GVN.cpp revision adb6f226714dcfae363f51b453c4590b0f42da5e
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)); 63static cl::opt<bool> EnableFullLoadPRE("enable-full-load-pre", cl::init(false)); 64 65//===----------------------------------------------------------------------===// 66// ValueTable Class 67//===----------------------------------------------------------------------===// 68 69/// This class holds the mapping between values and value numbers. It is used 70/// as an efficient mechanism to determine the expression-wise equivalence of 71/// two values. 72namespace { 73 struct Expression { 74 enum ExpressionOpcode { 75 ADD = Instruction::Add, 76 FADD = Instruction::FAdd, 77 SUB = Instruction::Sub, 78 FSUB = Instruction::FSub, 79 MUL = Instruction::Mul, 80 FMUL = Instruction::FMul, 81 UDIV = Instruction::UDiv, 82 SDIV = Instruction::SDiv, 83 FDIV = Instruction::FDiv, 84 UREM = Instruction::URem, 85 SREM = Instruction::SRem, 86 FREM = Instruction::FRem, 87 SHL = Instruction::Shl, 88 LSHR = Instruction::LShr, 89 ASHR = Instruction::AShr, 90 AND = Instruction::And, 91 OR = Instruction::Or, 92 XOR = Instruction::Xor, 93 TRUNC = Instruction::Trunc, 94 ZEXT = Instruction::ZExt, 95 SEXT = Instruction::SExt, 96 FPTOUI = Instruction::FPToUI, 97 FPTOSI = Instruction::FPToSI, 98 UITOFP = Instruction::UIToFP, 99 SITOFP = Instruction::SIToFP, 100 FPTRUNC = Instruction::FPTrunc, 101 FPEXT = Instruction::FPExt, 102 PTRTOINT = Instruction::PtrToInt, 103 INTTOPTR = Instruction::IntToPtr, 104 BITCAST = Instruction::BitCast, 105 ICMPEQ, ICMPNE, ICMPUGT, ICMPUGE, ICMPULT, ICMPULE, 106 ICMPSGT, ICMPSGE, ICMPSLT, ICMPSLE, FCMPOEQ, 107 FCMPOGT, FCMPOGE, FCMPOLT, FCMPOLE, FCMPONE, 108 FCMPORD, FCMPUNO, FCMPUEQ, FCMPUGT, FCMPUGE, 109 FCMPULT, FCMPULE, FCMPUNE, EXTRACT, INSERT, 110 SHUFFLE, SELECT, GEP, CALL, CONSTANT, 111 INSERTVALUE, EXTRACTVALUE, EMPTY, TOMBSTONE }; 112 113 ExpressionOpcode opcode; 114 const Type* type; 115 SmallVector<uint32_t, 4> varargs; 116 Value *function; 117 118 Expression() { } 119 Expression(ExpressionOpcode o) : opcode(o) { } 120 121 bool operator==(const Expression &other) const { 122 if (opcode != other.opcode) 123 return false; 124 else if (opcode == EMPTY || opcode == TOMBSTONE) 125 return true; 126 else if (type != other.type) 127 return false; 128 else if (function != other.function) 129 return false; 130 else { 131 if (varargs.size() != other.varargs.size()) 132 return false; 133 134 for (size_t i = 0; i < varargs.size(); ++i) 135 if (varargs[i] != other.varargs[i]) 136 return false; 137 138 return true; 139 } 140 } 141 142 bool operator!=(const Expression &other) const { 143 return !(*this == other); 144 } 145 }; 146 147 class ValueTable { 148 private: 149 DenseMap<Value*, uint32_t> valueNumbering; 150 DenseMap<Expression, uint32_t> expressionNumbering; 151 AliasAnalysis* AA; 152 MemoryDependenceAnalysis* MD; 153 DominatorTree* DT; 154 155 uint32_t nextValueNumber; 156 157 Expression::ExpressionOpcode getOpcode(CmpInst* C); 158 Expression create_expression(BinaryOperator* BO); 159 Expression create_expression(CmpInst* C); 160 Expression create_expression(ShuffleVectorInst* V); 161 Expression create_expression(ExtractElementInst* C); 162 Expression create_expression(InsertElementInst* V); 163 Expression create_expression(SelectInst* V); 164 Expression create_expression(CastInst* C); 165 Expression create_expression(GetElementPtrInst* G); 166 Expression create_expression(CallInst* C); 167 Expression create_expression(Constant* C); 168 Expression create_expression(ExtractValueInst* C); 169 Expression create_expression(InsertValueInst* C); 170 171 uint32_t lookup_or_add_call(CallInst* C); 172 public: 173 ValueTable() : nextValueNumber(1) { } 174 uint32_t lookup_or_add(Value *V); 175 uint32_t lookup(Value *V) const; 176 void add(Value *V, uint32_t num); 177 void clear(); 178 void erase(Value *v); 179 unsigned size(); 180 void setAliasAnalysis(AliasAnalysis* A) { AA = A; } 181 AliasAnalysis *getAliasAnalysis() const { return AA; } 182 void setMemDep(MemoryDependenceAnalysis* M) { MD = M; } 183 void setDomTree(DominatorTree* D) { DT = D; } 184 uint32_t getNextUnusedValueNumber() { return nextValueNumber; } 185 void verifyRemoved(const Value *) const; 186 }; 187} 188 189namespace llvm { 190template <> struct DenseMapInfo<Expression> { 191 static inline Expression getEmptyKey() { 192 return Expression(Expression::EMPTY); 193 } 194 195 static inline Expression getTombstoneKey() { 196 return Expression(Expression::TOMBSTONE); 197 } 198 199 static unsigned getHashValue(const Expression e) { 200 unsigned hash = e.opcode; 201 202 hash = ((unsigned)((uintptr_t)e.type >> 4) ^ 203 (unsigned)((uintptr_t)e.type >> 9)); 204 205 for (SmallVector<uint32_t, 4>::const_iterator I = e.varargs.begin(), 206 E = e.varargs.end(); I != E; ++I) 207 hash = *I + hash * 37; 208 209 hash = ((unsigned)((uintptr_t)e.function >> 4) ^ 210 (unsigned)((uintptr_t)e.function >> 9)) + 211 hash * 37; 212 213 return hash; 214 } 215 static bool isEqual(const Expression &LHS, const Expression &RHS) { 216 return LHS == RHS; 217 } 218}; 219 220template <> 221struct isPodLike<Expression> { static const bool value = true; }; 222 223} 224 225//===----------------------------------------------------------------------===// 226// ValueTable Internal Functions 227//===----------------------------------------------------------------------===// 228 229Expression::ExpressionOpcode ValueTable::getOpcode(CmpInst* C) { 230 if (isa<ICmpInst>(C)) { 231 switch (C->getPredicate()) { 232 default: // THIS SHOULD NEVER HAPPEN 233 llvm_unreachable("Comparison with unknown predicate?"); 234 case ICmpInst::ICMP_EQ: return Expression::ICMPEQ; 235 case ICmpInst::ICMP_NE: return Expression::ICMPNE; 236 case ICmpInst::ICMP_UGT: return Expression::ICMPUGT; 237 case ICmpInst::ICMP_UGE: return Expression::ICMPUGE; 238 case ICmpInst::ICMP_ULT: return Expression::ICMPULT; 239 case ICmpInst::ICMP_ULE: return Expression::ICMPULE; 240 case ICmpInst::ICMP_SGT: return Expression::ICMPSGT; 241 case ICmpInst::ICMP_SGE: return Expression::ICMPSGE; 242 case ICmpInst::ICMP_SLT: return Expression::ICMPSLT; 243 case ICmpInst::ICMP_SLE: return Expression::ICMPSLE; 244 } 245 } else { 246 switch (C->getPredicate()) { 247 default: // THIS SHOULD NEVER HAPPEN 248 llvm_unreachable("Comparison with unknown predicate?"); 249 case FCmpInst::FCMP_OEQ: return Expression::FCMPOEQ; 250 case FCmpInst::FCMP_OGT: return Expression::FCMPOGT; 251 case FCmpInst::FCMP_OGE: return Expression::FCMPOGE; 252 case FCmpInst::FCMP_OLT: return Expression::FCMPOLT; 253 case FCmpInst::FCMP_OLE: return Expression::FCMPOLE; 254 case FCmpInst::FCMP_ONE: return Expression::FCMPONE; 255 case FCmpInst::FCMP_ORD: return Expression::FCMPORD; 256 case FCmpInst::FCMP_UNO: return Expression::FCMPUNO; 257 case FCmpInst::FCMP_UEQ: return Expression::FCMPUEQ; 258 case FCmpInst::FCMP_UGT: return Expression::FCMPUGT; 259 case FCmpInst::FCMP_UGE: return Expression::FCMPUGE; 260 case FCmpInst::FCMP_ULT: return Expression::FCMPULT; 261 case FCmpInst::FCMP_ULE: return Expression::FCMPULE; 262 case FCmpInst::FCMP_UNE: return Expression::FCMPUNE; 263 } 264 } 265} 266 267Expression ValueTable::create_expression(CallInst* C) { 268 Expression e; 269 270 e.type = C->getType(); 271 e.function = C->getCalledFunction(); 272 e.opcode = Expression::CALL; 273 274 for (CallInst::op_iterator I = C->op_begin()+1, E = C->op_end(); 275 I != E; ++I) 276 e.varargs.push_back(lookup_or_add(*I)); 277 278 return e; 279} 280 281Expression ValueTable::create_expression(BinaryOperator* BO) { 282 Expression e; 283 e.varargs.push_back(lookup_or_add(BO->getOperand(0))); 284 e.varargs.push_back(lookup_or_add(BO->getOperand(1))); 285 e.function = 0; 286 e.type = BO->getType(); 287 e.opcode = static_cast<Expression::ExpressionOpcode>(BO->getOpcode()); 288 289 return e; 290} 291 292Expression ValueTable::create_expression(CmpInst* C) { 293 Expression e; 294 295 e.varargs.push_back(lookup_or_add(C->getOperand(0))); 296 e.varargs.push_back(lookup_or_add(C->getOperand(1))); 297 e.function = 0; 298 e.type = C->getType(); 299 e.opcode = getOpcode(C); 300 301 return e; 302} 303 304Expression ValueTable::create_expression(CastInst* C) { 305 Expression e; 306 307 e.varargs.push_back(lookup_or_add(C->getOperand(0))); 308 e.function = 0; 309 e.type = C->getType(); 310 e.opcode = static_cast<Expression::ExpressionOpcode>(C->getOpcode()); 311 312 return e; 313} 314 315Expression ValueTable::create_expression(ShuffleVectorInst* S) { 316 Expression e; 317 318 e.varargs.push_back(lookup_or_add(S->getOperand(0))); 319 e.varargs.push_back(lookup_or_add(S->getOperand(1))); 320 e.varargs.push_back(lookup_or_add(S->getOperand(2))); 321 e.function = 0; 322 e.type = S->getType(); 323 e.opcode = Expression::SHUFFLE; 324 325 return e; 326} 327 328Expression ValueTable::create_expression(ExtractElementInst* E) { 329 Expression e; 330 331 e.varargs.push_back(lookup_or_add(E->getOperand(0))); 332 e.varargs.push_back(lookup_or_add(E->getOperand(1))); 333 e.function = 0; 334 e.type = E->getType(); 335 e.opcode = Expression::EXTRACT; 336 337 return e; 338} 339 340Expression ValueTable::create_expression(InsertElementInst* I) { 341 Expression e; 342 343 e.varargs.push_back(lookup_or_add(I->getOperand(0))); 344 e.varargs.push_back(lookup_or_add(I->getOperand(1))); 345 e.varargs.push_back(lookup_or_add(I->getOperand(2))); 346 e.function = 0; 347 e.type = I->getType(); 348 e.opcode = Expression::INSERT; 349 350 return e; 351} 352 353Expression ValueTable::create_expression(SelectInst* I) { 354 Expression e; 355 356 e.varargs.push_back(lookup_or_add(I->getCondition())); 357 e.varargs.push_back(lookup_or_add(I->getTrueValue())); 358 e.varargs.push_back(lookup_or_add(I->getFalseValue())); 359 e.function = 0; 360 e.type = I->getType(); 361 e.opcode = Expression::SELECT; 362 363 return e; 364} 365 366Expression ValueTable::create_expression(GetElementPtrInst* G) { 367 Expression e; 368 369 e.varargs.push_back(lookup_or_add(G->getPointerOperand())); 370 e.function = 0; 371 e.type = G->getType(); 372 e.opcode = Expression::GEP; 373 374 for (GetElementPtrInst::op_iterator I = G->idx_begin(), E = G->idx_end(); 375 I != E; ++I) 376 e.varargs.push_back(lookup_or_add(*I)); 377 378 return e; 379} 380 381Expression ValueTable::create_expression(ExtractValueInst* E) { 382 Expression e; 383 384 e.varargs.push_back(lookup_or_add(E->getAggregateOperand())); 385 for (ExtractValueInst::idx_iterator II = E->idx_begin(), IE = E->idx_end(); 386 II != IE; ++II) 387 e.varargs.push_back(*II); 388 e.function = 0; 389 e.type = E->getType(); 390 e.opcode = Expression::EXTRACTVALUE; 391 392 return e; 393} 394 395Expression ValueTable::create_expression(InsertValueInst* E) { 396 Expression e; 397 398 e.varargs.push_back(lookup_or_add(E->getAggregateOperand())); 399 e.varargs.push_back(lookup_or_add(E->getInsertedValueOperand())); 400 for (InsertValueInst::idx_iterator II = E->idx_begin(), IE = E->idx_end(); 401 II != IE; ++II) 402 e.varargs.push_back(*II); 403 e.function = 0; 404 e.type = E->getType(); 405 e.opcode = Expression::INSERTVALUE; 406 407 return e; 408} 409 410//===----------------------------------------------------------------------===// 411// ValueTable External Functions 412//===----------------------------------------------------------------------===// 413 414/// add - Insert a value into the table with a specified value number. 415void ValueTable::add(Value *V, uint32_t num) { 416 valueNumbering.insert(std::make_pair(V, num)); 417} 418 419uint32_t ValueTable::lookup_or_add_call(CallInst* C) { 420 if (AA->doesNotAccessMemory(C)) { 421 Expression exp = create_expression(C); 422 uint32_t& e = expressionNumbering[exp]; 423 if (!e) e = nextValueNumber++; 424 valueNumbering[C] = e; 425 return e; 426 } else if (AA->onlyReadsMemory(C)) { 427 Expression exp = create_expression(C); 428 uint32_t& e = expressionNumbering[exp]; 429 if (!e) { 430 e = nextValueNumber++; 431 valueNumbering[C] = e; 432 return e; 433 } 434 if (!MD) { 435 e = nextValueNumber++; 436 valueNumbering[C] = e; 437 return e; 438 } 439 440 MemDepResult local_dep = MD->getDependency(C); 441 442 if (!local_dep.isDef() && !local_dep.isNonLocal()) { 443 valueNumbering[C] = nextValueNumber; 444 return nextValueNumber++; 445 } 446 447 if (local_dep.isDef()) { 448 CallInst* local_cdep = cast<CallInst>(local_dep.getInst()); 449 450 if (local_cdep->getNumOperands() != C->getNumOperands()) { 451 valueNumbering[C] = nextValueNumber; 452 return nextValueNumber++; 453 } 454 455 for (unsigned i = 1; i < C->getNumOperands(); ++i) { 456 uint32_t c_vn = lookup_or_add(C->getOperand(i)); 457 uint32_t cd_vn = lookup_or_add(local_cdep->getOperand(i)); 458 if (c_vn != cd_vn) { 459 valueNumbering[C] = nextValueNumber; 460 return nextValueNumber++; 461 } 462 } 463 464 uint32_t v = lookup_or_add(local_cdep); 465 valueNumbering[C] = v; 466 return v; 467 } 468 469 // Non-local case. 470 const MemoryDependenceAnalysis::NonLocalDepInfo &deps = 471 MD->getNonLocalCallDependency(CallSite(C)); 472 // FIXME: call/call dependencies for readonly calls should return def, not 473 // clobber! Move the checking logic to MemDep! 474 CallInst* cdep = 0; 475 476 // Check to see if we have a single dominating call instruction that is 477 // identical to C. 478 for (unsigned i = 0, e = deps.size(); i != e; ++i) { 479 const NonLocalDepEntry *I = &deps[i]; 480 // Ignore non-local dependencies. 481 if (I->getResult().isNonLocal()) 482 continue; 483 484 // We don't handle non-depedencies. If we already have a call, reject 485 // instruction dependencies. 486 if (I->getResult().isClobber() || cdep != 0) { 487 cdep = 0; 488 break; 489 } 490 491 CallInst *NonLocalDepCall = dyn_cast<CallInst>(I->getResult().getInst()); 492 // FIXME: All duplicated with non-local case. 493 if (NonLocalDepCall && DT->properlyDominates(I->getBB(), C->getParent())){ 494 cdep = NonLocalDepCall; 495 continue; 496 } 497 498 cdep = 0; 499 break; 500 } 501 502 if (!cdep) { 503 valueNumbering[C] = nextValueNumber; 504 return nextValueNumber++; 505 } 506 507 if (cdep->getNumOperands() != C->getNumOperands()) { 508 valueNumbering[C] = nextValueNumber; 509 return nextValueNumber++; 510 } 511 for (unsigned i = 1; i < C->getNumOperands(); ++i) { 512 uint32_t c_vn = lookup_or_add(C->getOperand(i)); 513 uint32_t cd_vn = lookup_or_add(cdep->getOperand(i)); 514 if (c_vn != cd_vn) { 515 valueNumbering[C] = nextValueNumber; 516 return nextValueNumber++; 517 } 518 } 519 520 uint32_t v = lookup_or_add(cdep); 521 valueNumbering[C] = v; 522 return v; 523 524 } else { 525 valueNumbering[C] = nextValueNumber; 526 return nextValueNumber++; 527 } 528} 529 530/// lookup_or_add - Returns the value number for the specified value, assigning 531/// it a new number if it did not have one before. 532uint32_t ValueTable::lookup_or_add(Value *V) { 533 DenseMap<Value*, uint32_t>::iterator VI = valueNumbering.find(V); 534 if (VI != valueNumbering.end()) 535 return VI->second; 536 537 if (!isa<Instruction>(V)) { 538 valueNumbering[V] = nextValueNumber; 539 return nextValueNumber++; 540 } 541 542 Instruction* I = cast<Instruction>(V); 543 Expression exp; 544 switch (I->getOpcode()) { 545 case Instruction::Call: 546 return lookup_or_add_call(cast<CallInst>(I)); 547 case Instruction::Add: 548 case Instruction::FAdd: 549 case Instruction::Sub: 550 case Instruction::FSub: 551 case Instruction::Mul: 552 case Instruction::FMul: 553 case Instruction::UDiv: 554 case Instruction::SDiv: 555 case Instruction::FDiv: 556 case Instruction::URem: 557 case Instruction::SRem: 558 case Instruction::FRem: 559 case Instruction::Shl: 560 case Instruction::LShr: 561 case Instruction::AShr: 562 case Instruction::And: 563 case Instruction::Or : 564 case Instruction::Xor: 565 exp = create_expression(cast<BinaryOperator>(I)); 566 break; 567 case Instruction::ICmp: 568 case Instruction::FCmp: 569 exp = create_expression(cast<CmpInst>(I)); 570 break; 571 case Instruction::Trunc: 572 case Instruction::ZExt: 573 case Instruction::SExt: 574 case Instruction::FPToUI: 575 case Instruction::FPToSI: 576 case Instruction::UIToFP: 577 case Instruction::SIToFP: 578 case Instruction::FPTrunc: 579 case Instruction::FPExt: 580 case Instruction::PtrToInt: 581 case Instruction::IntToPtr: 582 case Instruction::BitCast: 583 exp = create_expression(cast<CastInst>(I)); 584 break; 585 case Instruction::Select: 586 exp = create_expression(cast<SelectInst>(I)); 587 break; 588 case Instruction::ExtractElement: 589 exp = create_expression(cast<ExtractElementInst>(I)); 590 break; 591 case Instruction::InsertElement: 592 exp = create_expression(cast<InsertElementInst>(I)); 593 break; 594 case Instruction::ShuffleVector: 595 exp = create_expression(cast<ShuffleVectorInst>(I)); 596 break; 597 case Instruction::ExtractValue: 598 exp = create_expression(cast<ExtractValueInst>(I)); 599 break; 600 case Instruction::InsertValue: 601 exp = create_expression(cast<InsertValueInst>(I)); 602 break; 603 case Instruction::GetElementPtr: 604 exp = create_expression(cast<GetElementPtrInst>(I)); 605 break; 606 default: 607 valueNumbering[V] = nextValueNumber; 608 return nextValueNumber++; 609 } 610 611 uint32_t& e = expressionNumbering[exp]; 612 if (!e) e = nextValueNumber++; 613 valueNumbering[V] = e; 614 return e; 615} 616 617/// lookup - Returns the value number of the specified value. Fails if 618/// the value has not yet been numbered. 619uint32_t ValueTable::lookup(Value *V) const { 620 DenseMap<Value*, uint32_t>::const_iterator VI = valueNumbering.find(V); 621 assert(VI != valueNumbering.end() && "Value not numbered?"); 622 return VI->second; 623} 624 625/// clear - Remove all entries from the ValueTable 626void ValueTable::clear() { 627 valueNumbering.clear(); 628 expressionNumbering.clear(); 629 nextValueNumber = 1; 630} 631 632/// erase - Remove a value from the value numbering 633void ValueTable::erase(Value *V) { 634 valueNumbering.erase(V); 635} 636 637/// verifyRemoved - Verify that the value is removed from all internal data 638/// structures. 639void ValueTable::verifyRemoved(const Value *V) const { 640 for (DenseMap<Value*, uint32_t>::const_iterator 641 I = valueNumbering.begin(), E = valueNumbering.end(); I != E; ++I) { 642 assert(I->first != V && "Inst still occurs in value numbering map!"); 643 } 644} 645 646//===----------------------------------------------------------------------===// 647// GVN Pass 648//===----------------------------------------------------------------------===// 649 650namespace { 651 struct ValueNumberScope { 652 ValueNumberScope* parent; 653 DenseMap<uint32_t, Value*> table; 654 655 ValueNumberScope(ValueNumberScope* p) : parent(p) { } 656 }; 657} 658 659namespace { 660 661 class GVN : public FunctionPass { 662 bool runOnFunction(Function &F); 663 public: 664 static char ID; // Pass identification, replacement for typeid 665 explicit GVN(bool nopre = false, bool noloads = false) 666 : FunctionPass(&ID), NoPRE(nopre), NoLoads(noloads), MD(0) { } 667 668 private: 669 bool NoPRE; 670 bool NoLoads; 671 MemoryDependenceAnalysis *MD; 672 DominatorTree *DT; 673 674 ValueTable VN; 675 DenseMap<BasicBlock*, ValueNumberScope*> localAvail; 676 677 // This transformation requires dominator postdominator info 678 virtual void getAnalysisUsage(AnalysisUsage &AU) const { 679 AU.addRequired<DominatorTree>(); 680 if (!NoLoads) 681 AU.addRequired<MemoryDependenceAnalysis>(); 682 AU.addRequired<AliasAnalysis>(); 683 684 AU.addPreserved<DominatorTree>(); 685 AU.addPreserved<AliasAnalysis>(); 686 } 687 688 // Helper fuctions 689 // FIXME: eliminate or document these better 690 bool processLoad(LoadInst* L, 691 SmallVectorImpl<Instruction*> &toErase); 692 bool processInstruction(Instruction *I, 693 SmallVectorImpl<Instruction*> &toErase); 694 bool processNonLocalLoad(LoadInst* L, 695 SmallVectorImpl<Instruction*> &toErase); 696 bool processBlock(BasicBlock *BB); 697 void dump(DenseMap<uint32_t, Value*>& d); 698 bool iterateOnFunction(Function &F); 699 Value *CollapsePhi(PHINode* p); 700 bool performPRE(Function& F); 701 Value *lookupNumber(BasicBlock *BB, uint32_t num); 702 void cleanupGlobalSets(); 703 void verifyRemoved(const Instruction *I) const; 704 }; 705 706 char GVN::ID = 0; 707} 708 709// createGVNPass - The public interface to this file... 710FunctionPass *llvm::createGVNPass(bool NoPRE, bool NoLoads) { 711 return new GVN(NoPRE, NoLoads); 712} 713 714static RegisterPass<GVN> X("gvn", 715 "Global Value Numbering"); 716 717void GVN::dump(DenseMap<uint32_t, Value*>& d) { 718 errs() << "{\n"; 719 for (DenseMap<uint32_t, Value*>::iterator I = d.begin(), 720 E = d.end(); I != E; ++I) { 721 errs() << I->first << "\n"; 722 I->second->dump(); 723 } 724 errs() << "}\n"; 725} 726 727static bool isSafeReplacement(PHINode* p, Instruction *inst) { 728 if (!isa<PHINode>(inst)) 729 return true; 730 731 for (Instruction::use_iterator UI = p->use_begin(), E = p->use_end(); 732 UI != E; ++UI) 733 if (PHINode* use_phi = dyn_cast<PHINode>(UI)) 734 if (use_phi->getParent() == inst->getParent()) 735 return false; 736 737 return true; 738} 739 740Value *GVN::CollapsePhi(PHINode *PN) { 741 Value *ConstVal = PN->hasConstantValue(DT); 742 if (!ConstVal) return 0; 743 744 Instruction *Inst = dyn_cast<Instruction>(ConstVal); 745 if (!Inst) 746 return ConstVal; 747 748 if (DT->dominates(Inst, PN)) 749 if (isSafeReplacement(PN, Inst)) 750 return Inst; 751 return 0; 752} 753 754/// IsValueFullyAvailableInBlock - Return true if we can prove that the value 755/// we're analyzing is fully available in the specified block. As we go, keep 756/// track of which blocks we know are fully alive in FullyAvailableBlocks. This 757/// map is actually a tri-state map with the following values: 758/// 0) we know the block *is not* fully available. 759/// 1) we know the block *is* fully available. 760/// 2) we do not know whether the block is fully available or not, but we are 761/// currently speculating that it will be. 762/// 3) we are speculating for this block and have used that to speculate for 763/// other blocks. 764static bool IsValueFullyAvailableInBlock(BasicBlock *BB, 765 DenseMap<BasicBlock*, char> &FullyAvailableBlocks) { 766 // Optimistically assume that the block is fully available and check to see 767 // if we already know about this block in one lookup. 768 std::pair<DenseMap<BasicBlock*, char>::iterator, char> IV = 769 FullyAvailableBlocks.insert(std::make_pair(BB, 2)); 770 771 // If the entry already existed for this block, return the precomputed value. 772 if (!IV.second) { 773 // If this is a speculative "available" value, mark it as being used for 774 // speculation of other blocks. 775 if (IV.first->second == 2) 776 IV.first->second = 3; 777 return IV.first->second != 0; 778 } 779 780 // Otherwise, see if it is fully available in all predecessors. 781 pred_iterator PI = pred_begin(BB), PE = pred_end(BB); 782 783 // If this block has no predecessors, it isn't live-in here. 784 if (PI == PE) 785 goto SpeculationFailure; 786 787 for (; PI != PE; ++PI) 788 // If the value isn't fully available in one of our predecessors, then it 789 // isn't fully available in this block either. Undo our previous 790 // optimistic assumption and bail out. 791 if (!IsValueFullyAvailableInBlock(*PI, FullyAvailableBlocks)) 792 goto SpeculationFailure; 793 794 return true; 795 796// SpeculationFailure - If we get here, we found out that this is not, after 797// all, a fully-available block. We have a problem if we speculated on this and 798// used the speculation to mark other blocks as available. 799SpeculationFailure: 800 char &BBVal = FullyAvailableBlocks[BB]; 801 802 // If we didn't speculate on this, just return with it set to false. 803 if (BBVal == 2) { 804 BBVal = 0; 805 return false; 806 } 807 808 // If we did speculate on this value, we could have blocks set to 1 that are 809 // incorrect. Walk the (transitive) successors of this block and mark them as 810 // 0 if set to one. 811 SmallVector<BasicBlock*, 32> BBWorklist; 812 BBWorklist.push_back(BB); 813 814 do { 815 BasicBlock *Entry = BBWorklist.pop_back_val(); 816 // Note that this sets blocks to 0 (unavailable) if they happen to not 817 // already be in FullyAvailableBlocks. This is safe. 818 char &EntryVal = FullyAvailableBlocks[Entry]; 819 if (EntryVal == 0) continue; // Already unavailable. 820 821 // Mark as unavailable. 822 EntryVal = 0; 823 824 for (succ_iterator I = succ_begin(Entry), E = succ_end(Entry); I != E; ++I) 825 BBWorklist.push_back(*I); 826 } while (!BBWorklist.empty()); 827 828 return false; 829} 830 831 832/// CanCoerceMustAliasedValueToLoad - Return true if 833/// CoerceAvailableValueToLoadType will succeed. 834static bool CanCoerceMustAliasedValueToLoad(Value *StoredVal, 835 const Type *LoadTy, 836 const TargetData &TD) { 837 // If the loaded or stored value is an first class array or struct, don't try 838 // to transform them. We need to be able to bitcast to integer. 839 if (LoadTy->isStructTy() || LoadTy->isArrayTy() || 840 StoredVal->getType()->isStructTy() || 841 StoredVal->getType()->isArrayTy()) 842 return false; 843 844 // The store has to be at least as big as the load. 845 if (TD.getTypeSizeInBits(StoredVal->getType()) < 846 TD.getTypeSizeInBits(LoadTy)) 847 return false; 848 849 return true; 850} 851 852 853/// CoerceAvailableValueToLoadType - If we saw a store of a value to memory, and 854/// then a load from a must-aliased pointer of a different type, try to coerce 855/// the stored value. LoadedTy is the type of the load we want to replace and 856/// InsertPt is the place to insert new instructions. 857/// 858/// If we can't do it, return null. 859static Value *CoerceAvailableValueToLoadType(Value *StoredVal, 860 const Type *LoadedTy, 861 Instruction *InsertPt, 862 const TargetData &TD) { 863 if (!CanCoerceMustAliasedValueToLoad(StoredVal, LoadedTy, TD)) 864 return 0; 865 866 const Type *StoredValTy = StoredVal->getType(); 867 868 uint64_t StoreSize = TD.getTypeSizeInBits(StoredValTy); 869 uint64_t LoadSize = TD.getTypeSizeInBits(LoadedTy); 870 871 // If the store and reload are the same size, we can always reuse it. 872 if (StoreSize == LoadSize) { 873 if (StoredValTy->isPointerTy() && LoadedTy->isPointerTy()) { 874 // Pointer to Pointer -> use bitcast. 875 return new BitCastInst(StoredVal, LoadedTy, "", InsertPt); 876 } 877 878 // Convert source pointers to integers, which can be bitcast. 879 if (StoredValTy->isPointerTy()) { 880 StoredValTy = TD.getIntPtrType(StoredValTy->getContext()); 881 StoredVal = new PtrToIntInst(StoredVal, StoredValTy, "", InsertPt); 882 } 883 884 const Type *TypeToCastTo = LoadedTy; 885 if (TypeToCastTo->isPointerTy()) 886 TypeToCastTo = TD.getIntPtrType(StoredValTy->getContext()); 887 888 if (StoredValTy != TypeToCastTo) 889 StoredVal = new BitCastInst(StoredVal, TypeToCastTo, "", InsertPt); 890 891 // Cast to pointer if the load needs a pointer type. 892 if (LoadedTy->isPointerTy()) 893 StoredVal = new IntToPtrInst(StoredVal, LoadedTy, "", InsertPt); 894 895 return StoredVal; 896 } 897 898 // If the loaded value is smaller than the available value, then we can 899 // extract out a piece from it. If the available value is too small, then we 900 // can't do anything. 901 assert(StoreSize >= LoadSize && "CanCoerceMustAliasedValueToLoad fail"); 902 903 // Convert source pointers to integers, which can be manipulated. 904 if (StoredValTy->isPointerTy()) { 905 StoredValTy = TD.getIntPtrType(StoredValTy->getContext()); 906 StoredVal = new PtrToIntInst(StoredVal, StoredValTy, "", InsertPt); 907 } 908 909 // Convert vectors and fp to integer, which can be manipulated. 910 if (!StoredValTy->isIntegerTy()) { 911 StoredValTy = IntegerType::get(StoredValTy->getContext(), StoreSize); 912 StoredVal = new BitCastInst(StoredVal, StoredValTy, "", InsertPt); 913 } 914 915 // If this is a big-endian system, we need to shift the value down to the low 916 // bits so that a truncate will work. 917 if (TD.isBigEndian()) { 918 Constant *Val = ConstantInt::get(StoredVal->getType(), StoreSize-LoadSize); 919 StoredVal = BinaryOperator::CreateLShr(StoredVal, Val, "tmp", InsertPt); 920 } 921 922 // Truncate the integer to the right size now. 923 const Type *NewIntTy = IntegerType::get(StoredValTy->getContext(), LoadSize); 924 StoredVal = new TruncInst(StoredVal, NewIntTy, "trunc", InsertPt); 925 926 if (LoadedTy == NewIntTy) 927 return StoredVal; 928 929 // If the result is a pointer, inttoptr. 930 if (LoadedTy->isPointerTy()) 931 return new IntToPtrInst(StoredVal, LoadedTy, "inttoptr", InsertPt); 932 933 // Otherwise, bitcast. 934 return new BitCastInst(StoredVal, LoadedTy, "bitcast", InsertPt); 935} 936 937/// GetBaseWithConstantOffset - Analyze the specified pointer to see if it can 938/// be expressed as a base pointer plus a constant offset. Return the base and 939/// offset to the caller. 940static Value *GetBaseWithConstantOffset(Value *Ptr, int64_t &Offset, 941 const TargetData &TD) { 942 Operator *PtrOp = dyn_cast<Operator>(Ptr); 943 if (PtrOp == 0) return Ptr; 944 945 // Just look through bitcasts. 946 if (PtrOp->getOpcode() == Instruction::BitCast) 947 return GetBaseWithConstantOffset(PtrOp->getOperand(0), Offset, TD); 948 949 // If this is a GEP with constant indices, we can look through it. 950 GEPOperator *GEP = dyn_cast<GEPOperator>(PtrOp); 951 if (GEP == 0 || !GEP->hasAllConstantIndices()) return Ptr; 952 953 gep_type_iterator GTI = gep_type_begin(GEP); 954 for (User::op_iterator I = GEP->idx_begin(), E = GEP->idx_end(); I != E; 955 ++I, ++GTI) { 956 ConstantInt *OpC = cast<ConstantInt>(*I); 957 if (OpC->isZero()) continue; 958 959 // Handle a struct and array indices which add their offset to the pointer. 960 if (const StructType *STy = dyn_cast<StructType>(*GTI)) { 961 Offset += TD.getStructLayout(STy)->getElementOffset(OpC->getZExtValue()); 962 } else { 963 uint64_t Size = TD.getTypeAllocSize(GTI.getIndexedType()); 964 Offset += OpC->getSExtValue()*Size; 965 } 966 } 967 968 // Re-sign extend from the pointer size if needed to get overflow edge cases 969 // right. 970 unsigned PtrSize = TD.getPointerSizeInBits(); 971 if (PtrSize < 64) 972 Offset = (Offset << (64-PtrSize)) >> (64-PtrSize); 973 974 return GetBaseWithConstantOffset(GEP->getPointerOperand(), Offset, TD); 975} 976 977 978/// AnalyzeLoadFromClobberingWrite - This function is called when we have a 979/// memdep query of a load that ends up being a clobbering memory write (store, 980/// memset, memcpy, memmove). This means that the write *may* provide bits used 981/// by the load but we can't be sure because the pointers don't mustalias. 982/// 983/// Check this case to see if there is anything more we can do before we give 984/// up. This returns -1 if we have to give up, or a byte number in the stored 985/// value of the piece that feeds the load. 986static int AnalyzeLoadFromClobberingWrite(const Type *LoadTy, Value *LoadPtr, 987 Value *WritePtr, 988 uint64_t WriteSizeInBits, 989 const TargetData &TD) { 990 // If the loaded or stored value is an first class array or struct, don't try 991 // to transform them. We need to be able to bitcast to integer. 992 if (LoadTy->isStructTy() || LoadTy->isArrayTy()) 993 return -1; 994 995 int64_t StoreOffset = 0, LoadOffset = 0; 996 Value *StoreBase = GetBaseWithConstantOffset(WritePtr, StoreOffset, TD); 997 Value *LoadBase = 998 GetBaseWithConstantOffset(LoadPtr, LoadOffset, TD); 999 if (StoreBase != LoadBase) 1000 return -1; 1001 1002 // If the load and store are to the exact same address, they should have been 1003 // a must alias. AA must have gotten confused. 1004 // FIXME: Study to see if/when this happens. 1005 if (LoadOffset == StoreOffset) { 1006#if 0 1007 dbgs() << "STORE/LOAD DEP WITH COMMON POINTER MISSED:\n" 1008 << "Base = " << *StoreBase << "\n" 1009 << "Store Ptr = " << *WritePtr << "\n" 1010 << "Store Offs = " << StoreOffset << "\n" 1011 << "Load Ptr = " << *LoadPtr << "\n"; 1012 abort(); 1013#endif 1014 return -1; 1015 } 1016 1017 // If the load and store don't overlap at all, the store doesn't provide 1018 // anything to the load. In this case, they really don't alias at all, AA 1019 // must have gotten confused. 1020 // FIXME: Investigate cases where this bails out, e.g. rdar://7238614. Then 1021 // remove this check, as it is duplicated with what we have below. 1022 uint64_t LoadSize = TD.getTypeSizeInBits(LoadTy); 1023 1024 if ((WriteSizeInBits & 7) | (LoadSize & 7)) 1025 return -1; 1026 uint64_t StoreSize = WriteSizeInBits >> 3; // Convert to bytes. 1027 LoadSize >>= 3; 1028 1029 1030 bool isAAFailure = false; 1031 if (StoreOffset < LoadOffset) { 1032 isAAFailure = StoreOffset+int64_t(StoreSize) <= LoadOffset; 1033 } else { 1034 isAAFailure = LoadOffset+int64_t(LoadSize) <= StoreOffset; 1035 } 1036 if (isAAFailure) { 1037#if 0 1038 dbgs() << "STORE LOAD DEP WITH COMMON BASE:\n" 1039 << "Base = " << *StoreBase << "\n" 1040 << "Store Ptr = " << *WritePtr << "\n" 1041 << "Store Offs = " << StoreOffset << "\n" 1042 << "Load Ptr = " << *LoadPtr << "\n"; 1043 abort(); 1044#endif 1045 return -1; 1046 } 1047 1048 // If the Load isn't completely contained within the stored bits, we don't 1049 // have all the bits to feed it. We could do something crazy in the future 1050 // (issue a smaller load then merge the bits in) but this seems unlikely to be 1051 // valuable. 1052 if (StoreOffset > LoadOffset || 1053 StoreOffset+StoreSize < LoadOffset+LoadSize) 1054 return -1; 1055 1056 // Okay, we can do this transformation. Return the number of bytes into the 1057 // store that the load is. 1058 return LoadOffset-StoreOffset; 1059} 1060 1061/// AnalyzeLoadFromClobberingStore - This function is called when we have a 1062/// memdep query of a load that ends up being a clobbering store. 1063static int AnalyzeLoadFromClobberingStore(const Type *LoadTy, Value *LoadPtr, 1064 StoreInst *DepSI, 1065 const TargetData &TD) { 1066 // Cannot handle reading from store of first-class aggregate yet. 1067 if (DepSI->getOperand(0)->getType()->isStructTy() || 1068 DepSI->getOperand(0)->getType()->isArrayTy()) 1069 return -1; 1070 1071 Value *StorePtr = DepSI->getPointerOperand(); 1072 uint64_t StoreSize = TD.getTypeSizeInBits(DepSI->getOperand(0)->getType()); 1073 return AnalyzeLoadFromClobberingWrite(LoadTy, LoadPtr, 1074 StorePtr, StoreSize, TD); 1075} 1076 1077static int AnalyzeLoadFromClobberingMemInst(const Type *LoadTy, Value *LoadPtr, 1078 MemIntrinsic *MI, 1079 const TargetData &TD) { 1080 // If the mem operation is a non-constant size, we can't handle it. 1081 ConstantInt *SizeCst = dyn_cast<ConstantInt>(MI->getLength()); 1082 if (SizeCst == 0) return -1; 1083 uint64_t MemSizeInBits = SizeCst->getZExtValue()*8; 1084 1085 // If this is memset, we just need to see if the offset is valid in the size 1086 // of the memset.. 1087 if (MI->getIntrinsicID() == Intrinsic::memset) 1088 return AnalyzeLoadFromClobberingWrite(LoadTy, LoadPtr, MI->getDest(), 1089 MemSizeInBits, TD); 1090 1091 // If we have a memcpy/memmove, the only case we can handle is if this is a 1092 // copy from constant memory. In that case, we can read directly from the 1093 // constant memory. 1094 MemTransferInst *MTI = cast<MemTransferInst>(MI); 1095 1096 Constant *Src = dyn_cast<Constant>(MTI->getSource()); 1097 if (Src == 0) return -1; 1098 1099 GlobalVariable *GV = dyn_cast<GlobalVariable>(Src->getUnderlyingObject()); 1100 if (GV == 0 || !GV->isConstant()) return -1; 1101 1102 // See if the access is within the bounds of the transfer. 1103 int Offset = AnalyzeLoadFromClobberingWrite(LoadTy, LoadPtr, 1104 MI->getDest(), MemSizeInBits, TD); 1105 if (Offset == -1) 1106 return Offset; 1107 1108 // Otherwise, see if we can constant fold a load from the constant with the 1109 // offset applied as appropriate. 1110 Src = ConstantExpr::getBitCast(Src, 1111 llvm::Type::getInt8PtrTy(Src->getContext())); 1112 Constant *OffsetCst = 1113 ConstantInt::get(Type::getInt64Ty(Src->getContext()), (unsigned)Offset); 1114 Src = ConstantExpr::getGetElementPtr(Src, &OffsetCst, 1); 1115 Src = ConstantExpr::getBitCast(Src, PointerType::getUnqual(LoadTy)); 1116 if (ConstantFoldLoadFromConstPtr(Src, &TD)) 1117 return Offset; 1118 return -1; 1119} 1120 1121 1122/// GetStoreValueForLoad - This function is called when we have a 1123/// memdep query of a load that ends up being a clobbering store. This means 1124/// that the store *may* provide bits used by the load but we can't be sure 1125/// because the pointers don't mustalias. Check this case to see if there is 1126/// anything more we can do before we give up. 1127static Value *GetStoreValueForLoad(Value *SrcVal, unsigned Offset, 1128 const Type *LoadTy, 1129 Instruction *InsertPt, const TargetData &TD){ 1130 LLVMContext &Ctx = SrcVal->getType()->getContext(); 1131 1132 uint64_t StoreSize = TD.getTypeSizeInBits(SrcVal->getType())/8; 1133 uint64_t LoadSize = TD.getTypeSizeInBits(LoadTy)/8; 1134 1135 IRBuilder<> Builder(InsertPt->getParent(), InsertPt); 1136 1137 // Compute which bits of the stored value are being used by the load. Convert 1138 // to an integer type to start with. 1139 if (SrcVal->getType()->isPointerTy()) 1140 SrcVal = Builder.CreatePtrToInt(SrcVal, TD.getIntPtrType(Ctx), "tmp"); 1141 if (!SrcVal->getType()->isIntegerTy()) 1142 SrcVal = Builder.CreateBitCast(SrcVal, IntegerType::get(Ctx, StoreSize*8), 1143 "tmp"); 1144 1145 // Shift the bits to the least significant depending on endianness. 1146 unsigned ShiftAmt; 1147 if (TD.isLittleEndian()) 1148 ShiftAmt = Offset*8; 1149 else 1150 ShiftAmt = (StoreSize-LoadSize-Offset)*8; 1151 1152 if (ShiftAmt) 1153 SrcVal = Builder.CreateLShr(SrcVal, ShiftAmt, "tmp"); 1154 1155 if (LoadSize != StoreSize) 1156 SrcVal = Builder.CreateTrunc(SrcVal, IntegerType::get(Ctx, LoadSize*8), 1157 "tmp"); 1158 1159 return CoerceAvailableValueToLoadType(SrcVal, LoadTy, InsertPt, TD); 1160} 1161 1162/// GetMemInstValueForLoad - This function is called when we have a 1163/// memdep query of a load that ends up being a clobbering mem intrinsic. 1164static Value *GetMemInstValueForLoad(MemIntrinsic *SrcInst, unsigned Offset, 1165 const Type *LoadTy, Instruction *InsertPt, 1166 const TargetData &TD){ 1167 LLVMContext &Ctx = LoadTy->getContext(); 1168 uint64_t LoadSize = TD.getTypeSizeInBits(LoadTy)/8; 1169 1170 IRBuilder<> Builder(InsertPt->getParent(), InsertPt); 1171 1172 // We know that this method is only called when the mem transfer fully 1173 // provides the bits for the load. 1174 if (MemSetInst *MSI = dyn_cast<MemSetInst>(SrcInst)) { 1175 // memset(P, 'x', 1234) -> splat('x'), even if x is a variable, and 1176 // independently of what the offset is. 1177 Value *Val = MSI->getValue(); 1178 if (LoadSize != 1) 1179 Val = Builder.CreateZExt(Val, IntegerType::get(Ctx, LoadSize*8)); 1180 1181 Value *OneElt = Val; 1182 1183 // Splat the value out to the right number of bits. 1184 for (unsigned NumBytesSet = 1; NumBytesSet != LoadSize; ) { 1185 // If we can double the number of bytes set, do it. 1186 if (NumBytesSet*2 <= LoadSize) { 1187 Value *ShVal = Builder.CreateShl(Val, NumBytesSet*8); 1188 Val = Builder.CreateOr(Val, ShVal); 1189 NumBytesSet <<= 1; 1190 continue; 1191 } 1192 1193 // Otherwise insert one byte at a time. 1194 Value *ShVal = Builder.CreateShl(Val, 1*8); 1195 Val = Builder.CreateOr(OneElt, ShVal); 1196 ++NumBytesSet; 1197 } 1198 1199 return CoerceAvailableValueToLoadType(Val, LoadTy, InsertPt, TD); 1200 } 1201 1202 // Otherwise, this is a memcpy/memmove from a constant global. 1203 MemTransferInst *MTI = cast<MemTransferInst>(SrcInst); 1204 Constant *Src = cast<Constant>(MTI->getSource()); 1205 1206 // Otherwise, see if we can constant fold a load from the constant with the 1207 // offset applied as appropriate. 1208 Src = ConstantExpr::getBitCast(Src, 1209 llvm::Type::getInt8PtrTy(Src->getContext())); 1210 Constant *OffsetCst = 1211 ConstantInt::get(Type::getInt64Ty(Src->getContext()), (unsigned)Offset); 1212 Src = ConstantExpr::getGetElementPtr(Src, &OffsetCst, 1); 1213 Src = ConstantExpr::getBitCast(Src, PointerType::getUnqual(LoadTy)); 1214 return ConstantFoldLoadFromConstPtr(Src, &TD); 1215} 1216 1217 1218 1219struct AvailableValueInBlock { 1220 /// BB - The basic block in question. 1221 BasicBlock *BB; 1222 enum ValType { 1223 SimpleVal, // A simple offsetted value that is accessed. 1224 MemIntrin // A memory intrinsic which is loaded from. 1225 }; 1226 1227 /// V - The value that is live out of the block. 1228 PointerIntPair<Value *, 1, ValType> Val; 1229 1230 /// Offset - The byte offset in Val that is interesting for the load query. 1231 unsigned Offset; 1232 1233 static AvailableValueInBlock get(BasicBlock *BB, Value *V, 1234 unsigned Offset = 0) { 1235 AvailableValueInBlock Res; 1236 Res.BB = BB; 1237 Res.Val.setPointer(V); 1238 Res.Val.setInt(SimpleVal); 1239 Res.Offset = Offset; 1240 return Res; 1241 } 1242 1243 static AvailableValueInBlock getMI(BasicBlock *BB, MemIntrinsic *MI, 1244 unsigned Offset = 0) { 1245 AvailableValueInBlock Res; 1246 Res.BB = BB; 1247 Res.Val.setPointer(MI); 1248 Res.Val.setInt(MemIntrin); 1249 Res.Offset = Offset; 1250 return Res; 1251 } 1252 1253 bool isSimpleValue() const { return Val.getInt() == SimpleVal; } 1254 Value *getSimpleValue() const { 1255 assert(isSimpleValue() && "Wrong accessor"); 1256 return Val.getPointer(); 1257 } 1258 1259 MemIntrinsic *getMemIntrinValue() const { 1260 assert(!isSimpleValue() && "Wrong accessor"); 1261 return cast<MemIntrinsic>(Val.getPointer()); 1262 } 1263 1264 /// MaterializeAdjustedValue - Emit code into this block to adjust the value 1265 /// defined here to the specified type. This handles various coercion cases. 1266 Value *MaterializeAdjustedValue(const Type *LoadTy, 1267 const TargetData *TD) const { 1268 Value *Res; 1269 if (isSimpleValue()) { 1270 Res = getSimpleValue(); 1271 if (Res->getType() != LoadTy) { 1272 assert(TD && "Need target data to handle type mismatch case"); 1273 Res = GetStoreValueForLoad(Res, Offset, LoadTy, BB->getTerminator(), 1274 *TD); 1275 1276 DEBUG(errs() << "GVN COERCED NONLOCAL VAL:\nOffset: " << Offset << " " 1277 << *getSimpleValue() << '\n' 1278 << *Res << '\n' << "\n\n\n"); 1279 } 1280 } else { 1281 Res = GetMemInstValueForLoad(getMemIntrinValue(), Offset, 1282 LoadTy, BB->getTerminator(), *TD); 1283 DEBUG(errs() << "GVN COERCED NONLOCAL MEM INTRIN:\nOffset: " << Offset 1284 << " " << *getMemIntrinValue() << '\n' 1285 << *Res << '\n' << "\n\n\n"); 1286 } 1287 return Res; 1288 } 1289}; 1290 1291/// ConstructSSAForLoadSet - Given a set of loads specified by ValuesPerBlock, 1292/// construct SSA form, allowing us to eliminate LI. This returns the value 1293/// that should be used at LI's definition site. 1294static Value *ConstructSSAForLoadSet(LoadInst *LI, 1295 SmallVectorImpl<AvailableValueInBlock> &ValuesPerBlock, 1296 const TargetData *TD, 1297 const DominatorTree &DT, 1298 AliasAnalysis *AA) { 1299 // Check for the fully redundant, dominating load case. In this case, we can 1300 // just use the dominating value directly. 1301 if (ValuesPerBlock.size() == 1 && 1302 DT.properlyDominates(ValuesPerBlock[0].BB, LI->getParent())) 1303 return ValuesPerBlock[0].MaterializeAdjustedValue(LI->getType(), TD); 1304 1305 // Otherwise, we have to construct SSA form. 1306 SmallVector<PHINode*, 8> NewPHIs; 1307 SSAUpdater SSAUpdate(&NewPHIs); 1308 SSAUpdate.Initialize(LI); 1309 1310 const Type *LoadTy = LI->getType(); 1311 1312 for (unsigned i = 0, e = ValuesPerBlock.size(); i != e; ++i) { 1313 const AvailableValueInBlock &AV = ValuesPerBlock[i]; 1314 BasicBlock *BB = AV.BB; 1315 1316 if (SSAUpdate.HasValueForBlock(BB)) 1317 continue; 1318 1319 SSAUpdate.AddAvailableValue(BB, AV.MaterializeAdjustedValue(LoadTy, TD)); 1320 } 1321 1322 // Perform PHI construction. 1323 Value *V = SSAUpdate.GetValueInMiddleOfBlock(LI->getParent()); 1324 1325 // If new PHI nodes were created, notify alias analysis. 1326 if (V->getType()->isPointerTy()) 1327 for (unsigned i = 0, e = NewPHIs.size(); i != e; ++i) 1328 AA->copyValue(LI, NewPHIs[i]); 1329 1330 return V; 1331} 1332 1333static bool isLifetimeStart(Instruction *Inst) { 1334 if (IntrinsicInst* II = dyn_cast<IntrinsicInst>(Inst)) 1335 return II->getIntrinsicID() == Intrinsic::lifetime_start; 1336 return false; 1337} 1338 1339/// processNonLocalLoad - Attempt to eliminate a load whose dependencies are 1340/// non-local by performing PHI construction. 1341bool GVN::processNonLocalLoad(LoadInst *LI, 1342 SmallVectorImpl<Instruction*> &toErase) { 1343 // Find the non-local dependencies of the load. 1344 SmallVector<NonLocalDepResult, 64> Deps; 1345 MD->getNonLocalPointerDependency(LI->getOperand(0), true, LI->getParent(), 1346 Deps); 1347 //DEBUG(dbgs() << "INVESTIGATING NONLOCAL LOAD: " 1348 // << Deps.size() << *LI << '\n'); 1349 1350 // If we had to process more than one hundred blocks to find the 1351 // dependencies, this load isn't worth worrying about. Optimizing 1352 // it will be too expensive. 1353 if (Deps.size() > 100) 1354 return false; 1355 1356 // If we had a phi translation failure, we'll have a single entry which is a 1357 // clobber in the current block. Reject this early. 1358 if (Deps.size() == 1 && Deps[0].getResult().isClobber()) { 1359 DEBUG( 1360 dbgs() << "GVN: non-local load "; 1361 WriteAsOperand(dbgs(), LI); 1362 dbgs() << " is clobbered by " << *Deps[0].getResult().getInst() << '\n'; 1363 ); 1364 return false; 1365 } 1366 1367 // Filter out useless results (non-locals, etc). Keep track of the blocks 1368 // where we have a value available in repl, also keep track of whether we see 1369 // dependencies that produce an unknown value for the load (such as a call 1370 // that could potentially clobber the load). 1371 SmallVector<AvailableValueInBlock, 16> ValuesPerBlock; 1372 SmallVector<BasicBlock*, 16> UnavailableBlocks; 1373 1374 const TargetData *TD = 0; 1375 1376 for (unsigned i = 0, e = Deps.size(); i != e; ++i) { 1377 BasicBlock *DepBB = Deps[i].getBB(); 1378 MemDepResult DepInfo = Deps[i].getResult(); 1379 1380 if (DepInfo.isClobber()) { 1381 // The address being loaded in this non-local block may not be the same as 1382 // the pointer operand of the load if PHI translation occurs. Make sure 1383 // to consider the right address. 1384 Value *Address = Deps[i].getAddress(); 1385 1386 // If the dependence is to a store that writes to a superset of the bits 1387 // read by the load, we can extract the bits we need for the load from the 1388 // stored value. 1389 if (StoreInst *DepSI = dyn_cast<StoreInst>(DepInfo.getInst())) { 1390 if (TD == 0) 1391 TD = getAnalysisIfAvailable<TargetData>(); 1392 if (TD && Address) { 1393 int Offset = AnalyzeLoadFromClobberingStore(LI->getType(), Address, 1394 DepSI, *TD); 1395 if (Offset != -1) { 1396 ValuesPerBlock.push_back(AvailableValueInBlock::get(DepBB, 1397 DepSI->getOperand(0), 1398 Offset)); 1399 continue; 1400 } 1401 } 1402 } 1403 1404 // If the clobbering value is a memset/memcpy/memmove, see if we can 1405 // forward a value on from it. 1406 if (MemIntrinsic *DepMI = dyn_cast<MemIntrinsic>(DepInfo.getInst())) { 1407 if (TD == 0) 1408 TD = getAnalysisIfAvailable<TargetData>(); 1409 if (TD && Address) { 1410 int Offset = AnalyzeLoadFromClobberingMemInst(LI->getType(), Address, 1411 DepMI, *TD); 1412 if (Offset != -1) { 1413 ValuesPerBlock.push_back(AvailableValueInBlock::getMI(DepBB, DepMI, 1414 Offset)); 1415 continue; 1416 } 1417 } 1418 } 1419 1420 UnavailableBlocks.push_back(DepBB); 1421 continue; 1422 } 1423 1424 Instruction *DepInst = DepInfo.getInst(); 1425 1426 // Loading the allocation -> undef. 1427 if (isa<AllocaInst>(DepInst) || isMalloc(DepInst) || 1428 // Loading immediately after lifetime begin -> undef. 1429 isLifetimeStart(DepInst)) { 1430 ValuesPerBlock.push_back(AvailableValueInBlock::get(DepBB, 1431 UndefValue::get(LI->getType()))); 1432 continue; 1433 } 1434 1435 if (StoreInst *S = dyn_cast<StoreInst>(DepInst)) { 1436 // Reject loads and stores that are to the same address but are of 1437 // different types if we have to. 1438 if (S->getOperand(0)->getType() != LI->getType()) { 1439 if (TD == 0) 1440 TD = getAnalysisIfAvailable<TargetData>(); 1441 1442 // If the stored value is larger or equal to the loaded value, we can 1443 // reuse it. 1444 if (TD == 0 || !CanCoerceMustAliasedValueToLoad(S->getOperand(0), 1445 LI->getType(), *TD)) { 1446 UnavailableBlocks.push_back(DepBB); 1447 continue; 1448 } 1449 } 1450 1451 ValuesPerBlock.push_back(AvailableValueInBlock::get(DepBB, 1452 S->getOperand(0))); 1453 continue; 1454 } 1455 1456 if (LoadInst *LD = dyn_cast<LoadInst>(DepInst)) { 1457 // If the types mismatch and we can't handle it, reject reuse of the load. 1458 if (LD->getType() != LI->getType()) { 1459 if (TD == 0) 1460 TD = getAnalysisIfAvailable<TargetData>(); 1461 1462 // If the stored value is larger or equal to the loaded value, we can 1463 // reuse it. 1464 if (TD == 0 || !CanCoerceMustAliasedValueToLoad(LD, LI->getType(),*TD)){ 1465 UnavailableBlocks.push_back(DepBB); 1466 continue; 1467 } 1468 } 1469 ValuesPerBlock.push_back(AvailableValueInBlock::get(DepBB, LD)); 1470 continue; 1471 } 1472 1473 UnavailableBlocks.push_back(DepBB); 1474 continue; 1475 } 1476 1477 // If we have no predecessors that produce a known value for this load, exit 1478 // early. 1479 if (ValuesPerBlock.empty()) return false; 1480 1481 // If all of the instructions we depend on produce a known value for this 1482 // load, then it is fully redundant and we can use PHI insertion to compute 1483 // its value. Insert PHIs and remove the fully redundant value now. 1484 if (UnavailableBlocks.empty()) { 1485 DEBUG(dbgs() << "GVN REMOVING NONLOCAL LOAD: " << *LI << '\n'); 1486 1487 // Perform PHI construction. 1488 Value *V = ConstructSSAForLoadSet(LI, ValuesPerBlock, TD, *DT, 1489 VN.getAliasAnalysis()); 1490 LI->replaceAllUsesWith(V); 1491 1492 if (isa<PHINode>(V)) 1493 V->takeName(LI); 1494 if (V->getType()->isPointerTy()) 1495 MD->invalidateCachedPointerInfo(V); 1496 toErase.push_back(LI); 1497 NumGVNLoad++; 1498 return true; 1499 } 1500 1501 if (!EnablePRE || !EnableLoadPRE) 1502 return false; 1503 1504 // Okay, we have *some* definitions of the value. This means that the value 1505 // is available in some of our (transitive) predecessors. Lets think about 1506 // doing PRE of this load. This will involve inserting a new load into the 1507 // predecessor when it's not available. We could do this in general, but 1508 // prefer to not increase code size. As such, we only do this when we know 1509 // that we only have to insert *one* load (which means we're basically moving 1510 // the load, not inserting a new one). 1511 1512 SmallPtrSet<BasicBlock *, 4> Blockers; 1513 for (unsigned i = 0, e = UnavailableBlocks.size(); i != e; ++i) 1514 Blockers.insert(UnavailableBlocks[i]); 1515 1516 // Lets find first basic block with more than one predecessor. Walk backwards 1517 // through predecessors if needed. 1518 BasicBlock *LoadBB = LI->getParent(); 1519 BasicBlock *TmpBB = LoadBB; 1520 1521 bool isSinglePred = false; 1522 bool allSingleSucc = true; 1523 while (TmpBB->getSinglePredecessor()) { 1524 isSinglePred = true; 1525 TmpBB = TmpBB->getSinglePredecessor(); 1526 if (TmpBB == LoadBB) // Infinite (unreachable) loop. 1527 return false; 1528 if (Blockers.count(TmpBB)) 1529 return false; 1530 if (TmpBB->getTerminator()->getNumSuccessors() != 1) 1531 allSingleSucc = false; 1532 } 1533 1534 assert(TmpBB); 1535 LoadBB = TmpBB; 1536 1537 // If we have a repl set with LI itself in it, this means we have a loop where 1538 // at least one of the values is LI. Since this means that we won't be able 1539 // to eliminate LI even if we insert uses in the other predecessors, we will 1540 // end up increasing code size. Reject this by scanning for LI. 1541 if (!EnableFullLoadPRE) { 1542 for (unsigned i = 0, e = ValuesPerBlock.size(); i != e; ++i) 1543 if (ValuesPerBlock[i].isSimpleValue() && 1544 ValuesPerBlock[i].getSimpleValue() == LI) 1545 return false; 1546 } 1547 1548 // FIXME: It is extremely unclear what this loop is doing, other than 1549 // artificially restricting loadpre. 1550 if (isSinglePred) { 1551 bool isHot = false; 1552 for (unsigned i = 0, e = ValuesPerBlock.size(); i != e; ++i) { 1553 const AvailableValueInBlock &AV = ValuesPerBlock[i]; 1554 if (AV.isSimpleValue()) 1555 // "Hot" Instruction is in some loop (because it dominates its dep. 1556 // instruction). 1557 if (Instruction *I = dyn_cast<Instruction>(AV.getSimpleValue())) 1558 if (DT->dominates(LI, I)) { 1559 isHot = true; 1560 break; 1561 } 1562 } 1563 1564 // We are interested only in "hot" instructions. We don't want to do any 1565 // mis-optimizations here. 1566 if (!isHot) 1567 return false; 1568 } 1569 1570 // Check to see how many predecessors have the loaded value fully 1571 // available. 1572 DenseMap<BasicBlock*, Value*> PredLoads; 1573 DenseMap<BasicBlock*, char> FullyAvailableBlocks; 1574 for (unsigned i = 0, e = ValuesPerBlock.size(); i != e; ++i) 1575 FullyAvailableBlocks[ValuesPerBlock[i].BB] = true; 1576 for (unsigned i = 0, e = UnavailableBlocks.size(); i != e; ++i) 1577 FullyAvailableBlocks[UnavailableBlocks[i]] = false; 1578 1579 for (pred_iterator PI = pred_begin(LoadBB), E = pred_end(LoadBB); 1580 PI != E; ++PI) { 1581 BasicBlock *Pred = *PI; 1582 if (IsValueFullyAvailableInBlock(Pred, FullyAvailableBlocks)) { 1583 continue; 1584 } 1585 PredLoads[Pred] = 0; 1586 // We don't currently handle critical edges :( 1587 if (Pred->getTerminator()->getNumSuccessors() != 1) { 1588 DEBUG(dbgs() << "COULD NOT PRE LOAD BECAUSE OF CRITICAL EDGE '" 1589 << Pred->getName() << "': " << *LI << '\n'); 1590 return false; 1591 } 1592 } 1593 1594 // Decide whether PRE is profitable for this load. 1595 unsigned NumUnavailablePreds = PredLoads.size(); 1596 assert(NumUnavailablePreds != 0 && 1597 "Fully available value should be eliminated above!"); 1598 if (!EnableFullLoadPRE) { 1599 // If this load is unavailable in multiple predecessors, reject it. 1600 // FIXME: If we could restructure the CFG, we could make a common pred with 1601 // all the preds that don't have an available LI and insert a new load into 1602 // that one block. 1603 if (NumUnavailablePreds != 1) 1604 return false; 1605 } 1606 1607 // Check if the load can safely be moved to all the unavailable predecessors. 1608 bool CanDoPRE = true; 1609 SmallVector<Instruction*, 8> NewInsts; 1610 for (DenseMap<BasicBlock*, Value*>::iterator I = PredLoads.begin(), 1611 E = PredLoads.end(); I != E; ++I) { 1612 BasicBlock *UnavailablePred = I->first; 1613 1614 // Do PHI translation to get its value in the predecessor if necessary. The 1615 // returned pointer (if non-null) is guaranteed to dominate UnavailablePred. 1616 1617 // If all preds have a single successor, then we know it is safe to insert 1618 // the load on the pred (?!?), so we can insert code to materialize the 1619 // pointer if it is not available. 1620 PHITransAddr Address(LI->getOperand(0), TD); 1621 Value *LoadPtr = 0; 1622 if (allSingleSucc) { 1623 LoadPtr = Address.PHITranslateWithInsertion(LoadBB, UnavailablePred, 1624 *DT, NewInsts); 1625 } else { 1626 Address.PHITranslateValue(LoadBB, UnavailablePred); 1627 LoadPtr = Address.getAddr(); 1628 1629 // Make sure the value is live in the predecessor. 1630 if (Instruction *Inst = dyn_cast_or_null<Instruction>(LoadPtr)) 1631 if (!DT->dominates(Inst->getParent(), UnavailablePred)) 1632 LoadPtr = 0; 1633 } 1634 1635 // If we couldn't find or insert a computation of this phi translated value, 1636 // we fail PRE. 1637 if (LoadPtr == 0) { 1638 DEBUG(dbgs() << "COULDN'T INSERT PHI TRANSLATED VALUE OF: " 1639 << *LI->getOperand(0) << "\n"); 1640 CanDoPRE = false; 1641 break; 1642 } 1643 1644 // Make sure it is valid to move this load here. We have to watch out for: 1645 // @1 = getelementptr (i8* p, ... 1646 // test p and branch if == 0 1647 // load @1 1648 // It is valid to have the getelementptr before the test, even if p can be 0, 1649 // as getelementptr only does address arithmetic. 1650 // If we are not pushing the value through any multiple-successor blocks 1651 // we do not have this case. Otherwise, check that the load is safe to 1652 // put anywhere; this can be improved, but should be conservatively safe. 1653 if (!allSingleSucc && 1654 // FIXME: REEVALUTE THIS. 1655 !isSafeToLoadUnconditionally(LoadPtr, 1656 UnavailablePred->getTerminator(), 1657 LI->getAlignment(), TD)) { 1658 CanDoPRE = false; 1659 break; 1660 } 1661 1662 I->second = LoadPtr; 1663 } 1664 1665 if (!CanDoPRE) { 1666 while (!NewInsts.empty()) 1667 NewInsts.pop_back_val()->eraseFromParent(); 1668 return false; 1669 } 1670 1671 // Okay, we can eliminate this load by inserting a reload in the predecessor 1672 // and using PHI construction to get the value in the other predecessors, do 1673 // it. 1674 DEBUG(dbgs() << "GVN REMOVING PRE LOAD: " << *LI << '\n'); 1675 DEBUG(if (!NewInsts.empty()) 1676 dbgs() << "INSERTED " << NewInsts.size() << " INSTS: " 1677 << *NewInsts.back() << '\n'); 1678 1679 // Assign value numbers to the new instructions. 1680 for (unsigned i = 0, e = NewInsts.size(); i != e; ++i) { 1681 // FIXME: We really _ought_ to insert these value numbers into their 1682 // parent's availability map. However, in doing so, we risk getting into 1683 // ordering issues. If a block hasn't been processed yet, we would be 1684 // marking a value as AVAIL-IN, which isn't what we intend. 1685 VN.lookup_or_add(NewInsts[i]); 1686 } 1687 1688 for (DenseMap<BasicBlock*, Value*>::iterator I = PredLoads.begin(), 1689 E = PredLoads.end(); I != E; ++I) { 1690 BasicBlock *UnavailablePred = I->first; 1691 Value *LoadPtr = I->second; 1692 1693 Value *NewLoad = new LoadInst(LoadPtr, LI->getName()+".pre", false, 1694 LI->getAlignment(), 1695 UnavailablePred->getTerminator()); 1696 1697 // Add the newly created load. 1698 ValuesPerBlock.push_back(AvailableValueInBlock::get(UnavailablePred, 1699 NewLoad)); 1700 } 1701 1702 // Perform PHI construction. 1703 Value *V = ConstructSSAForLoadSet(LI, ValuesPerBlock, TD, *DT, 1704 VN.getAliasAnalysis()); 1705 LI->replaceAllUsesWith(V); 1706 if (isa<PHINode>(V)) 1707 V->takeName(LI); 1708 if (V->getType()->isPointerTy()) 1709 MD->invalidateCachedPointerInfo(V); 1710 toErase.push_back(LI); 1711 NumPRELoad++; 1712 return true; 1713} 1714 1715/// processLoad - Attempt to eliminate a load, first by eliminating it 1716/// locally, and then attempting non-local elimination if that fails. 1717bool GVN::processLoad(LoadInst *L, SmallVectorImpl<Instruction*> &toErase) { 1718 if (!MD) 1719 return false; 1720 1721 if (L->isVolatile()) 1722 return false; 1723 1724 // ... to a pointer that has been loaded from before... 1725 MemDepResult Dep = MD->getDependency(L); 1726 1727 // If the value isn't available, don't do anything! 1728 if (Dep.isClobber()) { 1729 // Check to see if we have something like this: 1730 // store i32 123, i32* %P 1731 // %A = bitcast i32* %P to i8* 1732 // %B = gep i8* %A, i32 1 1733 // %C = load i8* %B 1734 // 1735 // We could do that by recognizing if the clobber instructions are obviously 1736 // a common base + constant offset, and if the previous store (or memset) 1737 // completely covers this load. This sort of thing can happen in bitfield 1738 // access code. 1739 Value *AvailVal = 0; 1740 if (StoreInst *DepSI = dyn_cast<StoreInst>(Dep.getInst())) 1741 if (const TargetData *TD = getAnalysisIfAvailable<TargetData>()) { 1742 int Offset = AnalyzeLoadFromClobberingStore(L->getType(), 1743 L->getPointerOperand(), 1744 DepSI, *TD); 1745 if (Offset != -1) 1746 AvailVal = GetStoreValueForLoad(DepSI->getOperand(0), Offset, 1747 L->getType(), L, *TD); 1748 } 1749 1750 // If the clobbering value is a memset/memcpy/memmove, see if we can forward 1751 // a value on from it. 1752 if (MemIntrinsic *DepMI = dyn_cast<MemIntrinsic>(Dep.getInst())) { 1753 if (const TargetData *TD = getAnalysisIfAvailable<TargetData>()) { 1754 int Offset = AnalyzeLoadFromClobberingMemInst(L->getType(), 1755 L->getPointerOperand(), 1756 DepMI, *TD); 1757 if (Offset != -1) 1758 AvailVal = GetMemInstValueForLoad(DepMI, Offset, L->getType(), L,*TD); 1759 } 1760 } 1761 1762 if (AvailVal) { 1763 DEBUG(dbgs() << "GVN COERCED INST:\n" << *Dep.getInst() << '\n' 1764 << *AvailVal << '\n' << *L << "\n\n\n"); 1765 1766 // Replace the load! 1767 L->replaceAllUsesWith(AvailVal); 1768 if (AvailVal->getType()->isPointerTy()) 1769 MD->invalidateCachedPointerInfo(AvailVal); 1770 toErase.push_back(L); 1771 NumGVNLoad++; 1772 return true; 1773 } 1774 1775 DEBUG( 1776 // fast print dep, using operator<< on instruction would be too slow 1777 dbgs() << "GVN: load "; 1778 WriteAsOperand(dbgs(), L); 1779 Instruction *I = Dep.getInst(); 1780 dbgs() << " is clobbered by " << *I << '\n'; 1781 ); 1782 return false; 1783 } 1784 1785 // If it is defined in another block, try harder. 1786 if (Dep.isNonLocal()) 1787 return processNonLocalLoad(L, toErase); 1788 1789 Instruction *DepInst = Dep.getInst(); 1790 if (StoreInst *DepSI = dyn_cast<StoreInst>(DepInst)) { 1791 Value *StoredVal = DepSI->getOperand(0); 1792 1793 // The store and load are to a must-aliased pointer, but they may not 1794 // actually have the same type. See if we know how to reuse the stored 1795 // value (depending on its type). 1796 const TargetData *TD = 0; 1797 if (StoredVal->getType() != L->getType()) { 1798 if ((TD = getAnalysisIfAvailable<TargetData>())) { 1799 StoredVal = CoerceAvailableValueToLoadType(StoredVal, L->getType(), 1800 L, *TD); 1801 if (StoredVal == 0) 1802 return false; 1803 1804 DEBUG(dbgs() << "GVN COERCED STORE:\n" << *DepSI << '\n' << *StoredVal 1805 << '\n' << *L << "\n\n\n"); 1806 } 1807 else 1808 return false; 1809 } 1810 1811 // Remove it! 1812 L->replaceAllUsesWith(StoredVal); 1813 if (StoredVal->getType()->isPointerTy()) 1814 MD->invalidateCachedPointerInfo(StoredVal); 1815 toErase.push_back(L); 1816 NumGVNLoad++; 1817 return true; 1818 } 1819 1820 if (LoadInst *DepLI = dyn_cast<LoadInst>(DepInst)) { 1821 Value *AvailableVal = DepLI; 1822 1823 // The loads are of a must-aliased pointer, but they may not actually have 1824 // the same type. See if we know how to reuse the previously loaded value 1825 // (depending on its type). 1826 const TargetData *TD = 0; 1827 if (DepLI->getType() != L->getType()) { 1828 if ((TD = getAnalysisIfAvailable<TargetData>())) { 1829 AvailableVal = CoerceAvailableValueToLoadType(DepLI, L->getType(), L,*TD); 1830 if (AvailableVal == 0) 1831 return false; 1832 1833 DEBUG(dbgs() << "GVN COERCED LOAD:\n" << *DepLI << "\n" << *AvailableVal 1834 << "\n" << *L << "\n\n\n"); 1835 } 1836 else 1837 return false; 1838 } 1839 1840 // Remove it! 1841 L->replaceAllUsesWith(AvailableVal); 1842 if (DepLI->getType()->isPointerTy()) 1843 MD->invalidateCachedPointerInfo(DepLI); 1844 toErase.push_back(L); 1845 NumGVNLoad++; 1846 return true; 1847 } 1848 1849 // If this load really doesn't depend on anything, then we must be loading an 1850 // undef value. This can happen when loading for a fresh allocation with no 1851 // intervening stores, for example. 1852 if (isa<AllocaInst>(DepInst) || isMalloc(DepInst)) { 1853 L->replaceAllUsesWith(UndefValue::get(L->getType())); 1854 toErase.push_back(L); 1855 NumGVNLoad++; 1856 return true; 1857 } 1858 1859 // If this load occurs either right after a lifetime begin, 1860 // then the loaded value is undefined. 1861 if (IntrinsicInst* II = dyn_cast<IntrinsicInst>(DepInst)) { 1862 if (II->getIntrinsicID() == Intrinsic::lifetime_start) { 1863 L->replaceAllUsesWith(UndefValue::get(L->getType())); 1864 toErase.push_back(L); 1865 NumGVNLoad++; 1866 return true; 1867 } 1868 } 1869 1870 return false; 1871} 1872 1873Value *GVN::lookupNumber(BasicBlock *BB, uint32_t num) { 1874 DenseMap<BasicBlock*, ValueNumberScope*>::iterator I = localAvail.find(BB); 1875 if (I == localAvail.end()) 1876 return 0; 1877 1878 ValueNumberScope *Locals = I->second; 1879 while (Locals) { 1880 DenseMap<uint32_t, Value*>::iterator I = Locals->table.find(num); 1881 if (I != Locals->table.end()) 1882 return I->second; 1883 Locals = Locals->parent; 1884 } 1885 1886 return 0; 1887} 1888 1889 1890/// processInstruction - When calculating availability, handle an instruction 1891/// by inserting it into the appropriate sets 1892bool GVN::processInstruction(Instruction *I, 1893 SmallVectorImpl<Instruction*> &toErase) { 1894 // Ignore dbg info intrinsics. 1895 if (isa<DbgInfoIntrinsic>(I)) 1896 return false; 1897 1898 if (LoadInst *LI = dyn_cast<LoadInst>(I)) { 1899 bool Changed = processLoad(LI, toErase); 1900 1901 if (!Changed) { 1902 unsigned Num = VN.lookup_or_add(LI); 1903 localAvail[I->getParent()]->table.insert(std::make_pair(Num, LI)); 1904 } 1905 1906 return Changed; 1907 } 1908 1909 uint32_t NextNum = VN.getNextUnusedValueNumber(); 1910 unsigned Num = VN.lookup_or_add(I); 1911 1912 if (BranchInst *BI = dyn_cast<BranchInst>(I)) { 1913 localAvail[I->getParent()]->table.insert(std::make_pair(Num, I)); 1914 1915 if (!BI->isConditional() || isa<Constant>(BI->getCondition())) 1916 return false; 1917 1918 Value *BranchCond = BI->getCondition(); 1919 uint32_t CondVN = VN.lookup_or_add(BranchCond); 1920 1921 BasicBlock *TrueSucc = BI->getSuccessor(0); 1922 BasicBlock *FalseSucc = BI->getSuccessor(1); 1923 1924 if (TrueSucc->getSinglePredecessor()) 1925 localAvail[TrueSucc]->table[CondVN] = 1926 ConstantInt::getTrue(TrueSucc->getContext()); 1927 if (FalseSucc->getSinglePredecessor()) 1928 localAvail[FalseSucc]->table[CondVN] = 1929 ConstantInt::getFalse(TrueSucc->getContext()); 1930 1931 return false; 1932 1933 // Allocations are always uniquely numbered, so we can save time and memory 1934 // by fast failing them. 1935 } else if (isa<AllocaInst>(I) || isa<TerminatorInst>(I)) { 1936 localAvail[I->getParent()]->table.insert(std::make_pair(Num, I)); 1937 return false; 1938 } 1939 1940 // Collapse PHI nodes 1941 if (PHINode* p = dyn_cast<PHINode>(I)) { 1942 Value *constVal = CollapsePhi(p); 1943 1944 if (constVal) { 1945 p->replaceAllUsesWith(constVal); 1946 if (MD && constVal->getType()->isPointerTy()) 1947 MD->invalidateCachedPointerInfo(constVal); 1948 VN.erase(p); 1949 1950 toErase.push_back(p); 1951 } else { 1952 localAvail[I->getParent()]->table.insert(std::make_pair(Num, I)); 1953 } 1954 1955 // If the number we were assigned was a brand new VN, then we don't 1956 // need to do a lookup to see if the number already exists 1957 // somewhere in the domtree: it can't! 1958 } else if (Num == NextNum) { 1959 localAvail[I->getParent()]->table.insert(std::make_pair(Num, I)); 1960 1961 // Perform fast-path value-number based elimination of values inherited from 1962 // dominators. 1963 } else if (Value *repl = lookupNumber(I->getParent(), Num)) { 1964 // Remove it! 1965 VN.erase(I); 1966 I->replaceAllUsesWith(repl); 1967 if (MD && repl->getType()->isPointerTy()) 1968 MD->invalidateCachedPointerInfo(repl); 1969 toErase.push_back(I); 1970 return true; 1971 1972 } else { 1973 localAvail[I->getParent()]->table.insert(std::make_pair(Num, I)); 1974 } 1975 1976 return false; 1977} 1978 1979/// runOnFunction - This is the main transformation entry point for a function. 1980bool GVN::runOnFunction(Function& F) { 1981 if (!NoLoads) 1982 MD = &getAnalysis<MemoryDependenceAnalysis>(); 1983 DT = &getAnalysis<DominatorTree>(); 1984 VN.setAliasAnalysis(&getAnalysis<AliasAnalysis>()); 1985 VN.setMemDep(MD); 1986 VN.setDomTree(DT); 1987 1988 bool Changed = false; 1989 bool ShouldContinue = true; 1990 1991 // Merge unconditional branches, allowing PRE to catch more 1992 // optimization opportunities. 1993 for (Function::iterator FI = F.begin(), FE = F.end(); FI != FE; ) { 1994 BasicBlock *BB = FI; 1995 ++FI; 1996 bool removedBlock = MergeBlockIntoPredecessor(BB, this); 1997 if (removedBlock) NumGVNBlocks++; 1998 1999 Changed |= removedBlock; 2000 } 2001 2002 unsigned Iteration = 0; 2003 2004 while (ShouldContinue) { 2005 DEBUG(dbgs() << "GVN iteration: " << Iteration << "\n"); 2006 ShouldContinue = iterateOnFunction(F); 2007 Changed |= ShouldContinue; 2008 ++Iteration; 2009 } 2010 2011 if (EnablePRE) { 2012 bool PREChanged = true; 2013 while (PREChanged) { 2014 PREChanged = performPRE(F); 2015 Changed |= PREChanged; 2016 } 2017 } 2018 // FIXME: Should perform GVN again after PRE does something. PRE can move 2019 // computations into blocks where they become fully redundant. Note that 2020 // we can't do this until PRE's critical edge splitting updates memdep. 2021 // Actually, when this happens, we should just fully integrate PRE into GVN. 2022 2023 cleanupGlobalSets(); 2024 2025 return Changed; 2026} 2027 2028 2029bool GVN::processBlock(BasicBlock *BB) { 2030 // FIXME: Kill off toErase by doing erasing eagerly in a helper function (and 2031 // incrementing BI before processing an instruction). 2032 SmallVector<Instruction*, 8> toErase; 2033 bool ChangedFunction = false; 2034 2035 for (BasicBlock::iterator BI = BB->begin(), BE = BB->end(); 2036 BI != BE;) { 2037 ChangedFunction |= processInstruction(BI, toErase); 2038 if (toErase.empty()) { 2039 ++BI; 2040 continue; 2041 } 2042 2043 // If we need some instructions deleted, do it now. 2044 NumGVNInstr += toErase.size(); 2045 2046 // Avoid iterator invalidation. 2047 bool AtStart = BI == BB->begin(); 2048 if (!AtStart) 2049 --BI; 2050 2051 for (SmallVector<Instruction*, 4>::iterator I = toErase.begin(), 2052 E = toErase.end(); I != E; ++I) { 2053 DEBUG(dbgs() << "GVN removed: " << **I << '\n'); 2054 if (MD) MD->removeInstruction(*I); 2055 (*I)->eraseFromParent(); 2056 DEBUG(verifyRemoved(*I)); 2057 } 2058 toErase.clear(); 2059 2060 if (AtStart) 2061 BI = BB->begin(); 2062 else 2063 ++BI; 2064 } 2065 2066 return ChangedFunction; 2067} 2068 2069/// performPRE - Perform a purely local form of PRE that looks for diamond 2070/// control flow patterns and attempts to perform simple PRE at the join point. 2071bool GVN::performPRE(Function &F) { 2072 bool Changed = false; 2073 SmallVector<std::pair<TerminatorInst*, unsigned>, 4> toSplit; 2074 DenseMap<BasicBlock*, Value*> predMap; 2075 for (df_iterator<BasicBlock*> DI = df_begin(&F.getEntryBlock()), 2076 DE = df_end(&F.getEntryBlock()); DI != DE; ++DI) { 2077 BasicBlock *CurrentBlock = *DI; 2078 2079 // Nothing to PRE in the entry block. 2080 if (CurrentBlock == &F.getEntryBlock()) continue; 2081 2082 for (BasicBlock::iterator BI = CurrentBlock->begin(), 2083 BE = CurrentBlock->end(); BI != BE; ) { 2084 Instruction *CurInst = BI++; 2085 2086 if (isa<AllocaInst>(CurInst) || 2087 isa<TerminatorInst>(CurInst) || isa<PHINode>(CurInst) || 2088 CurInst->getType()->isVoidTy() || 2089 CurInst->mayReadFromMemory() || CurInst->mayHaveSideEffects() || 2090 isa<DbgInfoIntrinsic>(CurInst)) 2091 continue; 2092 2093 uint32_t ValNo = VN.lookup(CurInst); 2094 2095 // Look for the predecessors for PRE opportunities. We're 2096 // only trying to solve the basic diamond case, where 2097 // a value is computed in the successor and one predecessor, 2098 // but not the other. We also explicitly disallow cases 2099 // where the successor is its own predecessor, because they're 2100 // more complicated to get right. 2101 unsigned NumWith = 0; 2102 unsigned NumWithout = 0; 2103 BasicBlock *PREPred = 0; 2104 predMap.clear(); 2105 2106 for (pred_iterator PI = pred_begin(CurrentBlock), 2107 PE = pred_end(CurrentBlock); PI != PE; ++PI) { 2108 // We're not interested in PRE where the block is its 2109 // own predecessor, or in blocks with predecessors 2110 // that are not reachable. 2111 if (*PI == CurrentBlock) { 2112 NumWithout = 2; 2113 break; 2114 } else if (!localAvail.count(*PI)) { 2115 NumWithout = 2; 2116 break; 2117 } 2118 2119 DenseMap<uint32_t, Value*>::iterator predV = 2120 localAvail[*PI]->table.find(ValNo); 2121 if (predV == localAvail[*PI]->table.end()) { 2122 PREPred = *PI; 2123 NumWithout++; 2124 } else if (predV->second == CurInst) { 2125 NumWithout = 2; 2126 } else { 2127 predMap[*PI] = predV->second; 2128 NumWith++; 2129 } 2130 } 2131 2132 // Don't do PRE when it might increase code size, i.e. when 2133 // we would need to insert instructions in more than one pred. 2134 if (NumWithout != 1 || NumWith == 0) 2135 continue; 2136 2137 // Don't do PRE across indirect branch. 2138 if (isa<IndirectBrInst>(PREPred->getTerminator())) 2139 continue; 2140 2141 // We can't do PRE safely on a critical edge, so instead we schedule 2142 // the edge to be split and perform the PRE the next time we iterate 2143 // on the function. 2144 unsigned SuccNum = SuccessorNumber(PREPred, CurrentBlock); 2145 if (isCriticalEdge(PREPred->getTerminator(), SuccNum)) { 2146 toSplit.push_back(std::make_pair(PREPred->getTerminator(), SuccNum)); 2147 continue; 2148 } 2149 2150 // Instantiate the expression in the predecessor that lacked it. 2151 // Because we are going top-down through the block, all value numbers 2152 // will be available in the predecessor by the time we need them. Any 2153 // that weren't originally present will have been instantiated earlier 2154 // in this loop. 2155 Instruction *PREInstr = CurInst->clone(); 2156 bool success = true; 2157 for (unsigned i = 0, e = CurInst->getNumOperands(); i != e; ++i) { 2158 Value *Op = PREInstr->getOperand(i); 2159 if (isa<Argument>(Op) || isa<Constant>(Op) || isa<GlobalValue>(Op)) 2160 continue; 2161 2162 if (Value *V = lookupNumber(PREPred, VN.lookup(Op))) { 2163 PREInstr->setOperand(i, V); 2164 } else { 2165 success = false; 2166 break; 2167 } 2168 } 2169 2170 // Fail out if we encounter an operand that is not available in 2171 // the PRE predecessor. This is typically because of loads which 2172 // are not value numbered precisely. 2173 if (!success) { 2174 delete PREInstr; 2175 DEBUG(verifyRemoved(PREInstr)); 2176 continue; 2177 } 2178 2179 PREInstr->insertBefore(PREPred->getTerminator()); 2180 PREInstr->setName(CurInst->getName() + ".pre"); 2181 predMap[PREPred] = PREInstr; 2182 VN.add(PREInstr, ValNo); 2183 NumGVNPRE++; 2184 2185 // Update the availability map to include the new instruction. 2186 localAvail[PREPred]->table.insert(std::make_pair(ValNo, PREInstr)); 2187 2188 // Create a PHI to make the value available in this block. 2189 PHINode* Phi = PHINode::Create(CurInst->getType(), 2190 CurInst->getName() + ".pre-phi", 2191 CurrentBlock->begin()); 2192 for (pred_iterator PI = pred_begin(CurrentBlock), 2193 PE = pred_end(CurrentBlock); PI != PE; ++PI) 2194 Phi->addIncoming(predMap[*PI], *PI); 2195 2196 VN.add(Phi, ValNo); 2197 localAvail[CurrentBlock]->table[ValNo] = Phi; 2198 2199 CurInst->replaceAllUsesWith(Phi); 2200 if (MD && Phi->getType()->isPointerTy()) 2201 MD->invalidateCachedPointerInfo(Phi); 2202 VN.erase(CurInst); 2203 2204 DEBUG(dbgs() << "GVN PRE removed: " << *CurInst << '\n'); 2205 if (MD) MD->removeInstruction(CurInst); 2206 CurInst->eraseFromParent(); 2207 DEBUG(verifyRemoved(CurInst)); 2208 Changed = true; 2209 } 2210 } 2211 2212 for (SmallVector<std::pair<TerminatorInst*, unsigned>, 4>::iterator 2213 I = toSplit.begin(), E = toSplit.end(); I != E; ++I) 2214 SplitCriticalEdge(I->first, I->second, this); 2215 2216 return Changed || toSplit.size(); 2217} 2218 2219/// iterateOnFunction - Executes one iteration of GVN 2220bool GVN::iterateOnFunction(Function &F) { 2221 cleanupGlobalSets(); 2222 2223 for (df_iterator<DomTreeNode*> DI = df_begin(DT->getRootNode()), 2224 DE = df_end(DT->getRootNode()); DI != DE; ++DI) { 2225 if (DI->getIDom()) 2226 localAvail[DI->getBlock()] = 2227 new ValueNumberScope(localAvail[DI->getIDom()->getBlock()]); 2228 else 2229 localAvail[DI->getBlock()] = new ValueNumberScope(0); 2230 } 2231 2232 // Top-down walk of the dominator tree 2233 bool Changed = false; 2234#if 0 2235 // Needed for value numbering with phi construction to work. 2236 ReversePostOrderTraversal<Function*> RPOT(&F); 2237 for (ReversePostOrderTraversal<Function*>::rpo_iterator RI = RPOT.begin(), 2238 RE = RPOT.end(); RI != RE; ++RI) 2239 Changed |= processBlock(*RI); 2240#else 2241 for (df_iterator<DomTreeNode*> DI = df_begin(DT->getRootNode()), 2242 DE = df_end(DT->getRootNode()); DI != DE; ++DI) 2243 Changed |= processBlock(DI->getBlock()); 2244#endif 2245 2246 return Changed; 2247} 2248 2249void GVN::cleanupGlobalSets() { 2250 VN.clear(); 2251 2252 for (DenseMap<BasicBlock*, ValueNumberScope*>::iterator 2253 I = localAvail.begin(), E = localAvail.end(); I != E; ++I) 2254 delete I->second; 2255 localAvail.clear(); 2256} 2257 2258/// verifyRemoved - Verify that the specified instruction does not occur in our 2259/// internal data structures. 2260void GVN::verifyRemoved(const Instruction *Inst) const { 2261 VN.verifyRemoved(Inst); 2262 2263 // Walk through the value number scope to make sure the instruction isn't 2264 // ferreted away in it. 2265 for (DenseMap<BasicBlock*, ValueNumberScope*>::const_iterator 2266 I = localAvail.begin(), E = localAvail.end(); I != E; ++I) { 2267 const ValueNumberScope *VNS = I->second; 2268 2269 while (VNS) { 2270 for (DenseMap<uint32_t, Value*>::const_iterator 2271 II = VNS->table.begin(), IE = VNS->table.end(); II != IE; ++II) { 2272 assert(II->second != Inst && "Inst still in value numbering scope!"); 2273 } 2274 2275 VNS = VNS->parent; 2276 } 2277 } 2278} 2279