 972 973 return GetBaseWithConstantOffset(GEP->getPointerOperand(), Offset, TD); 974} 975 976 977/// AnalyzeLoadFromClobberingWrite - This function is called when we have a 978/// memdep query of a load that ends up being a clobbering memory write (store, 979/// memset, memcpy, memmove). This means that the write *may* provide bits used 980/// by the load but we can't be sure because the pointers don't mustalias. 981/// 982/// Check this case to see if there is anything more we can do before we give 983/// up. This returns -1 if we have to give up, or a byte number in the stored 984/// value of the piece that feeds the load. 985static int AnalyzeLoadFromClobberingWrite(const Type *LoadTy, Value *LoadPtr, 986 Value *WritePtr, 987 uint64_t WriteSizeInBits, 988 const TargetData &TD) { 989 // If the loaded or stored value is an first class array or struct, don't try 990 // to transform them. We need to be able to bitcast to integer. 991 if (isa<StructType>(LoadTy) || isa<ArrayType>(LoadTy)) 992 return -1; 993 994 int64_t StoreOffset = 0, LoadOffset = 0; 995 Value *StoreBase = GetBaseWithConstantOffset(WritePtr, StoreOffset, TD); 996 Value *LoadBase = 997 GetBaseWithConstantOffset(LoadPtr, LoadOffset, TD); 998 if (StoreBase != LoadBase) 999 return -1; 1000 1001 // If the load and store are to the exact same address, they should have been 1002 // a must alias. AA must have gotten confused. 1003 // FIXME: Study to see if/when this happens. 1004 if (LoadOffset == StoreOffset) { 1005#if 0 1006 dbgs() << "STORE/LOAD DEP WITH COMMON POINTER MISSED:\n" 1007 << "Base = " << *StoreBase << "\n" 1008 << "Store Ptr = " << *WritePtr << "\n" 1009 << "Store Offs = " << StoreOffset << "\n" 1010 << "Load Ptr = " << *LoadPtr << "\n"; 1011 abort(); 1012#endif 1013 return -1; 1014 } 1015 1016 // If the load and store don't overlap at all, the store doesn't provide 1017 // anything to the load. In this case, they really don't alias at all, AA 1018 // must have gotten confused. 1019 // FIXME: Investigate cases where this bails out, e.g. rdar://7238614. Then 1020 // remove this check, as it is duplicated with what we have below. 1021 uint64_t LoadSize = TD.getTypeSizeInBits(LoadTy); 1022 1023 if ((WriteSizeInBits & 7) | (LoadSize & 7)) 1024 return -1; 1025 uint64_t StoreSize = WriteSizeInBits >> 3; // Convert to bytes. 1026 LoadSize >>= 3; 1027 1028 1029 bool isAAFailure = false; 1030 if (StoreOffset < LoadOffset) { 1031 isAAFailure = StoreOffset+int64_t(StoreSize) <= LoadOffset; 1032 } else { 1033 isAAFailure = LoadOffset+int64_t(LoadSize) <= StoreOffset; 1034 } 1035 if (isAAFailure) { 1036#if 0 1037 dbgs() << "STORE LOAD DEP WITH COMMON BASE:\n" 1038 << "Base = " << *StoreBase << "\n" 1039 << "Store Ptr = " << *WritePtr << "\n" 1040 << "Store Offs = " << StoreOffset << "\n" 1041 << "Load Ptr = " << *LoadPtr << "\n"; 1042 abort(); 1043#endif 1044 return -1; 1045 } 1046 1047 // If the Load isn't completely contained within the stored bits, we don't 1048 // have all the bits to feed it. We could do something crazy in the future 1049 // (issue a smaller load then merge the bits in) but this seems unlikely to be 1050 // valuable. 1051 if (StoreOffset > LoadOffset || 1052 StoreOffset+StoreSize < LoadOffset+LoadSize) 1053 return -1; 1054 1055 // Okay, we can do this transformation. Return the number of bytes into the 1056 // store that the load is. 1057 return LoadOffset-StoreOffset; 1058} 1059 1060/// AnalyzeLoadFromClobberingStore - This function is called when we have a 1061/// memdep query of a load that ends up being a clobbering store. 1062static int AnalyzeLoadFromClobberingStore(const Type *LoadTy, Value *LoadPtr, 1063 StoreInst *DepSI, 1064 const TargetData &TD) { 1065 // Cannot handle reading from store of first-class aggregate yet. 1066 if (isa<StructType>(DepSI->getOperand(0)->getType()) || 1067 isa<ArrayType>(DepSI->getOperand(0)->getType())) 1068 return -1; 1069 1070 Value *StorePtr = DepSI->getPointerOperand(); 1071 uint64_t StoreSize = TD.getTypeSizeInBits(DepSI->getOperand(0)->getType()); 1072 return AnalyzeLoadFromClobberingWrite(LoadTy, LoadPtr, 1073 StorePtr, StoreSize, TD); 1074} 1075 1076static int AnalyzeLoadFromClobberingMemInst(const Type *LoadTy, Value *LoadPtr, 1077 MemIntrinsic *MI, 1078 const TargetData &TD) { 1079 // If the mem operation is a non-constant size, we can't handle it. 1080 ConstantInt *SizeCst = dyn_cast<ConstantInt>(MI->getLength()); 1081 if (SizeCst == 0) return -1; 1082 uint64_t MemSizeInBits = SizeCst->getZExtValue()*8; 1083 1084 // If this is memset, we just need to see if the offset is valid in the size 1085 // of the memset.. 1086 if (MI->getIntrinsicID() == Intrinsic::memset) 1087 return AnalyzeLoadFromClobberingWrite(LoadTy, LoadPtr, MI->getDest(), 1088 MemSizeInBits, TD); 1089 1090 // If we have a memcpy/memmove, the only case we can handle is if this is a 1091 // copy from constant memory. In that case, we can read directly from the 1092 // constant memory. 1093 MemTransferInst *MTI = cast<MemTransferInst>(MI); 1094 1095 Constant *Src = dyn_cast<Constant>(MTI->getSource()); 1096 if (Src == 0) return -1; 1097 1098 GlobalVariable *GV = dyn_cast<GlobalVariable>(Src->getUnderlyingObject()); 1099 if (GV == 0 || !GV->isConstant()) return -1; 1100 1101 // See if the access is within the bounds of the transfer. 1102 int Offset = AnalyzeLoadFromClobberingWrite(LoadTy, LoadPtr, 1103 MI->getDest(), MemSizeInBits, TD); 1104 if (Offset == -1) 1105 return Offset; 1106 1107 // Otherwise, see if we can constant fold a load from the constant with the 1108 // offset applied as appropriate. 1109 Src = ConstantExpr::getBitCast(Src, 1110 llvm::Type::getInt8PtrTy(Src->getContext())); 1111 Constant *OffsetCst = 1112 ConstantInt::get(Type::getInt64Ty(Src->getContext()), (unsigned)Offset); 1113 Src = ConstantExpr::getGetElementPtr(Src, &OffsetCst, 1); 1114 Src = ConstantExpr::getBitCast(Src, PointerType::getUnqual(LoadTy)); 1115 if (ConstantFoldLoadFromConstPtr(Src, &TD)) 1116 return Offset; 1117 return -1; 1118} 1119 1120 1121/// GetStoreValueForLoad - This function is called when we have a 1122/// memdep query of a load that ends up being a clobbering store. This means 1123/// that the store *may* provide bits used by the load but we can't be sure 1124/// because the pointers don't mustalias. Check this case to see if there is 1125/// anything more we can do before we give up. 1126static Value *GetStoreValueForLoad(Value *SrcVal, unsigned Offset, 1127 const Type *LoadTy, 1128 Instruction *InsertPt, const TargetData &TD){ 1129 LLVMContext &Ctx = SrcVal->getType()->getContext(); 1130 1131 uint64_t StoreSize = TD.getTypeSizeInBits(SrcVal->getType())/8; 1132 uint64_t LoadSize = TD.getTypeSizeInBits(LoadTy)/8; 1133 1134 IRBuilder<> Builder(InsertPt->getParent(), InsertPt); 1135 1136 // Compute which bits of the stored value are being used by the load. Convert 1137 // to an integer type to start with. 1138 if (isa<PointerType>(SrcVal->getType())) 1139 SrcVal = Builder.CreatePtrToInt(SrcVal, TD.getIntPtrType(Ctx), "tmp"); 1140 if (!isa<IntegerType>(SrcVal->getType())) 1141 SrcVal = Builder.CreateBitCast(SrcVal, IntegerType::get(Ctx, StoreSize*8), 1142 "tmp"); 1143 1144 // Shift the bits to the least significant depending on endianness. 1145 unsigned ShiftAmt; 1146 if (TD.isLittleEndian()) 1147 ShiftAmt = Offset*8; 1148 else 1149 ShiftAmt = (StoreSize-LoadSize-Offset)*8; 1150 1151 if (ShiftAmt) 1152 SrcVal = Builder.CreateLShr(SrcVal, ShiftAmt, "tmp"); 1153 1154 if (LoadSize != StoreSize) 1155 SrcVal = Builder.CreateTrunc(SrcVal, IntegerType::get(Ctx, LoadSize*8), 1156 "tmp"); 1157 1158 return CoerceAvailableValueToLoadType(SrcVal, LoadTy, InsertPt, TD); 1159} 1160 1161/// GetMemInstValueForLoad - This function is called when we have a 1162/// memdep query of a load that ends up being a clobbering mem intrinsic. 1163static Value *GetMemInstValueForLoad(MemIntrinsic *SrcInst, unsigned Offset, 1164 const Type *LoadTy, Instruction *InsertPt, 1165 const TargetData &TD){ 1166 LLVMContext &Ctx = LoadTy->getContext(); 1167 uint64_t LoadSize = TD.getTypeSizeInBits(LoadTy)/8; 1168 1169 IRBuilder<> Builder(InsertPt->getParent(), InsertPt); 1170 1171 // We know that this method is only called when the mem transfer fully 1172 // provides the bits for the load. 1173 if (MemSetInst *MSI = dyn_cast<MemSetInst>(SrcInst)) { 1174 // memset(P, 'x', 1234) -> splat('x'), even if x is a variable, and 1175 // independently of what the offset is. 1176 Value *Val = MSI->getValue(); 1177 if (LoadSize != 1) 1178 Val = Builder.CreateZExt(Val, IntegerType::get(Ctx, LoadSize*8)); 1179 1180 Value *OneElt = Val; 1181 1182 // Splat the value out to the right number of bits. 1183 for (unsigned NumBytesSet = 1; NumBytesSet != LoadSize; ) { 1184 // If we can double the number of bytes set, do it. 1185 if (NumBytesSet*2 <= LoadSize) { 1186 Value *ShVal = Builder.CreateShl(Val, NumBytesSet*8); 1187 Val = Builder.CreateOr(Val, ShVal); 1188 NumBytesSet <<= 1; 1189 continue; 1190 } 1191 1192 // Otherwise insert one byte at a time. 1193 Value *ShVal = Builder.CreateShl(Val, 1*8); 1194 Val = Builder.CreateOr(OneElt, ShVal); 1195 ++NumBytesSet; 1196 } 1197 1198 return CoerceAvailableValueToLoadType(Val, LoadTy, InsertPt, TD); 1199 } 1200 1201 // Otherwise, this is a memcpy/memmove from a constant global. 1202 MemTransferInst *MTI = cast<MemTransferInst>(SrcInst); 1203 Constant *Src = cast<Constant>(MTI->getSource()); 1204 1205 // Otherwise, see if we can constant fold a load from the constant with the 1206 // offset applied as appropriate. 1207 Src = ConstantExpr::getBitCast(Src, 1208 llvm::Type::getInt8PtrTy(Src->getContext())); 1209 Constant *OffsetCst = 1210 ConstantInt::get(Type::getInt64Ty(Src->getContext()), (unsigned)Offset); 1211 Src = ConstantExpr::getGetElementPtr(Src, &OffsetCst, 1); 1212 Src = ConstantExpr::getBitCast(Src, PointerType::getUnqual(LoadTy)); 1213 return ConstantFoldLoadFromConstPtr(Src, &TD); 1214} 1215 1216 1217 1218struct AvailableValueInBlock { 1219 /// BB - The basic block in question. 1220 BasicBlock *BB; 1221 enum ValType { 1222 SimpleVal, // A simple offsetted value that is accessed. 1223 MemIntrin // A memory intrinsic which is loaded from. 1224 }; 1225 1226 /// V - The value that is live out of the block. 1227 PointerIntPair<Value *, 1, ValType> Val; 1228 1229 /// Offset - The byte offset in Val that is interesting for the load query. 1230 unsigned Offset; 1231 1232 static AvailableValueInBlock get(BasicBlock *BB, Value *V, 1233 unsigned Offset = 0) { 1234 AvailableValueInBlock Res; 1235 Res.BB = BB; 1236 Res.Val.setPointer(V); 1237 Res.Val.setInt(SimpleVal); 1238 Res.Offset = Offset; 1239 return Res; 1240 } 1241 1242 static AvailableValueInBlock getMI(BasicBlock *BB, MemIntrinsic *MI, 1243 unsigned Offset = 0) { 1244 AvailableValueInBlock Res; 1245 Res.BB = BB; 1246 Res.Val.setPointer(MI); 1247 Res.Val.setInt(MemIntrin); 1248 Res.Offset = Offset; 1249 return Res; 1250 } 1251 1252 bool isSimpleValue() const { return Val.getInt() == SimpleVal; } 1253 Value *getSimpleValue() const { 1254 assert(isSimpleValue() && "Wrong accessor"); 1255 return Val.getPointer(); 1256 } 1257 1258 MemIntrinsic *getMemIntrinValue() const { 1259 assert(!isSimpleValue() && "Wrong accessor"); 1260 return cast<MemIntrinsic>(Val.getPointer()); 1261 } 1262 1263 /// MaterializeAdjustedValue - Emit code into this block to adjust the value 1264 /// defined here to the specified type. This handles various coercion cases. 1265 Value *MaterializeAdjustedValue(const Type *LoadTy, 1266 const TargetData *TD) const { 1267 Value *Res; 1268 if (isSimpleValue()) { 1269 Res = getSimpleValue(); 1270 if (Res->getType() != LoadTy) { 1271 assert(TD && "Need target data to handle type mismatch case"); 1272 Res = GetStoreValueForLoad(Res, Offset, LoadTy, BB->getTerminator(), 1273 *TD); 1274 1275 DEBUG(errs() << "GVN COERCED NONLOCAL VAL:\nOffset: " << Offset << " " 1276 << *getSimpleValue() << '\n' 1277 << *Res << '\n' << "\n\n\n"); 1278 } 1279 } else { 1280 Res = GetMemInstValueForLoad(getMemIntrinValue(), Offset, 1281 LoadTy, BB->getTerminator(), *TD); 1282 DEBUG(errs() << "GVN COERCED NONLOCAL MEM INTRIN:\nOffset: " << Offset 1283 << " " << *getMemIntrinValue() << '\n' 1284 << *Res << '\n' << "\n\n\n"); 1285 } 1286 return Res; 1287 } 1288}; 1289 1290/// ConstructSSAForLoadSet - Given a set of loads specified by ValuesPerBlock, 1291/// construct SSA form, allowing us to eliminate LI. This returns the value 1292/// that should be used at LI's definition site. 1293static Value *ConstructSSAForLoadSet(LoadInst *LI, 1294 SmallVectorImpl<AvailableValueInBlock> &ValuesPerBlock, 1295 const TargetData *TD, 1296 const DominatorTree &DT, 1297 AliasAnalysis *AA) { 1298 // Check for the fully redundant, dominating load case. In this case, we can 1299 // just use the dominating value directly. 1300 if (ValuesPerBlock.size() == 1 && 1301 DT.properlyDominates(ValuesPerBlock[0].BB, LI->getParent())) 1302 return ValuesPerBlock[0].MaterializeAdjustedValue(LI->getType(), TD); 1303 1304 // Otherwise, we have to construct SSA form. 1305 SmallVector<PHINode*, 8> NewPHIs; 1306 SSAUpdater SSAUpdate(&NewPHIs); 1307 SSAUpdate.Initialize(LI); 1308 1309 const Type *LoadTy = LI->getType(); 1310 1311 for (unsigned i = 0, e = ValuesPerBlock.size(); i != e; ++i) { 1312 const AvailableValueInBlock &AV = ValuesPerBlock[i]; 1313 BasicBlock *BB = AV.BB; 1314 1315 if (SSAUpdate.HasValueForBlock(BB)) 1316 continue; 1317 1318 SSAUpdate.AddAvailableValue(BB, AV.MaterializeAdjustedValue(LoadTy, TD)); 1319 } 1320 1321 // Perform PHI construction. 1322 Value *V = SSAUpdate.GetValueInMiddleOfBlock(LI->getParent()); 1323 1324 // If new PHI nodes were created, notify alias analysis. 1325 if (isa<PointerType>(V->getType())) 1326 for (unsigned i = 0, e = NewPHIs.size(); i != e; ++i) 1327 AA->copyValue(LI, NewPHIs[i]); 1328 1329 return V; 1330} 1331 1332static bool isLifetimeStart(Instruction *Inst) { 1333 if (IntrinsicInst* II = dyn_cast<IntrinsicInst>(Inst)) 1334 return II->getIntrinsicID() == Intrinsic::lifetime_start; 1335 return false; 1336} 1337 1338/// processNonLocalLoad - Attempt to eliminate a load whose dependencies are 1339/// non-local by performing PHI construction. 1340bool GVN::processNonLocalLoad(LoadInst *LI, 1341 SmallVectorImpl<Instruction*> &toErase) { 1342 // Find the non-local dependencies of the load. 1343 SmallVector<NonLocalDepResult, 64> Deps; 1344 MD->getNonLocalPointerDependency(LI->getOperand(0), true, LI->getParent(), 1345 Deps); 1346 //DEBUG(dbgs() << "INVESTIGATING NONLOCAL LOAD: " 1347 // << Deps.size() << *LI << '\n'); 1348 1349 // If we had to process more than one hundred blocks to find the 1350 // dependencies, this load isn't worth worrying about. Optimizing 1351 // it will be too expensive. 1352 if (Deps.size() > 100) 1353 return false; 1354 1355 // If we had a phi translation failure, we'll have a single entry which is a 1356 // clobber in the current block. Reject this early. 1357 if (Deps.size() == 1 && Deps[0].getResult().isClobber()) { 1358 DEBUG( 1359 dbgs() << "GVN: non-local load "; 1360 WriteAsOperand(dbgs(), LI); 1361 dbgs() << " is clobbered by " << *Deps[0].getResult().getInst() << '\n'; 1362 ); 1363 return false; 1364 } 1365 1366 // Filter out useless results (non-locals, etc). Keep track of the blocks 1367 // where we have a value available in repl, also keep track of whether we see 1368 // dependencies that produce an unknown value for the load (such as a call 1369 // that could potentially clobber the load). 1370 SmallVector<AvailableValueInBlock, 16> ValuesPerBlock; 1371 SmallVector<BasicBlock*, 16> UnavailableBlocks; 1372 1373 const TargetData *TD = 0; 1374 1375 for (unsigned i = 0, e = Deps.size(); i != e; ++i) { 1376 BasicBlock *DepBB = Deps[i].getBB(); 1377 MemDepResult DepInfo = Deps[i].getResult(); 1378 1379 if (DepInfo.isClobber()) { 1380 // The address being loaded in this non-local block may not be the same as 1381 // the pointer operand of the load if PHI translation occurs. Make sure 1382 // to consider the right address. 1383 Value *Address = Deps[i].getAddress(); 1384 1385 // If the dependence is to a store that writes to a superset of the bits 1386 // read by the load, we can extract the bits we need for the load from the 1387 // stored value. 1388 if (StoreInst *DepSI = dyn_cast<StoreInst>(DepInfo.getInst())) { 1389 if (TD == 0) 1390 TD = getAnalysisIfAvailable<TargetData>(); 1391 if (TD && Address) { 1392 int Offset = AnalyzeLoadFromClobberingStore(LI->getType(), Address, 1393 DepSI, *TD); 1394 if (Offset != -1) { 1395 ValuesPerBlock.push_back(AvailableValueInBlock::get(DepBB, 1396 DepSI->getOperand(0), 1397 Offset)); 1398 continue; 1399 } 1400 } 1401 } 1402 1403 // If the clobbering value is a memset/memcpy/memmove, see if we can 1404 // forward a value on from it. 1405 if (MemIntrinsic *DepMI = dyn_cast<MemIntrinsic>(DepInfo.getInst())) { 1406 if (TD == 0) 1407 TD = getAnalysisIfAvailable<TargetData>(); 1408 if (TD && Address) { 1409 int Offset = AnalyzeLoadFromClobberingMemInst(LI->getType(), Address, 1410 DepMI, *TD); 1411 if (Offset != -1) { 1412 ValuesPerBlock.push_back(AvailableValueInBlock::getMI(DepBB, DepMI, 1413 Offset)); 1414 continue; 1415 } 1416 } 1417 } 1418 1419 UnavailableBlocks.push_back(DepBB); 1420 continue; 1421 } 1422 1423 Instruction *DepInst = DepInfo.getInst(); 1424 1425 // Loading the allocation -> undef. 1426 if (isa<AllocaInst>(DepInst) || isMalloc(DepInst) || 1427 // Loading immediately after lifetime begin -> undef. 1428 isLifetimeStart(DepInst)) { 1429 ValuesPerBlock.push_back(AvailableValueInBlock::get(DepBB, 1430 UndefValue::get(LI->getType()))); 1431 continue; 1432 } 1433 1434 if (StoreInst *S = dyn_cast<StoreInst>(DepInst)) { 1435 // Reject loads and stores that are to the same address but are of 1436 // different types if we have to. 1437 if (S->getOperand(0)->getType() != LI->getType()) { 1438 if (TD == 0) 1439 TD = getAnalysisIfAvailable<TargetData>(); 1440 1441 // If the stored value is larger or equal to the loaded value, we can 1442 // reuse it. 1443 if (TD == 0 || !CanCoerceMustAliasedValueToLoad(S->getOperand(0), 1444 LI->getType(), *TD)) { 1445 UnavailableBlocks.push_back(DepBB); 1446 continue; 1447 } 1448 } 1449 1450 ValuesPerBlock.push_back(AvailableValueInBlock::get(DepBB, 1451 S->getOperand(0))); 1452 continue; 1453 } 1454 1455 if (LoadInst *LD = dyn_cast<LoadInst>(DepInst)) { 1456 // If the types mismatch and we can't handle it, reject reuse of the load. 1457 if (LD->getType() != LI->getType()) { 1458 if (TD == 0) 1459 TD = getAnalysisIfAvailable<TargetData>(); 1460 1461 // If the stored value is larger or equal to the loaded value, we can 1462 // reuse it. 1463 if (TD == 0 || !CanCoerceMustAliasedValueToLoad(LD, LI->getType(),*TD)){ 1464 UnavailableBlocks.push_back(DepBB); 1465 continue; 1466 } 1467 } 1468 ValuesPerBlock.push_back(AvailableValueInBlock::get(DepBB, LD)); 1469 continue; 1470 } 1471 1472 UnavailableBlocks.push_back(DepBB); 1473 continue; 1474 } 1475 1476 // If we have no predecessors that produce a known value for this load, exit 1477 // early. 1478 if (ValuesPerBlock.empty()) return false; 1479 1480 // If all of the instructions we depend on produce a known value for this 1481 // load, then it is fully redundant and we can use PHI insertion to compute 1482 // its value. Insert PHIs and remove the fully redundant value now. 1483 if (UnavailableBlocks.empty()) { 1484 DEBUG(dbgs() << "GVN REMOVING NONLOCAL LOAD: " << *LI << '\n'); 1485 1486 // Perform PHI construction. 1487 Value *V = ConstructSSAForLoadSet(LI, ValuesPerBlock, TD, *DT, 1488 VN.getAliasAnalysis()); 1489 LI->replaceAllUsesWith(V); 1490 1491 if (isa<PHINode>(V)) 1492 V->takeName(LI); 1493 if (isa<PointerType>(V->getType())) 1494 MD->invalidateCachedPointerInfo(V); 1495 toErase.push_back(LI); 1496 NumGVNLoad++; 1497 return true; 1498 } 1499 1500 if (!EnablePRE || !EnableLoadPRE) 1501 return false; 1502 1503 // Okay, we have *some* definitions of the value. This means that the value 1504 // is available in some of our (transitive) predecessors. Lets think about 1505 // doing PRE of this load. This will involve inserting a new load into the 1506 // predecessor when it's not available. We could do this in general, but 1507 // prefer to not increase code size. As such, we only do this when we know 1508 // that we only have to insert *one* load (which means we're basically moving 1509 // the load, not inserting a new one). 1510 1511 SmallPtrSet<BasicBlock *, 4> Blockers; 1512 for (unsigned i = 0, e = UnavailableBlocks.size(); i != e; ++i) 1513 Blockers.insert(UnavailableBlocks[i]); 1514 1515 // Lets find first basic block with more than one predecessor. Walk backwards 1516 // through predecessors if needed. 1517 BasicBlock *LoadBB = LI->getParent(); 1518 BasicBlock *TmpBB = LoadBB; 1519 1520 bool isSinglePred = false; 1521 bool allSingleSucc = true; 1522 while (TmpBB->getSinglePredecessor()) { 1523 isSinglePred = true; 1524 TmpBB = TmpBB->getSinglePredecessor(); 1525 if (TmpBB == LoadBB) // Infinite (unreachable) loop. 1526 return false; 1527 if (Blockers.count(TmpBB)) 1528 return false; 1529 if (TmpBB->getTerminator()->getNumSuccessors() != 1) 1530 allSingleSucc = false; 1531 } 1532 1533 assert(TmpBB); 1534 LoadBB = TmpBB; 1535 1536 // If we have a repl set with LI itself in it, this means we have a loop where 1537 // at least one of the values is LI. Since this means that we won't be able 1538 // to eliminate LI even if we insert uses in the other predecessors, we will 1539 // end up increasing code size. Reject this by scanning for LI. 1540 for (unsigned i = 0, e = ValuesPerBlock.size(); i != e; ++i) 1541 if (ValuesPerBlock[i].isSimpleValue() && 1542 ValuesPerBlock[i].getSimpleValue() == LI) 1543 return false; 1544 1545 // FIXME: It is extremely unclear what this loop is doing, other than 1546 // artificially restricting loadpre. 1547 if (isSinglePred) { 1548 bool isHot = false; 1549 for (unsigned i = 0, e = ValuesPerBlock.size(); i != e; ++i) { 1550 const AvailableValueInBlock &AV = ValuesPerBlock[i]; 1551 if (AV.isSimpleValue()) 1552 // "Hot" Instruction is in some loop (because it dominates its dep. 1553 // instruction). 1554 if (Instruction *I = dyn_cast<Instruction>(AV.getSimpleValue())) 1555 if (DT->dominates(LI, I)) { 1556 isHot = true; 1557 break; 1558 } 1559 } 1560 1561 // We are interested only in "hot" instructions. We don't want to do any 1562 // mis-optimizations here. 1563 if (!isHot) 1564 return false; 1565 } 1566 1567 // Okay, we have some hope :). Check to see if the loaded value is fully 1568 // available in all but one predecessor. 1569 // FIXME: If we could restructure the CFG, we could make a common pred with 1570 // all the preds that don't have an available LI and insert a new load into 1571 // that one block. 1572 BasicBlock *UnavailablePred = 0; 1573 1574 DenseMap<BasicBlock*, char> FullyAvailableBlocks; 1575 for (unsigned i = 0, e = ValuesPerBlock.size(); i != e; ++i) 1576 FullyAvailableBlocks[ValuesPerBlock[i].BB] = true; 1577 for (unsigned i = 0, e = UnavailableBlocks.size(); i != e; ++i) 1578 FullyAvailableBlocks[UnavailableBlocks[i]] = false; 1579 1580 for (pred_iterator PI = pred_begin(LoadBB), E = pred_end(LoadBB); 1581 PI != E; ++PI) { 1582 if (IsValueFullyAvailableInBlock(*PI, FullyAvailableBlocks)) 1583 continue; 1584 1585 // If this load is not available in multiple predecessors, reject it. 1586 if (UnavailablePred && UnavailablePred != *PI) 1587 return false; 1588 UnavailablePred = *PI; 1589 } 1590 1591 assert(UnavailablePred != 0 && 1592 "Fully available value should be eliminated above!"); 1593 1594 // We don't currently handle critical edges :( 1595 if (UnavailablePred->getTerminator()->getNumSuccessors() != 1) { 1596 DEBUG(dbgs() << "COULD NOT PRE LOAD BECAUSE OF CRITICAL EDGE '" 1597 << UnavailablePred->getName() << "': " << *LI << '\n'); 1598 return false; 1599 } 1600 1601 // Do PHI translation to get its value in the predecessor if necessary. The 1602 // returned pointer (if non-null) is guaranteed to dominate UnavailablePred. 1603 // 1604 SmallVector<Instruction*, 8> NewInsts; 1605 1606 // If all preds have a single successor, then we know it is safe to insert the 1607 // load on the pred (?!?), so we can insert code to materialize the pointer if 1608 // it is not available. 1609 PHITransAddr Address(LI->getOperand(0), TD); 1610 Value *LoadPtr = 0; 1611 if (allSingleSucc) { 1612 LoadPtr = Address.PHITranslateWithInsertion(LoadBB, UnavailablePred, 1613 *DT, NewInsts); 1614 } else { 1615 Address.PHITranslateValue(LoadBB, UnavailablePred); 1616 LoadPtr = Address.getAddr(); 1617 1618 // Make sure the value is live in the predecessor. 1619 if (Instruction *Inst = dyn_cast_or_null<Instruction>(LoadPtr)) 1620 if (!DT->dominates(Inst->getParent(), UnavailablePred)) 1621 LoadPtr = 0; 1622 } 1623 1624 // If we couldn't find or insert a computation of this phi translated value, 1625 // we fail PRE. 1626 if (LoadPtr == 0) { 1627 assert(NewInsts.empty() && "Shouldn't insert insts on failure"); 1628 DEBUG(dbgs() << "COULDN'T INSERT PHI TRANSLATED VALUE OF: " 1629 << *LI->getOperand(0) << "\n"); 1630 return false; 1631 } 1632 1633 // Assign value numbers to these new instructions. 1634 for (unsigned i = 0, e = NewInsts.size(); i != e; ++i) { 1635 // FIXME: We really _ought_ to insert these value numbers into their 1636 // parent's availability map. However, in doing so, we risk getting into 1637 // ordering issues. If a block hasn't been processed yet, we would be 1638 // marking a value as AVAIL-IN, which isn't what we intend. 1639 VN.lookup_or_add(NewInsts[i]); 1640 } 1641 1642 // Make sure it is valid to move this load here. We have to watch out for: 1643 // @1 = getelementptr (i8* p, ... 1644 // test p and branch if == 0 1645 // load @1 1646 // It is valid to have the getelementptr before the test, even if p can be 0, 1647 // as getelementptr only does address arithmetic. 1648 // If we are not pushing the value through any multiple-successor blocks 1649 // we do not have this case. Otherwise, check that the load is safe to 1650 // put anywhere; this can be improved, but should be conservatively safe. 1651 if (!allSingleSucc && 1652 // FIXME: REEVALUTE THIS. 1653 !isSafeToLoadUnconditionally(LoadPtr, 1654 UnavailablePred->getTerminator(), TD)) { 1655 assert(NewInsts.empty() && "Should not have inserted instructions"); 1656 return false; 1657 } 1658 1659 // Okay, we can eliminate this load by inserting a reload in the predecessor 1660 // and using PHI construction to get the value in the other predecessors, do 1661 // it. 1662 DEBUG(dbgs() << "GVN REMOVING PRE LOAD: " << *LI << '\n'); 1663 DEBUG(if (!NewInsts.empty()) 1664 dbgs() << "INSERTED " << NewInsts.size() << " INSTS: " 1665 << *NewInsts.back() << '\n'); 1666 1667 Value *NewLoad = new LoadInst(LoadPtr, LI->getName()+".pre", false, 1668 LI->getAlignment(), 1669 UnavailablePred->getTerminator()); 1670 1671 // Add the newly created load. 1672 ValuesPerBlock.push_back(AvailableValueInBlock::get(UnavailablePred,NewLoad)); 1673 1674 // Perform PHI construction. 1675 Value *V = ConstructSSAForLoadSet(LI, ValuesPerBlock, TD, *DT, 1676 VN.getAliasAnalysis()); 1677 LI->replaceAllUsesWith(V); 1678 if (isa<PHINode>(V)) 1679 V->takeName(LI); 1680 if (isa<PointerType>(V->getType())) 1681 MD->invalidateCachedPointerInfo(V); 1682 toErase.push_back(LI); 1683 NumPRELoad++; 1684 return true; 1685} 1686 1687/// processLoad - Attempt to eliminate a load, first by eliminating it 1688/// locally, and then attempting non-local elimination if that fails. 1689bool GVN::processLoad(LoadInst *L, SmallVectorImpl<Instruction*> &toErase) { 1690 if (!MD) 1691 return false; 1692 1693 if (L->isVolatile()) 1694 return false; 1695 1696 // ... to a pointer that has been loaded from before... 1697 MemDepResult Dep = MD->getDependency(L); 1698 1699 // If the value isn't available, don't do anything! 1700 if (Dep.isClobber()) { 1701 // Check to see if we have something like this: 1702 // store i32 123, i32* %P 1703 // %A = bitcast i32* %P to i8* 1704 // %B = gep i8* %A, i32 1 1705 // %C = load i8* %B 1706 // 1707 // We could do that by recognizing if the clobber instructions are obviously 1708 // a common base + constant offset, and if the previous store (or memset) 1709 // completely covers this load. This sort of thing can happen in bitfield 1710 // access code. 1711 Value *AvailVal = 0; 1712 if (StoreInst *DepSI = dyn_cast<StoreInst>(Dep.getInst())) 1713 if (const TargetData *TD = getAnalysisIfAvailable<TargetData>()) { 1714 int Offset = AnalyzeLoadFromClobberingStore(L->getType(), 1715 L->getPointerOperand(), 1716 DepSI, *TD); 1717 if (Offset != -1) 1718 AvailVal = GetStoreValueForLoad(DepSI->getOperand(0), Offset, 1719 L->getType(), L, *TD); 1720 } 1721 1722 // If the clobbering value is a memset/memcpy/memmove, see if we can forward 1723 // a value on from it. 1724 if (MemIntrinsic *DepMI = dyn_cast<MemIntrinsic>(Dep.getInst())) { 1725 if (const TargetData *TD = getAnalysisIfAvailable<TargetData>()) { 1726 int Offset = AnalyzeLoadFromClobberingMemInst(L->getType(), 1727 L->getPointerOperand(), 1728 DepMI, *TD); 1729 if (Offset != -1) 1730 AvailVal = GetMemInstValueForLoad(DepMI, Offset, L->getType(), L,*TD); 1731 } 1732 } 1733 1734 if (AvailVal) { 1735 DEBUG(dbgs() << "GVN COERCED INST:\n" << *Dep.getInst() << '\n' 1736 << *AvailVal << '\n' << *L << "\n\n\n"); 1737 1738 // Replace the load! 1739 L->replaceAllUsesWith(AvailVal); 1740 if (isa<PointerType>(AvailVal->getType())) 1741 MD->invalidateCachedPointerInfo(AvailVal); 1742 toErase.push_back(L); 1743 NumGVNLoad++; 1744 return true; 1745 } 1746 1747 DEBUG( 1748 // fast print dep, using operator<< on instruction would be too slow 1749 dbgs() << "GVN: load "; 1750 WriteAsOperand(dbgs(), L); 1751 Instruction *I = Dep.getInst(); 1752 dbgs() << " is clobbered by " << *I << '\n'; 1753 ); 1754 return false; 1755 } 1756 1757 // If it is defined in another block, try harder. 1758 if (Dep.isNonLocal()) 1759 return processNonLocalLoad(L, toErase); 1760 1761 Instruction *DepInst = Dep.getInst(); 1762 if (StoreInst *DepSI = dyn_cast<StoreInst>(DepInst)) { 1763 Value *StoredVal = DepSI->getOperand(0); 1764 1765 // The store and load are to a must-aliased pointer, but they may not 1766 // actually have the same type. See if we know how to reuse the stored 1767 // value (depending on its type). 1768 const TargetData *TD = 0; 1769 if (StoredVal->getType() != L->getType()) { 1770 if ((TD = getAnalysisIfAvailable<TargetData>())) { 1771 StoredVal = CoerceAvailableValueToLoadType(StoredVal, L->getType(), 1772 L, *TD); 1773 if (StoredVal == 0) 1774 return false; 1775 1776 DEBUG(dbgs() << "GVN COERCED STORE:\n" << *DepSI << '\n' << *StoredVal 1777 << '\n' << *L << "\n\n\n"); 1778 } 1779 else 1780 return false; 1781 } 1782 1783 // Remove it! 1784 L->replaceAllUsesWith(StoredVal); 1785 if (isa<PointerType>(StoredVal->getType())) 1786 MD->invalidateCachedPointerInfo(StoredVal); 1787 toErase.push_back(L); 1788 NumGVNLoad++; 1789 return true; 1790 } 1791 1792 if (LoadInst *DepLI = dyn_cast<LoadInst>(DepInst)) { 1793 Value *AvailableVal = DepLI; 1794 1795 // The loads are of a must-aliased pointer, but they may not actually have 1796 // the same type. See if we know how to reuse the previously loaded value 1797 // (depending on its type). 1798 const TargetData *TD = 0; 1799 if (DepLI->getType() != L->getType()) { 1800 if ((TD = getAnalysisIfAvailable<TargetData>())) { 1801 AvailableVal = CoerceAvailableValueToLoadType(DepLI, L->getType(), L,*TD); 1802 if (AvailableVal == 0) 1803 return false; 1804 1805 DEBUG(dbgs() << "GVN COERCED LOAD:\n" << *DepLI << "\n" << *AvailableVal 1806 << "\n" << *L << "\n\n\n"); 1807 } 1808 else 1809 return false; 1810 } 1811 1812 // Remove it! 1813 L->replaceAllUsesWith(AvailableVal); 1814 if (isa<PointerType>(DepLI->getType())) 1815 MD->invalidateCachedPointerInfo(DepLI); 1816 toErase.push_back(L); 1817 NumGVNLoad++; 1818 return true; 1819 } 1820 1821 // If this load really doesn't depend on anything, then we must be loading an 1822 // undef value. This can happen when loading for a fresh allocation with no 1823 // intervening stores, for example. 1824 if (isa<AllocaInst>(DepInst) || isMalloc(DepInst)) { 1825 L->replaceAllUsesWith(UndefValue::get(L->getType())); 1826 toErase.push_back(L); 1827 NumGVNLoad++; 1828 return true; 1829 } 1830 1831 // If this load occurs either right after a lifetime begin, 1832 // then the loaded value is undefined. 1833 if (IntrinsicInst* II = dyn_cast<IntrinsicInst>(DepInst)) { 1834 if (II->getIntrinsicID() == Intrinsic::lifetime_start) { 1835 L->replaceAllUsesWith(UndefValue::get(L->getType())); 1836 toErase.push_back(L); 1837 NumGVNLoad++; 1838 return true; 1839 } 1840 } 1841 1842 return false; 1843} 1844 1845Value *GVN::lookupNumber(BasicBlock *BB, uint32_t num) { 1846 DenseMap<BasicBlock*, ValueNumberScope*>::iterator I = localAvail.find(BB); 1847 if (I == localAvail.end()) 1848 return 0; 1849 1850 ValueNumberScope *Locals = I->second; 1851 while (Locals) { 1852 DenseMap<uint32_t, Value*>::iterator I = Locals->table.find(num); 1853 if (I != Locals->table.end()) 1854 return I->second; 1855 Locals = Locals->parent; 1856 } 1857 1858 return 0; 1859} 1860 1861 1862/// processInstruction - When calculating availability, handle an instruction 1863/// by inserting it into the appropriate sets 1864bool GVN::processInstruction(Instruction *I, 1865 SmallVectorImpl<Instruction*> &toErase) { 1866 if (LoadInst *LI = dyn_cast<LoadInst>(I)) { 1867 bool Changed = processLoad(LI, toErase); 1868 1869 if (!Changed) { 1870 unsigned Num = VN.lookup_or_add(LI); 1871 localAvail[I->getParent()]->table.insert(std::make_pair(Num, LI)); 1872 } 1873 1874 return Changed; 1875 } 1876 1877 uint32_t NextNum = VN.getNextUnusedValueNumber(); 1878 unsigned Num = VN.lookup_or_add(I); 1879 1880 if (BranchInst *BI = dyn_cast<BranchInst>(I)) { 1881 localAvail[I->getParent()]->table.insert(std::make_pair(Num, I)); 1882 1883 if (!BI->isConditional() || isa<Constant>(BI->getCondition())) 1884 return false; 1885 1886 Value *BranchCond = BI->getCondition(); 1887 uint32_t CondVN = VN.lookup_or_add(BranchCond); 1888 1889 BasicBlock *TrueSucc = BI->getSuccessor(0); 1890 BasicBlock *FalseSucc = BI->getSuccessor(1); 1891 1892 if (TrueSucc->getSinglePredecessor()) 1893 localAvail[TrueSucc]->table[CondVN] = 1894 ConstantInt::getTrue(TrueSucc->getContext()); 1895 if (FalseSucc->getSinglePredecessor()) 1896 localAvail[FalseSucc]->table[CondVN] = 1897 ConstantInt::getFalse(TrueSucc->getContext()); 1898 1899 return false; 1900 1901 // Allocations are always uniquely numbered, so we can save time and memory 1902 // by fast failing them. 1903 } else if (isa<AllocaInst>(I) || isa<TerminatorInst>(I)) { 1904 localAvail[I->getParent()]->table.insert(std::make_pair(Num, I)); 1905 return false; 1906 } 1907 1908 // Collapse PHI nodes 1909 if (PHINode* p = dyn_cast<PHINode>(I)) { 1910 Value *constVal = CollapsePhi(p); 1911 1912 if (constVal) { 1913 p->replaceAllUsesWith(constVal); 1914 if (MD && isa<PointerType>(constVal->getType())) 1915 MD->invalidateCachedPointerInfo(constVal); 1916 VN.erase(p); 1917 1918 toErase.push_back(p); 1919 } else { 1920 localAvail[I->getParent()]->table.insert(std::make_pair(Num, I)); 1921 } 1922 1923 // If the number we were assigned was a brand new VN, then we don't 1924 // need to do a lookup to see if the number already exists 1925 // somewhere in the domtree: it can't! 1926 } else if (Num == NextNum) { 1927 localAvail[I->getParent()]->table.insert(std::make_pair(Num, I)); 1928 1929 // Perform fast-path value-number based elimination of values inherited from 1930 // dominators. 1931 } else if (Value *repl = lookupNumber(I->getParent(), Num)) { 1932 // Remove it! 1933 VN.erase(I); 1934 I->replaceAllUsesWith(repl); 1935 if (MD && isa<PointerType>(repl->getType())) 1936 MD->invalidateCachedPointerInfo(repl); 1937 toErase.push_back(I); 1938 return true; 1939 1940 } else { 1941 localAvail[I->getParent()]->table.insert(std::make_pair(Num, I)); 1942 } 1943 1944 return false; 1945} 1946 1947/// runOnFunction - This is the main transformation entry point for a function. 1948bool GVN::runOnFunction(Function& F) { 1949 if (!NoLoads) 1950 MD = &getAnalysis<MemoryDependenceAnalysis>(); 1951 DT = &getAnalysis<DominatorTree>(); 1952 VN.setAliasAnalysis(&getAnalysis<AliasAnalysis>()); 1953 VN.setMemDep(MD); 1954 VN.setDomTree(DT); 1955 1956 bool Changed = false; 1957 bool ShouldContinue = true; 1958 1959 // Merge unconditional branches, allowing PRE to catch more 1960 // optimization opportunities. 1961 for (Function::iterator FI = F.begin(), FE = F.end(); FI != FE; ) { 1962 BasicBlock *BB = FI; 1963 ++FI; 1964 bool removedBlock = MergeBlockIntoPredecessor(BB, this); 1965 if (removedBlock) NumGVNBlocks++; 1966 1967 Changed |= removedBlock; 1968 } 1969 1970 unsigned Iteration = 0; 1971 1972 while (ShouldContinue) { 1973 DEBUG(dbgs() << "GVN iteration: " << Iteration << "\n"); 1974 ShouldContinue = iterateOnFunction(F); 1975 Changed |= ShouldContinue; 1976 ++Iteration; 1977 } 1978 1979 if (EnablePRE) { 1980 bool PREChanged = true; 1981 while (PREChanged) { 1982 PREChanged = performPRE(F); 1983 Changed |= PREChanged; 1984 } 1985 } 1986 // FIXME: Should perform GVN again after PRE does something. PRE can move 1987 // computations into blocks where they become fully redundant. Note that 1988 // we can't do this until PRE's critical edge splitting updates memdep. 1989 // Actually, when this happens, we should just fully integrate PRE into GVN. 1990 1991 cleanupGlobalSets(); 1992 1993 return Changed; 1994} 1995 1996 1997bool GVN::processBlock(BasicBlock *BB) { 1998 // FIXME: Kill off toErase by doing erasing eagerly in a helper function (and 1999 // incrementing BI before processing an instruction). 2000 SmallVector<Instruction*, 8> toErase; 2001 bool ChangedFunction = false; 2002 2003 for (BasicBlock::iterator BI = BB->begin(), BE = BB->end(); 2004 BI != BE;) { 2005 ChangedFunction |= processInstruction(BI, toErase); 2006 if (toErase.empty()) { 2007 ++BI; 2008 continue; 2009 } 2010 2011 // If we need some instructions deleted, do it now. 2012 NumGVNInstr += toErase.size(); 2013 2014 // Avoid iterator invalidation. 2015 bool AtStart = BI == BB->begin(); 2016 if (!AtStart) 2017 --BI; 2018 2019 for (SmallVector<Instruction*, 4>::iterator I = toErase.begin(), 2020 E = toErase.end(); I != E; ++I) { 2021 DEBUG(dbgs() << "GVN removed: " << **I << '\n'); 2022 if (MD) MD->removeInstruction(*I); 2023 (*I)->eraseFromParent(); 2024 DEBUG(verifyRemoved(*I)); 2025 } 2026 toErase.clear(); 2027 2028 if (AtStart) 2029 BI = BB->begin(); 2030 else 2031 ++BI; 2032 } 2033 2034 return ChangedFunction; 2035} 2036 2037/// performPRE - Perform a purely local form of PRE that looks for diamond 2038/// control flow patterns and attempts to perform simple PRE at the join point. 2039bool GVN::performPRE(Function &F) { 2040 bool Changed = false; 2041 SmallVector<std::pair<TerminatorInst*, unsigned>, 4> toSplit; 2042 DenseMap<BasicBlock*, Value*> predMap; 2043 for (df_iterator<BasicBlock*> DI = df_begin(&F.getEntryBlock()), 2044 DE = df_end(&F.getEntryBlock()); DI != DE; ++DI) { 2045 BasicBlock *CurrentBlock = *DI; 2046 2047 // Nothing to PRE in the entry block. 2048 if (CurrentBlock == &F.getEntryBlock()) continue; 2049 2050 for (BasicBlock::iterator BI = CurrentBlock->begin(), 2051 BE = CurrentBlock->end(); BI != BE; ) { 2052 Instruction *CurInst = BI++; 2053 2054 if (isa<AllocaInst>(CurInst) || 2055 isa<TerminatorInst>(CurInst) || isa<PHINode>(CurInst) || 2056 CurInst->getType()->isVoidTy() || 2057 CurInst->mayReadFromMemory() || CurInst->mayHaveSideEffects() || 2058 isa<DbgInfoIntrinsic>(CurInst)) 2059 continue; 2060 2061 uint32_t ValNo = VN.lookup(CurInst); 2062 2063 // Look for the predecessors for PRE opportunities. We're 2064 // only trying to solve the basic diamond case, where 2065 // a value is computed in the successor and one predecessor, 2066 // but not the other. We also explicitly disallow cases 2067 // where the successor is its own predecessor, because they're 2068 // more complicated to get right. 2069 unsigned NumWith = 0; 2070 unsigned NumWithout = 0; 2071 BasicBlock *PREPred = 0; 2072 predMap.clear(); 2073 2074 for (pred_iterator PI = pred_begin(CurrentBlock), 2075 PE = pred_end(CurrentBlock); PI != PE; ++PI) { 2076 // We're not interested in PRE where the block is its 2077 // own predecessor, on in blocks with predecessors 2078 // that are not reachable. 2079 if (*PI == CurrentBlock) { 2080 NumWithout = 2; 2081 break; 2082 } else if (!localAvail.count(*PI)) { 2083 NumWithout = 2; 2084 break; 2085 } 2086 2087 DenseMap<uint32_t, Value*>::iterator predV = 2088 localAvail[*PI]->table.find(ValNo); 2089 if (predV == localAvail[*PI]->table.end()) { 2090 PREPred = *PI; 2091 NumWithout++; 2092 } else if (predV->second == CurInst) { 2093 NumWithout = 2; 2094 } else { 2095 predMap[*PI] = predV->second; 2096 NumWith++; 2097 } 2098 } 2099 2100 // Don't do PRE when it might increase code size, i.e. when 2101 // we would need to insert instructions in more than one pred. 2102 if (NumWithout != 1 || NumWith == 0) 2103 continue; 2104 2105 // Don't do PRE across indirect branch. 2106 if (isa<IndirectBrInst>(PREPred->getTerminator())) 2107 continue; 2108 2109 // We can't do PRE safely on a critical edge, so instead we schedule 2110 // the edge to be split and perform the PRE the next time we iterate 2111 // on the function. 2112 unsigned SuccNum = 0; 2113 for (unsigned i = 0, e = PREPred->getTerminator()->getNumSuccessors(); 2114 i != e; ++i) 2115 if (PREPred->getTerminator()->getSuccessor(i) == CurrentBlock) { 2116 SuccNum = i; 2117 break; 2118 } 2119 2120 if (isCriticalEdge(PREPred->getTerminator(), SuccNum)) { 2121 toSplit.push_back(std::make_pair(PREPred->getTerminator(), SuccNum)); 2122 continue; 2123 } 2124 2125 // Instantiate the expression the in predecessor that lacked it. 2126 // Because we are going top-down through the block, all value numbers 2127 // will be available in the predecessor by the time we need them. Any 2128 // that weren't original present will have been instantiated earlier 2129 // in this loop. 2130 Instruction *PREInstr = CurInst->clone(); 2131 bool success = true; 2132 for (unsigned i = 0, e = CurInst->getNumOperands(); i != e; ++i) { 2133 Value *Op = PREInstr->getOperand(i); 2134 if (isa<Argument>(Op) || isa<Constant>(Op) || isa<GlobalValue>(Op)) 2135 continue; 2136 2137 if (Value *V = lookupNumber(PREPred, VN.lookup(Op))) { 2138 PREInstr->setOperand(i, V); 2139 } else { 2140 success = false; 2141 break; 2142 } 2143 } 2144 2145 // Fail out if we encounter an operand that is not available in 2146 // the PRE predecessor. This is typically because of loads which 2147 // are not value numbered precisely. 2148 if (!success) { 2149 delete PREInstr; 2150 DEBUG(verifyRemoved(PREInstr)); 2151 continue; 2152 } 2153 2154 PREInstr->insertBefore(PREPred->getTerminator()); 2155 PREInstr->setName(CurInst->getName() + ".pre"); 2156 predMap[PREPred] = PREInstr; 2157 VN.add(PREInstr, ValNo); 2158 NumGVNPRE++; 2159 2160 // Update the availability map to include the new instruction. 2161 localAvail[PREPred]->table.insert(std::make_pair(ValNo, PREInstr)); 2162 2163 // Create a PHI to make the value available in this block. 2164 PHINode* Phi = PHINode::Create(CurInst->getType(), 2165 CurInst->getName() + ".pre-phi", 2166 CurrentBlock->begin()); 2167 for (pred_iterator PI = pred_begin(CurrentBlock), 2168 PE = pred_end(CurrentBlock); PI != PE; ++PI) 2169 Phi->addIncoming(predMap[*PI], *PI); 2170 2171 VN.add(Phi, ValNo); 2172 localAvail[CurrentBlock]->table[ValNo] = Phi; 2173 2174 CurInst->replaceAllUsesWith(Phi); 2175 if (MD && isa<PointerType>(Phi->getType())) 2176 MD->invalidateCachedPointerInfo(Phi); 2177 VN.erase(CurInst); 2178 2179 DEBUG(dbgs() << "GVN PRE removed: " << *CurInst << '\n'); 2180 if (MD) MD->removeInstruction(CurInst); 2181 CurInst->eraseFromParent(); 2182 DEBUG(verifyRemoved(CurInst)); 2183 Changed = true; 2184 } 2185 } 2186 2187 for (SmallVector<std::pair<TerminatorInst*, unsigned>, 4>::iterator 2188 I = toSplit.begin(), E = toSplit.end(); I != E; ++I) 2189 SplitCriticalEdge(I->first, I->second, this); 2190 2191 return Changed || toSplit.size(); 2192} 2193 2194/// iterateOnFunction - Executes one iteration of GVN 2195bool GVN::iterateOnFunction(Function &F) { 2196 cleanupGlobalSets(); 2197 2198 for (df_iterator<DomTreeNode*> DI = df_begin(DT->getRootNode()), 2199 DE = df_end(DT->getRootNode()); DI != DE; ++DI) { 2200 if (DI->getIDom()) 2201 localAvail[DI->getBlock()] = 2202 new ValueNumberScope(localAvail[DI->getIDom()->getBlock()]); 2203 else 2204 localAvail[DI->getBlock()] = new ValueNumberScope(0); 2205 } 2206 2207 // Top-down walk of the dominator tree 2208 bool Changed = false; 2209#if 0 2210 // Needed for value numbering with phi construction to work. 2211 ReversePostOrderTraversal<Function*> RPOT(&F); 2212 for (ReversePostOrderTraversal<Function*>::rpo_iterator RI = RPOT.begin(), 2213 RE = RPOT.end(); RI != RE; ++RI) 2214 Changed |= processBlock(*RI); 2215#else 2216 for (df_iterator<DomTreeNode*> DI = df_begin(DT->getRootNode()), 2217 DE = df_end(DT->getRootNode()); DI != DE; ++DI) 2218 Changed |= processBlock(DI->getBlock()); 2219#endif 2220 2221 return Changed; 2222} 2223 2224void GVN::cleanupGlobalSets() { 2225 VN.clear(); 2226 2227 for (DenseMap<BasicBlock*, ValueNumberScope*>::iterator 2228 I = localAvail.begin(), E = localAvail.end(); I != E; ++I) 2229 delete I->second; 2230 localAvail.clear(); 2231} 2232 2233/// verifyRemoved - Verify that the specified instruction does not occur in our 2234/// internal data structures. 2235void GVN::verifyRemoved(const Instruction *Inst) const { 2236 VN.verifyRemoved(Inst); 2237 2238 // Walk through the value number scope to make sure the instruction isn't 2239 // ferreted away in it. 2240 for (DenseMap<BasicBlock*, ValueNumberScope*>::const_iterator 2241 I = localAvail.begin(), E = localAvail.end(); I != E; ++I) { 2242 const ValueNumberScope *VNS = I->second; 2243 2244 while (VNS) { 2245 for (DenseMap<uint32_t, Value*>::const_iterator 2246 II = VNS->table.begin(), IE = VNS->table.end(); II != IE; ++II) { 2247 assert(II->second != Inst && "Inst still in value numbering scope!"); 2248 } 2249 2250 VNS = VNS->parent; 2251 } 2252 } 2253} 2254