1 //===- MemorySSA.h - Build Memory SSA ---------------------------*- C++ -*-===// 2 // 3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. 4 // See https://llvm.org/LICENSE.txt for license information. 5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception 6 // 7 //===----------------------------------------------------------------------===// 8 // 9 /// \file 10 /// This file exposes an interface to building/using memory SSA to 11 /// walk memory instructions using a use/def graph. 12 /// 13 /// Memory SSA class builds an SSA form that links together memory access 14 /// instructions such as loads, stores, atomics, and calls. Additionally, it 15 /// does a trivial form of "heap versioning" Every time the memory state changes 16 /// in the program, we generate a new heap version. It generates 17 /// MemoryDef/Uses/Phis that are overlayed on top of the existing instructions. 18 /// 19 /// As a trivial example, 20 /// define i32 @main() #0 { 21 /// entry: 22 /// %call = call noalias i8* @_Znwm(i64 4) #2 23 /// %0 = bitcast i8* %call to i32* 24 /// %call1 = call noalias i8* @_Znwm(i64 4) #2 25 /// %1 = bitcast i8* %call1 to i32* 26 /// store i32 5, i32* %0, align 4 27 /// store i32 7, i32* %1, align 4 28 /// %2 = load i32* %0, align 4 29 /// %3 = load i32* %1, align 4 30 /// %add = add nsw i32 %2, %3 31 /// ret i32 %add 32 /// } 33 /// 34 /// Will become 35 /// define i32 @main() #0 { 36 /// entry: 37 /// ; 1 = MemoryDef(0) 38 /// %call = call noalias i8* @_Znwm(i64 4) #3 39 /// %2 = bitcast i8* %call to i32* 40 /// ; 2 = MemoryDef(1) 41 /// %call1 = call noalias i8* @_Znwm(i64 4) #3 42 /// %4 = bitcast i8* %call1 to i32* 43 /// ; 3 = MemoryDef(2) 44 /// store i32 5, i32* %2, align 4 45 /// ; 4 = MemoryDef(3) 46 /// store i32 7, i32* %4, align 4 47 /// ; MemoryUse(3) 48 /// %7 = load i32* %2, align 4 49 /// ; MemoryUse(4) 50 /// %8 = load i32* %4, align 4 51 /// %add = add nsw i32 %7, %8 52 /// ret i32 %add 53 /// } 54 /// 55 /// Given this form, all the stores that could ever effect the load at %8 can be 56 /// gotten by using the MemoryUse associated with it, and walking from use to 57 /// def until you hit the top of the function. 58 /// 59 /// Each def also has a list of users associated with it, so you can walk from 60 /// both def to users, and users to defs. Note that we disambiguate MemoryUses, 61 /// but not the RHS of MemoryDefs. You can see this above at %7, which would 62 /// otherwise be a MemoryUse(4). Being disambiguated means that for a given 63 /// store, all the MemoryUses on its use lists are may-aliases of that store 64 /// (but the MemoryDefs on its use list may not be). 65 /// 66 /// MemoryDefs are not disambiguated because it would require multiple reaching 67 /// definitions, which would require multiple phis, and multiple memoryaccesses 68 /// per instruction. 69 /// 70 /// In addition to the def/use graph described above, MemoryDefs also contain 71 /// an "optimized" definition use. The "optimized" use points to some def 72 /// reachable through the memory def chain. The optimized def *may* (but is 73 /// not required to) alias the original MemoryDef, but no def *closer* to the 74 /// source def may alias it. As the name implies, the purpose of the optimized 75 /// use is to allow caching of clobber searches for memory defs. The optimized 76 /// def may be nullptr, in which case clients must walk the defining access 77 /// chain. 78 /// 79 /// When iterating the uses of a MemoryDef, both defining uses and optimized 80 /// uses will be encountered. If only one type is needed, the client must 81 /// filter the use walk. 82 // 83 //===----------------------------------------------------------------------===// 84 85 #ifndef LLVM_ANALYSIS_MEMORYSSA_H 86 #define LLVM_ANALYSIS_MEMORYSSA_H 87 88 #include "llvm/ADT/DenseMap.h" 89 #include "llvm/ADT/SmallPtrSet.h" 90 #include "llvm/ADT/SmallVector.h" 91 #include "llvm/ADT/ilist_node.h" 92 #include "llvm/ADT/iterator_range.h" 93 #include "llvm/Analysis/AliasAnalysis.h" 94 #include "llvm/Analysis/MemoryLocation.h" 95 #include "llvm/Analysis/PHITransAddr.h" 96 #include "llvm/IR/DerivedUser.h" 97 #include "llvm/IR/Dominators.h" 98 #include "llvm/IR/Type.h" 99 #include "llvm/IR/User.h" 100 #include "llvm/Pass.h" 101 #include "llvm/Support/Compiler.h" 102 #include <algorithm> 103 #include <cassert> 104 #include <cstddef> 105 #include <iterator> 106 #include <memory> 107 #include <utility> 108 109 namespace llvm { 110 111 template <class GraphType> struct GraphTraits; 112 class Function; 113 class Loop; 114 class LLVMContext; 115 class MemoryAccess; 116 class MemorySSAWalker; 117 class Module; 118 class raw_ostream; 119 120 namespace MSSAHelpers { 121 122 struct AllAccessTag {}; 123 struct DefsOnlyTag {}; 124 125 } // end namespace MSSAHelpers 126 127 enum : unsigned { 128 // Used to signify what the default invalid ID is for MemoryAccess's 129 // getID() 130 INVALID_MEMORYACCESS_ID = -1U 131 }; 132 133 template <class T> class memoryaccess_def_iterator_base; 134 using memoryaccess_def_iterator = memoryaccess_def_iterator_base<MemoryAccess>; 135 using const_memoryaccess_def_iterator = 136 memoryaccess_def_iterator_base<const MemoryAccess>; 137 138 // The base for all memory accesses. All memory accesses in a block are 139 // linked together using an intrusive list. 140 class MemoryAccess 141 : public DerivedUser, 142 public ilist_node<MemoryAccess, ilist_tag<MSSAHelpers::AllAccessTag>>, 143 public ilist_node<MemoryAccess, ilist_tag<MSSAHelpers::DefsOnlyTag>> { 144 public: 145 using AllAccessType = 146 ilist_node<MemoryAccess, ilist_tag<MSSAHelpers::AllAccessTag>>; 147 using DefsOnlyType = 148 ilist_node<MemoryAccess, ilist_tag<MSSAHelpers::DefsOnlyTag>>; 149 150 MemoryAccess(const MemoryAccess &) = delete; 151 MemoryAccess &operator=(const MemoryAccess &) = delete; 152 153 void *operator new(size_t) = delete; 154 155 // Methods for support type inquiry through isa, cast, and 156 // dyn_cast classof(const Value * V)157 static bool classof(const Value *V) { 158 unsigned ID = V->getValueID(); 159 return ID == MemoryUseVal || ID == MemoryPhiVal || ID == MemoryDefVal; 160 } 161 getBlock()162 BasicBlock *getBlock() const { return Block; } 163 164 LLVM_ABI void print(raw_ostream &OS) const; 165 LLVM_ABI void dump() const; 166 167 /// The user iterators for a memory access 168 using iterator = user_iterator; 169 using const_iterator = const_user_iterator; 170 171 /// This iterator walks over all of the defs in a given 172 /// MemoryAccess. For MemoryPhi nodes, this walks arguments. For 173 /// MemoryUse/MemoryDef, this walks the defining access. 174 memoryaccess_def_iterator defs_begin(); 175 const_memoryaccess_def_iterator defs_begin() const; 176 memoryaccess_def_iterator defs_end(); 177 const_memoryaccess_def_iterator defs_end() const; 178 179 /// Get the iterators for the all access list and the defs only list 180 /// We default to the all access list. getIterator()181 AllAccessType::self_iterator getIterator() { 182 return this->AllAccessType::getIterator(); 183 } getIterator()184 AllAccessType::const_self_iterator getIterator() const { 185 return this->AllAccessType::getIterator(); 186 } getReverseIterator()187 AllAccessType::reverse_self_iterator getReverseIterator() { 188 return this->AllAccessType::getReverseIterator(); 189 } getReverseIterator()190 AllAccessType::const_reverse_self_iterator getReverseIterator() const { 191 return this->AllAccessType::getReverseIterator(); 192 } getDefsIterator()193 DefsOnlyType::self_iterator getDefsIterator() { 194 return this->DefsOnlyType::getIterator(); 195 } getDefsIterator()196 DefsOnlyType::const_self_iterator getDefsIterator() const { 197 return this->DefsOnlyType::getIterator(); 198 } getReverseDefsIterator()199 DefsOnlyType::reverse_self_iterator getReverseDefsIterator() { 200 return this->DefsOnlyType::getReverseIterator(); 201 } getReverseDefsIterator()202 DefsOnlyType::const_reverse_self_iterator getReverseDefsIterator() const { 203 return this->DefsOnlyType::getReverseIterator(); 204 } 205 206 protected: 207 friend class MemoryDef; 208 friend class MemoryPhi; 209 friend class MemorySSA; 210 friend class MemoryUse; 211 friend class MemoryUseOrDef; 212 213 /// Used by MemorySSA to change the block of a MemoryAccess when it is 214 /// moved. setBlock(BasicBlock * BB)215 void setBlock(BasicBlock *BB) { Block = BB; } 216 217 /// Used for debugging and tracking things about MemoryAccesses. 218 /// Guaranteed unique among MemoryAccesses, no guarantees otherwise. 219 inline unsigned getID() const; 220 MemoryAccess(LLVMContext & C,unsigned Vty,DeleteValueTy DeleteValue,BasicBlock * BB,AllocInfo AllocInfo)221 MemoryAccess(LLVMContext &C, unsigned Vty, DeleteValueTy DeleteValue, 222 BasicBlock *BB, AllocInfo AllocInfo) 223 : DerivedUser(Type::getVoidTy(C), Vty, AllocInfo, DeleteValue), 224 Block(BB) {} 225 226 // Use deleteValue() to delete a generic MemoryAccess. 227 ~MemoryAccess() = default; 228 229 private: 230 BasicBlock *Block; 231 }; 232 233 template <> 234 struct ilist_alloc_traits<MemoryAccess> { 235 static void deleteNode(MemoryAccess *MA) { MA->deleteValue(); } 236 }; 237 238 inline raw_ostream &operator<<(raw_ostream &OS, const MemoryAccess &MA) { 239 MA.print(OS); 240 return OS; 241 } 242 243 /// Class that has the common methods + fields of memory uses/defs. It's 244 /// a little awkward to have, but there are many cases where we want either a 245 /// use or def, and there are many cases where uses are needed (defs aren't 246 /// acceptable), and vice-versa. 247 /// 248 /// This class should never be instantiated directly; make a MemoryUse or 249 /// MemoryDef instead. 250 class MemoryUseOrDef : public MemoryAccess { 251 public: 252 void *operator new(size_t) = delete; 253 254 DECLARE_TRANSPARENT_OPERAND_ACCESSORS(MemoryAccess); 255 256 /// Get the instruction that this MemoryUse represents. 257 Instruction *getMemoryInst() const { return MemoryInstruction; } 258 259 /// Get the access that produces the memory state used by this Use. 260 MemoryAccess *getDefiningAccess() const { return getOperand(0); } 261 262 static bool classof(const Value *MA) { 263 return MA->getValueID() == MemoryUseVal || MA->getValueID() == MemoryDefVal; 264 } 265 266 /// Do we have an optimized use? 267 inline bool isOptimized() const; 268 /// Return the MemoryAccess associated with the optimized use, or nullptr. 269 inline MemoryAccess *getOptimized() const; 270 /// Sets the optimized use for a MemoryDef. 271 inline void setOptimized(MemoryAccess *); 272 273 /// Reset the ID of what this MemoryUse was optimized to, causing it to 274 /// be rewalked by the walker if necessary. 275 /// This really should only be called by tests. 276 inline void resetOptimized(); 277 278 protected: 279 friend class MemorySSA; 280 friend class MemorySSAUpdater; 281 282 MemoryUseOrDef(LLVMContext &C, MemoryAccess *DMA, unsigned Vty, 283 DeleteValueTy DeleteValue, Instruction *MI, BasicBlock *BB, 284 AllocInfo AllocInfo) 285 : MemoryAccess(C, Vty, DeleteValue, BB, AllocInfo), 286 MemoryInstruction(MI) { 287 setDefiningAccess(DMA); 288 } 289 290 // Use deleteValue() to delete a generic MemoryUseOrDef. 291 ~MemoryUseOrDef() = default; 292 293 void setDefiningAccess(MemoryAccess *DMA, bool Optimized = false) { 294 if (!Optimized) { 295 setOperand(0, DMA); 296 return; 297 } 298 setOptimized(DMA); 299 } 300 301 private: 302 Instruction *MemoryInstruction; 303 }; 304 305 /// Represents read-only accesses to memory 306 /// 307 /// In particular, the set of Instructions that will be represented by 308 /// MemoryUse's is exactly the set of Instructions for which 309 /// AliasAnalysis::getModRefInfo returns "Ref". 310 class MemoryUse final : public MemoryUseOrDef { 311 constexpr static IntrusiveOperandsAllocMarker AllocMarker{1}; 312 313 public: 314 DECLARE_TRANSPARENT_OPERAND_ACCESSORS(MemoryAccess); 315 316 MemoryUse(LLVMContext &C, MemoryAccess *DMA, Instruction *MI, BasicBlock *BB) 317 : MemoryUseOrDef(C, DMA, MemoryUseVal, deleteMe, MI, BB, AllocMarker) {} 318 319 // allocate space for exactly one operand 320 void *operator new(size_t S) { return User::operator new(S, AllocMarker); } 321 void operator delete(void *Ptr) { User::operator delete(Ptr); } 322 323 static bool classof(const Value *MA) { 324 return MA->getValueID() == MemoryUseVal; 325 } 326 327 LLVM_ABI void print(raw_ostream &OS) const; 328 329 void setOptimized(MemoryAccess *DMA) { 330 OptimizedID = DMA->getID(); 331 setOperand(0, DMA); 332 } 333 334 /// Whether the MemoryUse is optimized. If ensureOptimizedUses() was called, 335 /// uses will usually be optimized, but this is not guaranteed (e.g. due to 336 /// invalidation and optimization limits.) 337 bool isOptimized() const { 338 return getDefiningAccess() && OptimizedID == getDefiningAccess()->getID(); 339 } 340 341 MemoryAccess *getOptimized() const { 342 return getDefiningAccess(); 343 } 344 345 void resetOptimized() { 346 OptimizedID = INVALID_MEMORYACCESS_ID; 347 } 348 349 protected: 350 friend class MemorySSA; 351 352 private: 353 static void deleteMe(DerivedUser *Self); 354 355 unsigned OptimizedID = INVALID_MEMORYACCESS_ID; 356 }; 357 358 template <> 359 struct OperandTraits<MemoryUse> : public FixedNumOperandTraits<MemoryUse, 1> {}; 360 DEFINE_TRANSPARENT_OPERAND_ACCESSORS(MemoryUse, MemoryAccess) 361 362 /// Represents a read-write access to memory, whether it is a must-alias, 363 /// or a may-alias. 364 /// 365 /// In particular, the set of Instructions that will be represented by 366 /// MemoryDef's is exactly the set of Instructions for which 367 /// AliasAnalysis::getModRefInfo returns "Mod" or "ModRef". 368 /// Note that, in order to provide def-def chains, all defs also have a use 369 /// associated with them. This use points to the nearest reaching 370 /// MemoryDef/MemoryPhi. 371 class MemoryDef final : public MemoryUseOrDef { 372 constexpr static IntrusiveOperandsAllocMarker AllocMarker{2}; 373 374 public: 375 friend class MemorySSA; 376 377 DECLARE_TRANSPARENT_OPERAND_ACCESSORS(MemoryAccess); 378 379 MemoryDef(LLVMContext &C, MemoryAccess *DMA, Instruction *MI, BasicBlock *BB, 380 unsigned Ver) 381 : MemoryUseOrDef(C, DMA, MemoryDefVal, deleteMe, MI, BB, AllocMarker), 382 ID(Ver) {} 383 384 // allocate space for exactly two operands 385 void *operator new(size_t S) { return User::operator new(S, AllocMarker); } 386 void operator delete(void *Ptr) { User::operator delete(Ptr); } 387 388 static bool classof(const Value *MA) { 389 return MA->getValueID() == MemoryDefVal; 390 } 391 392 void setOptimized(MemoryAccess *MA) { 393 setOperand(1, MA); 394 OptimizedID = MA->getID(); 395 } 396 397 MemoryAccess *getOptimized() const { 398 return cast_or_null<MemoryAccess>(getOperand(1)); 399 } 400 401 bool isOptimized() const { 402 return getOptimized() && OptimizedID == getOptimized()->getID(); 403 } 404 405 void resetOptimized() { 406 OptimizedID = INVALID_MEMORYACCESS_ID; 407 setOperand(1, nullptr); 408 } 409 410 LLVM_ABI void print(raw_ostream &OS) const; 411 412 unsigned getID() const { return ID; } 413 414 private: 415 static void deleteMe(DerivedUser *Self); 416 417 const unsigned ID; 418 unsigned OptimizedID = INVALID_MEMORYACCESS_ID; 419 }; 420 421 template <> 422 struct OperandTraits<MemoryDef> : public FixedNumOperandTraits<MemoryDef, 2> {}; 423 DEFINE_TRANSPARENT_OPERAND_ACCESSORS(MemoryDef, MemoryAccess) 424 425 template <> 426 struct OperandTraits<MemoryUseOrDef> { 427 static Use *op_begin(MemoryUseOrDef *MUD) { 428 if (auto *MU = dyn_cast<MemoryUse>(MUD)) 429 return OperandTraits<MemoryUse>::op_begin(MU); 430 return OperandTraits<MemoryDef>::op_begin(cast<MemoryDef>(MUD)); 431 } 432 433 static Use *op_end(MemoryUseOrDef *MUD) { 434 if (auto *MU = dyn_cast<MemoryUse>(MUD)) 435 return OperandTraits<MemoryUse>::op_end(MU); 436 return OperandTraits<MemoryDef>::op_end(cast<MemoryDef>(MUD)); 437 } 438 439 static unsigned operands(const MemoryUseOrDef *MUD) { 440 if (const auto *MU = dyn_cast<MemoryUse>(MUD)) 441 return OperandTraits<MemoryUse>::operands(MU); 442 return OperandTraits<MemoryDef>::operands(cast<MemoryDef>(MUD)); 443 } 444 }; 445 DEFINE_TRANSPARENT_OPERAND_ACCESSORS(MemoryUseOrDef, MemoryAccess) 446 447 /// Represents phi nodes for memory accesses. 448 /// 449 /// These have the same semantic as regular phi nodes, with the exception that 450 /// only one phi will ever exist in a given basic block. 451 /// Guaranteeing one phi per block means guaranteeing there is only ever one 452 /// valid reaching MemoryDef/MemoryPHI along each path to the phi node. 453 /// This is ensured by not allowing disambiguation of the RHS of a MemoryDef or 454 /// a MemoryPhi's operands. 455 /// That is, given 456 /// if (a) { 457 /// store %a 458 /// store %b 459 /// } 460 /// it *must* be transformed into 461 /// if (a) { 462 /// 1 = MemoryDef(liveOnEntry) 463 /// store %a 464 /// 2 = MemoryDef(1) 465 /// store %b 466 /// } 467 /// and *not* 468 /// if (a) { 469 /// 1 = MemoryDef(liveOnEntry) 470 /// store %a 471 /// 2 = MemoryDef(liveOnEntry) 472 /// store %b 473 /// } 474 /// even if the two stores do not conflict. Otherwise, both 1 and 2 reach the 475 /// end of the branch, and if there are not two phi nodes, one will be 476 /// disconnected completely from the SSA graph below that point. 477 /// Because MemoryUse's do not generate new definitions, they do not have this 478 /// issue. 479 class MemoryPhi final : public MemoryAccess { 480 constexpr static HungOffOperandsAllocMarker AllocMarker{}; 481 482 // allocate space for exactly zero operands 483 void *operator new(size_t S) { return User::operator new(S, AllocMarker); } 484 485 public: 486 void operator delete(void *Ptr) { User::operator delete(Ptr); } 487 488 /// Provide fast operand accessors 489 DECLARE_TRANSPARENT_OPERAND_ACCESSORS(MemoryAccess); 490 491 MemoryPhi(LLVMContext &C, BasicBlock *BB, unsigned Ver, unsigned NumPreds = 0) 492 : MemoryAccess(C, MemoryPhiVal, deleteMe, BB, AllocMarker), ID(Ver), 493 ReservedSpace(NumPreds) { 494 allocHungoffUses(ReservedSpace); 495 } 496 497 // Block iterator interface. This provides access to the list of incoming 498 // basic blocks, which parallels the list of incoming values. 499 using block_iterator = BasicBlock **; 500 using const_block_iterator = BasicBlock *const *; 501 502 block_iterator block_begin() { 503 return reinterpret_cast<block_iterator>(op_begin() + ReservedSpace); 504 } 505 506 const_block_iterator block_begin() const { 507 return reinterpret_cast<const_block_iterator>(op_begin() + ReservedSpace); 508 } 509 510 block_iterator block_end() { return block_begin() + getNumOperands(); } 511 512 const_block_iterator block_end() const { 513 return block_begin() + getNumOperands(); 514 } 515 516 iterator_range<block_iterator> blocks() { 517 return make_range(block_begin(), block_end()); 518 } 519 520 iterator_range<const_block_iterator> blocks() const { 521 return make_range(block_begin(), block_end()); 522 } 523 524 op_range incoming_values() { return operands(); } 525 526 const_op_range incoming_values() const { return operands(); } 527 528 /// Return the number of incoming edges 529 unsigned getNumIncomingValues() const { return getNumOperands(); } 530 531 /// Return incoming value number x 532 MemoryAccess *getIncomingValue(unsigned I) const { return getOperand(I); } 533 void setIncomingValue(unsigned I, MemoryAccess *V) { 534 assert(V && "PHI node got a null value!"); 535 setOperand(I, V); 536 } 537 538 static unsigned getOperandNumForIncomingValue(unsigned I) { return I; } 539 static unsigned getIncomingValueNumForOperand(unsigned I) { return I; } 540 541 /// Return incoming basic block number @p i. 542 BasicBlock *getIncomingBlock(unsigned I) const { return block_begin()[I]; } 543 544 /// Return incoming basic block corresponding 545 /// to an operand of the PHI. 546 BasicBlock *getIncomingBlock(const Use &U) const { 547 assert(this == U.getUser() && "Iterator doesn't point to PHI's Uses?"); 548 return getIncomingBlock(unsigned(&U - op_begin())); 549 } 550 551 /// Return incoming basic block corresponding 552 /// to value use iterator. 553 BasicBlock *getIncomingBlock(MemoryAccess::const_user_iterator I) const { 554 return getIncomingBlock(I.getUse()); 555 } 556 557 void setIncomingBlock(unsigned I, BasicBlock *BB) { 558 assert(BB && "PHI node got a null basic block!"); 559 block_begin()[I] = BB; 560 } 561 562 /// Add an incoming value to the end of the PHI list 563 void addIncoming(MemoryAccess *V, BasicBlock *BB) { 564 if (getNumOperands() == ReservedSpace) 565 growOperands(); // Get more space! 566 // Initialize some new operands. 567 setNumHungOffUseOperands(getNumOperands() + 1); 568 setIncomingValue(getNumOperands() - 1, V); 569 setIncomingBlock(getNumOperands() - 1, BB); 570 } 571 572 /// Return the first index of the specified basic 573 /// block in the value list for this PHI. Returns -1 if no instance. 574 int getBasicBlockIndex(const BasicBlock *BB) const { 575 for (unsigned I = 0, E = getNumOperands(); I != E; ++I) 576 if (block_begin()[I] == BB) 577 return I; 578 return -1; 579 } 580 581 MemoryAccess *getIncomingValueForBlock(const BasicBlock *BB) const { 582 int Idx = getBasicBlockIndex(BB); 583 assert(Idx >= 0 && "Invalid basic block argument!"); 584 return getIncomingValue(Idx); 585 } 586 587 // After deleting incoming position I, the order of incoming may be changed. 588 void unorderedDeleteIncoming(unsigned I) { 589 unsigned E = getNumOperands(); 590 assert(I < E && "Cannot remove out of bounds Phi entry."); 591 // MemoryPhi must have at least two incoming values, otherwise the MemoryPhi 592 // itself should be deleted. 593 assert(E >= 2 && "Cannot only remove incoming values in MemoryPhis with " 594 "at least 2 values."); 595 setIncomingValue(I, getIncomingValue(E - 1)); 596 setIncomingBlock(I, block_begin()[E - 1]); 597 setOperand(E - 1, nullptr); 598 block_begin()[E - 1] = nullptr; 599 setNumHungOffUseOperands(getNumOperands() - 1); 600 } 601 602 // After deleting entries that satisfy Pred, remaining entries may have 603 // changed order. 604 template <typename Fn> void unorderedDeleteIncomingIf(Fn &&Pred) { 605 for (unsigned I = 0, E = getNumOperands(); I != E; ++I) 606 if (Pred(getIncomingValue(I), getIncomingBlock(I))) { 607 unorderedDeleteIncoming(I); 608 E = getNumOperands(); 609 --I; 610 } 611 assert(getNumOperands() >= 1 && 612 "Cannot remove all incoming blocks in a MemoryPhi."); 613 } 614 615 // After deleting incoming block BB, the incoming blocks order may be changed. 616 void unorderedDeleteIncomingBlock(const BasicBlock *BB) { 617 unorderedDeleteIncomingIf( 618 [&](const MemoryAccess *, const BasicBlock *B) { return BB == B; }); 619 } 620 621 // After deleting incoming memory access MA, the incoming accesses order may 622 // be changed. 623 void unorderedDeleteIncomingValue(const MemoryAccess *MA) { 624 unorderedDeleteIncomingIf( 625 [&](const MemoryAccess *M, const BasicBlock *) { return MA == M; }); 626 } 627 628 static bool classof(const Value *V) { 629 return V->getValueID() == MemoryPhiVal; 630 } 631 632 LLVM_ABI void print(raw_ostream &OS) const; 633 634 unsigned getID() const { return ID; } 635 636 protected: 637 friend class MemorySSA; 638 639 /// this is more complicated than the generic 640 /// User::allocHungoffUses, because we have to allocate Uses for the incoming 641 /// values and pointers to the incoming blocks, all in one allocation. 642 void allocHungoffUses(unsigned N) { 643 User::allocHungoffUses(N, /* IsPhi */ true); 644 } 645 646 private: 647 // For debugging only 648 const unsigned ID; 649 unsigned ReservedSpace; 650 651 /// This grows the operand list in response to a push_back style of 652 /// operation. This grows the number of ops by 1.5 times. 653 void growOperands() { 654 unsigned E = getNumOperands(); 655 // 2 op PHI nodes are VERY common, so reserve at least enough for that. 656 ReservedSpace = std::max(E + E / 2, 2u); 657 growHungoffUses(ReservedSpace, /* IsPhi */ true); 658 } 659 660 static void deleteMe(DerivedUser *Self); 661 }; 662 663 inline unsigned MemoryAccess::getID() const { 664 assert((isa<MemoryDef>(this) || isa<MemoryPhi>(this)) && 665 "only memory defs and phis have ids"); 666 if (const auto *MD = dyn_cast<MemoryDef>(this)) 667 return MD->getID(); 668 return cast<MemoryPhi>(this)->getID(); 669 } 670 671 inline bool MemoryUseOrDef::isOptimized() const { 672 if (const auto *MD = dyn_cast<MemoryDef>(this)) 673 return MD->isOptimized(); 674 return cast<MemoryUse>(this)->isOptimized(); 675 } 676 677 inline MemoryAccess *MemoryUseOrDef::getOptimized() const { 678 if (const auto *MD = dyn_cast<MemoryDef>(this)) 679 return MD->getOptimized(); 680 return cast<MemoryUse>(this)->getOptimized(); 681 } 682 683 inline void MemoryUseOrDef::setOptimized(MemoryAccess *MA) { 684 if (auto *MD = dyn_cast<MemoryDef>(this)) 685 MD->setOptimized(MA); 686 else 687 cast<MemoryUse>(this)->setOptimized(MA); 688 } 689 690 inline void MemoryUseOrDef::resetOptimized() { 691 if (auto *MD = dyn_cast<MemoryDef>(this)) 692 MD->resetOptimized(); 693 else 694 cast<MemoryUse>(this)->resetOptimized(); 695 } 696 697 template <> struct OperandTraits<MemoryPhi> : public HungoffOperandTraits {}; 698 DEFINE_TRANSPARENT_OPERAND_ACCESSORS(MemoryPhi, MemoryAccess) 699 700 /// Encapsulates MemorySSA, including all data associated with memory 701 /// accesses. 702 class MemorySSA { 703 public: 704 LLVM_ABI MemorySSA(Function &, AliasAnalysis *, DominatorTree *); 705 LLVM_ABI MemorySSA(Loop &, AliasAnalysis *, DominatorTree *); 706 707 // MemorySSA must remain where it's constructed; Walkers it creates store 708 // pointers to it. 709 MemorySSA(MemorySSA &&) = delete; 710 711 LLVM_ABI ~MemorySSA(); 712 713 LLVM_ABI MemorySSAWalker *getWalker(); 714 LLVM_ABI MemorySSAWalker *getSkipSelfWalker(); 715 716 /// Given a memory Mod/Ref'ing instruction, get the MemorySSA 717 /// access associated with it. If passed a basic block gets the memory phi 718 /// node that exists for that block, if there is one. Otherwise, this will get 719 /// a MemoryUseOrDef. 720 MemoryUseOrDef *getMemoryAccess(const Instruction *I) const { 721 return cast_or_null<MemoryUseOrDef>(ValueToMemoryAccess.lookup(I)); 722 } 723 724 MemoryPhi *getMemoryAccess(const BasicBlock *BB) const { 725 return cast_or_null<MemoryPhi>(ValueToMemoryAccess.lookup(cast<Value>(BB))); 726 } 727 728 DominatorTree &getDomTree() const { return *DT; } 729 730 LLVM_ABI void dump() const; 731 LLVM_ABI void print(raw_ostream &) const; 732 733 /// Return true if \p MA represents the live on entry value 734 /// 735 /// Loads and stores from pointer arguments and other global values may be 736 /// defined by memory operations that do not occur in the current function, so 737 /// they may be live on entry to the function. MemorySSA represents such 738 /// memory state by the live on entry definition, which is guaranteed to occur 739 /// before any other memory access in the function. 740 inline bool isLiveOnEntryDef(const MemoryAccess *MA) const { 741 return MA == LiveOnEntryDef.get(); 742 } 743 744 inline MemoryAccess *getLiveOnEntryDef() const { 745 return LiveOnEntryDef.get(); 746 } 747 748 // Sadly, iplists, by default, owns and deletes pointers added to the 749 // list. It's not currently possible to have two iplists for the same type, 750 // where one owns the pointers, and one does not. This is because the traits 751 // are per-type, not per-tag. If this ever changes, we should make the 752 // DefList an iplist. 753 using AccessList = iplist<MemoryAccess, ilist_tag<MSSAHelpers::AllAccessTag>>; 754 using DefsList = 755 simple_ilist<MemoryAccess, ilist_tag<MSSAHelpers::DefsOnlyTag>>; 756 757 /// Return the list of MemoryAccess's for a given basic block. 758 /// 759 /// This list is not modifiable by the user. 760 const AccessList *getBlockAccesses(const BasicBlock *BB) const { 761 return getWritableBlockAccesses(BB); 762 } 763 764 /// Return the list of MemoryDef's and MemoryPhi's for a given basic 765 /// block. 766 /// 767 /// This list is not modifiable by the user. 768 const DefsList *getBlockDefs(const BasicBlock *BB) const { 769 return getWritableBlockDefs(BB); 770 } 771 772 /// Given two memory accesses in the same basic block, determine 773 /// whether MemoryAccess \p A dominates MemoryAccess \p B. 774 LLVM_ABI bool locallyDominates(const MemoryAccess *A, 775 const MemoryAccess *B) const; 776 777 /// Given two memory accesses in potentially different blocks, 778 /// determine whether MemoryAccess \p A dominates MemoryAccess \p B. 779 LLVM_ABI bool dominates(const MemoryAccess *A, const MemoryAccess *B) const; 780 781 /// Given a MemoryAccess and a Use, determine whether MemoryAccess \p A 782 /// dominates Use \p B. 783 LLVM_ABI bool dominates(const MemoryAccess *A, const Use &B) const; 784 785 enum class VerificationLevel { Fast, Full }; 786 /// Verify that MemorySSA is self consistent (IE definitions dominate 787 /// all uses, uses appear in the right places). This is used by unit tests. 788 LLVM_ABI void 789 verifyMemorySSA(VerificationLevel = VerificationLevel::Fast) const; 790 791 /// Used in various insertion functions to specify whether we are talking 792 /// about the beginning or end of a block. 793 enum InsertionPlace { Beginning, End, BeforeTerminator }; 794 795 /// By default, uses are *not* optimized during MemorySSA construction. 796 /// Calling this method will attempt to optimize all MemoryUses, if this has 797 /// not happened yet for this MemorySSA instance. This should be done if you 798 /// plan to query the clobbering access for most uses, or if you walk the 799 /// def-use chain of uses. 800 LLVM_ABI void ensureOptimizedUses(); 801 802 AliasAnalysis &getAA() { return *AA; } 803 804 protected: 805 // Used by Memory SSA dumpers and wrapper pass 806 friend class MemorySSAUpdater; 807 808 template <typename IterT> 809 void verifyOrderingDominationAndDefUses( 810 IterT Blocks, VerificationLevel = VerificationLevel::Fast) const; 811 template <typename IterT> void verifyDominationNumbers(IterT Blocks) const; 812 template <typename IterT> void verifyPrevDefInPhis(IterT Blocks) const; 813 814 // This is used by the use optimizer and updater. 815 AccessList *getWritableBlockAccesses(const BasicBlock *BB) const { 816 auto It = PerBlockAccesses.find(BB); 817 return It == PerBlockAccesses.end() ? nullptr : It->second.get(); 818 } 819 820 // This is used by the use optimizer and updater. 821 DefsList *getWritableBlockDefs(const BasicBlock *BB) const { 822 auto It = PerBlockDefs.find(BB); 823 return It == PerBlockDefs.end() ? nullptr : It->second.get(); 824 } 825 826 // These is used by the updater to perform various internal MemorySSA 827 // machinsations. They do not always leave the IR in a correct state, and 828 // relies on the updater to fixup what it breaks, so it is not public. 829 830 LLVM_ABI void moveTo(MemoryUseOrDef *What, BasicBlock *BB, 831 AccessList::iterator Where); 832 LLVM_ABI void moveTo(MemoryAccess *What, BasicBlock *BB, 833 InsertionPlace Point); 834 835 // Rename the dominator tree branch rooted at BB. 836 void renamePass(BasicBlock *BB, MemoryAccess *IncomingVal, 837 SmallPtrSetImpl<BasicBlock *> &Visited) { 838 renamePass(DT->getNode(BB), IncomingVal, Visited, true, true); 839 } 840 841 LLVM_ABI void removeFromLookups(MemoryAccess *); 842 LLVM_ABI void removeFromLists(MemoryAccess *, bool ShouldDelete = true); 843 LLVM_ABI void insertIntoListsForBlock(MemoryAccess *, const BasicBlock *, 844 InsertionPlace); 845 LLVM_ABI void insertIntoListsBefore(MemoryAccess *, const BasicBlock *, 846 AccessList::iterator); 847 LLVM_ABI MemoryUseOrDef * 848 createDefinedAccess(Instruction *, MemoryAccess *, 849 const MemoryUseOrDef *Template = nullptr, 850 bool CreationMustSucceed = true); 851 852 private: 853 class ClobberWalkerBase; 854 class CachingWalker; 855 class SkipSelfWalker; 856 class OptimizeUses; 857 858 CachingWalker *getWalkerImpl(); 859 template <typename IterT> 860 void buildMemorySSA(BatchAAResults &BAA, IterT Blocks); 861 862 void prepareForMoveTo(MemoryAccess *, BasicBlock *); 863 void verifyUseInDefs(MemoryAccess *, MemoryAccess *) const; 864 865 using AccessMap = DenseMap<const BasicBlock *, std::unique_ptr<AccessList>>; 866 using DefsMap = DenseMap<const BasicBlock *, std::unique_ptr<DefsList>>; 867 868 void markUnreachableAsLiveOnEntry(BasicBlock *BB); 869 MemoryPhi *createMemoryPhi(BasicBlock *BB); 870 template <typename AliasAnalysisType> 871 MemoryUseOrDef *createNewAccess(Instruction *, AliasAnalysisType *, 872 const MemoryUseOrDef *Template = nullptr); 873 void placePHINodes(const SmallPtrSetImpl<BasicBlock *> &); 874 MemoryAccess *renameBlock(BasicBlock *, MemoryAccess *, bool); 875 void renameSuccessorPhis(BasicBlock *, MemoryAccess *, bool); 876 LLVM_ABI void renamePass(DomTreeNode *, MemoryAccess *IncomingVal, 877 SmallPtrSetImpl<BasicBlock *> &Visited, 878 bool SkipVisited = false, 879 bool RenameAllUses = false); 880 AccessList *getOrCreateAccessList(const BasicBlock *); 881 DefsList *getOrCreateDefsList(const BasicBlock *); 882 void renumberBlock(const BasicBlock *) const; 883 AliasAnalysis *AA = nullptr; 884 DominatorTree *DT; 885 Function *F = nullptr; 886 Loop *L = nullptr; 887 888 // Memory SSA mappings 889 DenseMap<const Value *, MemoryAccess *> ValueToMemoryAccess; 890 891 // These two mappings contain the main block to access/def mappings for 892 // MemorySSA. The list contained in PerBlockAccesses really owns all the 893 // MemoryAccesses. 894 // Both maps maintain the invariant that if a block is found in them, the 895 // corresponding list is not empty, and if a block is not found in them, the 896 // corresponding list is empty. 897 AccessMap PerBlockAccesses; 898 DefsMap PerBlockDefs; 899 std::unique_ptr<MemoryAccess, ValueDeleter> LiveOnEntryDef; 900 901 // Domination mappings 902 // Note that the numbering is local to a block, even though the map is 903 // global. 904 mutable SmallPtrSet<const BasicBlock *, 16> BlockNumberingValid; 905 mutable DenseMap<const MemoryAccess *, unsigned long> BlockNumbering; 906 907 // Memory SSA building info 908 std::unique_ptr<ClobberWalkerBase> WalkerBase; 909 std::unique_ptr<CachingWalker> Walker; 910 std::unique_ptr<SkipSelfWalker> SkipWalker; 911 unsigned NextID = 0; 912 bool IsOptimized = false; 913 }; 914 915 /// Enables verification of MemorySSA. 916 /// 917 /// The checks which this flag enables is exensive and disabled by default 918 /// unless `EXPENSIVE_CHECKS` is defined. The flag `-verify-memoryssa` can be 919 /// used to selectively enable the verification without re-compilation. 920 LLVM_ABI extern bool VerifyMemorySSA; 921 922 // Internal MemorySSA utils, for use by MemorySSA classes and walkers 923 class MemorySSAUtil { 924 protected: 925 friend class GVNHoist; 926 friend class MemorySSAWalker; 927 928 // This function should not be used by new passes. 929 LLVM_ABI static bool defClobbersUseOrDef(MemoryDef *MD, 930 const MemoryUseOrDef *MU, 931 AliasAnalysis &AA); 932 }; 933 934 /// An analysis that produces \c MemorySSA for a function. 935 /// 936 class MemorySSAAnalysis : public AnalysisInfoMixin<MemorySSAAnalysis> { 937 friend AnalysisInfoMixin<MemorySSAAnalysis>; 938 939 LLVM_ABI static AnalysisKey Key; 940 941 public: 942 // Wrap MemorySSA result to ensure address stability of internal MemorySSA 943 // pointers after construction. Use a wrapper class instead of plain 944 // unique_ptr<MemorySSA> to avoid build breakage on MSVC. 945 struct Result { 946 Result(std::unique_ptr<MemorySSA> &&MSSA) : MSSA(std::move(MSSA)) {} 947 948 MemorySSA &getMSSA() { return *MSSA; } 949 950 std::unique_ptr<MemorySSA> MSSA; 951 952 LLVM_ABI bool invalidate(Function &F, const PreservedAnalyses &PA, 953 FunctionAnalysisManager::Invalidator &Inv); 954 }; 955 956 LLVM_ABI Result run(Function &F, FunctionAnalysisManager &AM); 957 }; 958 959 /// Printer pass for \c MemorySSA. 960 class MemorySSAPrinterPass : public PassInfoMixin<MemorySSAPrinterPass> { 961 raw_ostream &OS; 962 bool EnsureOptimizedUses; 963 964 public: 965 explicit MemorySSAPrinterPass(raw_ostream &OS, bool EnsureOptimizedUses) 966 : OS(OS), EnsureOptimizedUses(EnsureOptimizedUses) {} 967 968 LLVM_ABI PreservedAnalyses run(Function &F, FunctionAnalysisManager &AM); 969 970 static bool isRequired() { return true; } 971 }; 972 973 /// Printer pass for \c MemorySSA via the walker. 974 class MemorySSAWalkerPrinterPass 975 : public PassInfoMixin<MemorySSAWalkerPrinterPass> { 976 raw_ostream &OS; 977 978 public: 979 explicit MemorySSAWalkerPrinterPass(raw_ostream &OS) : OS(OS) {} 980 981 LLVM_ABI PreservedAnalyses run(Function &F, FunctionAnalysisManager &AM); 982 983 static bool isRequired() { return true; } 984 }; 985 986 /// Verifier pass for \c MemorySSA. 987 struct MemorySSAVerifierPass : PassInfoMixin<MemorySSAVerifierPass> { 988 LLVM_ABI PreservedAnalyses run(Function &F, FunctionAnalysisManager &AM); 989 static bool isRequired() { return true; } 990 }; 991 992 /// Legacy analysis pass which computes \c MemorySSA. 993 class LLVM_ABI MemorySSAWrapperPass : public FunctionPass { 994 public: 995 MemorySSAWrapperPass(); 996 997 static char ID; 998 999 bool runOnFunction(Function &) override; 1000 void releaseMemory() override; 1001 MemorySSA &getMSSA() { return *MSSA; } 1002 const MemorySSA &getMSSA() const { return *MSSA; } 1003 1004 void getAnalysisUsage(AnalysisUsage &AU) const override; 1005 1006 void verifyAnalysis() const override; 1007 void print(raw_ostream &OS, const Module *M = nullptr) const override; 1008 1009 private: 1010 std::unique_ptr<MemorySSA> MSSA; 1011 }; 1012 1013 /// This is the generic walker interface for walkers of MemorySSA. 1014 /// Walkers are used to be able to further disambiguate the def-use chains 1015 /// MemorySSA gives you, or otherwise produce better info than MemorySSA gives 1016 /// you. 1017 /// In particular, while the def-use chains provide basic information, and are 1018 /// guaranteed to give, for example, the nearest may-aliasing MemoryDef for a 1019 /// MemoryUse as AliasAnalysis considers it, a user mant want better or other 1020 /// information. In particular, they may want to use SCEV info to further 1021 /// disambiguate memory accesses, or they may want the nearest dominating 1022 /// may-aliasing MemoryDef for a call or a store. This API enables a 1023 /// standardized interface to getting and using that info. 1024 class MemorySSAWalker { 1025 public: 1026 LLVM_ABI MemorySSAWalker(MemorySSA *); 1027 virtual ~MemorySSAWalker() = default; 1028 1029 using MemoryAccessSet = SmallVector<MemoryAccess *, 8>; 1030 1031 /// Given a memory Mod/Ref/ModRef'ing instruction, calling this 1032 /// will give you the nearest dominating MemoryAccess that Mod's the location 1033 /// the instruction accesses (by skipping any def which AA can prove does not 1034 /// alias the location(s) accessed by the instruction given). 1035 /// 1036 /// Note that this will return a single access, and it must dominate the 1037 /// Instruction, so if an operand of a MemoryPhi node Mod's the instruction, 1038 /// this will return the MemoryPhi, not the operand. This means that 1039 /// given: 1040 /// if (a) { 1041 /// 1 = MemoryDef(liveOnEntry) 1042 /// store %a 1043 /// } else { 1044 /// 2 = MemoryDef(liveOnEntry) 1045 /// store %b 1046 /// } 1047 /// 3 = MemoryPhi(2, 1) 1048 /// MemoryUse(3) 1049 /// load %a 1050 /// 1051 /// calling this API on load(%a) will return the MemoryPhi, not the MemoryDef 1052 /// in the if (a) branch. 1053 MemoryAccess *getClobberingMemoryAccess(const Instruction *I, 1054 BatchAAResults &AA) { 1055 MemoryAccess *MA = MSSA->getMemoryAccess(I); 1056 assert(MA && "Handed an instruction that MemorySSA doesn't recognize?"); 1057 return getClobberingMemoryAccess(MA, AA); 1058 } 1059 1060 /// Does the same thing as getClobberingMemoryAccess(const Instruction *I), 1061 /// but takes a MemoryAccess instead of an Instruction. 1062 virtual MemoryAccess *getClobberingMemoryAccess(MemoryAccess *, 1063 BatchAAResults &AA) = 0; 1064 1065 /// Given a potentially clobbering memory access and a new location, 1066 /// calling this will give you the nearest dominating clobbering MemoryAccess 1067 /// (by skipping non-aliasing def links). 1068 /// 1069 /// This version of the function is mainly used to disambiguate phi translated 1070 /// pointers, where the value of a pointer may have changed from the initial 1071 /// memory access. Note that this expects to be handed either a MemoryUse, 1072 /// or an already potentially clobbering access. Unlike the above API, if 1073 /// given a MemoryDef that clobbers the pointer as the starting access, it 1074 /// will return that MemoryDef, whereas the above would return the clobber 1075 /// starting from the use side of the memory def. 1076 virtual MemoryAccess *getClobberingMemoryAccess(MemoryAccess *, 1077 const MemoryLocation &, 1078 BatchAAResults &AA) = 0; 1079 1080 MemoryAccess *getClobberingMemoryAccess(const Instruction *I) { 1081 BatchAAResults BAA(MSSA->getAA()); 1082 return getClobberingMemoryAccess(I, BAA); 1083 } 1084 1085 MemoryAccess *getClobberingMemoryAccess(MemoryAccess *MA) { 1086 BatchAAResults BAA(MSSA->getAA()); 1087 return getClobberingMemoryAccess(MA, BAA); 1088 } 1089 1090 MemoryAccess *getClobberingMemoryAccess(MemoryAccess *MA, 1091 const MemoryLocation &Loc) { 1092 BatchAAResults BAA(MSSA->getAA()); 1093 return getClobberingMemoryAccess(MA, Loc, BAA); 1094 } 1095 1096 /// Given a memory access, invalidate anything this walker knows about 1097 /// that access. 1098 /// This API is used by walkers that store information to perform basic cache 1099 /// invalidation. This will be called by MemorySSA at appropriate times for 1100 /// the walker it uses or returns. 1101 virtual void invalidateInfo(MemoryAccess *) {} 1102 1103 protected: 1104 friend class MemorySSA; // For updating MSSA pointer in MemorySSA move 1105 // constructor. 1106 MemorySSA *MSSA; 1107 }; 1108 1109 /// A MemorySSAWalker that does no alias queries, or anything else. It 1110 /// simply returns the links as they were constructed by the builder. 1111 class LLVM_ABI DoNothingMemorySSAWalker final : public MemorySSAWalker { 1112 public: 1113 // Keep the overrides below from hiding the Instruction overload of 1114 // getClobberingMemoryAccess. 1115 using MemorySSAWalker::getClobberingMemoryAccess; 1116 1117 MemoryAccess *getClobberingMemoryAccess(MemoryAccess *, 1118 BatchAAResults &) override; 1119 MemoryAccess *getClobberingMemoryAccess(MemoryAccess *, 1120 const MemoryLocation &, 1121 BatchAAResults &) override; 1122 }; 1123 1124 using MemoryAccessPair = std::pair<MemoryAccess *, MemoryLocation>; 1125 using ConstMemoryAccessPair = std::pair<const MemoryAccess *, MemoryLocation>; 1126 1127 /// Iterator base class used to implement const and non-const iterators 1128 /// over the defining accesses of a MemoryAccess. 1129 template <class T> 1130 class memoryaccess_def_iterator_base 1131 : public iterator_facade_base<memoryaccess_def_iterator_base<T>, 1132 std::forward_iterator_tag, T, ptrdiff_t, T *, 1133 T *> { 1134 using BaseT = typename memoryaccess_def_iterator_base::iterator_facade_base; 1135 1136 public: 1137 memoryaccess_def_iterator_base(T *Start) : Access(Start) {} 1138 memoryaccess_def_iterator_base() = default; 1139 1140 bool operator==(const memoryaccess_def_iterator_base &Other) const { 1141 return Access == Other.Access && (!Access || ArgNo == Other.ArgNo); 1142 } 1143 1144 // This is a bit ugly, but for MemoryPHI's, unlike PHINodes, you can't get the 1145 // block from the operand in constant time (In a PHINode, the uselist has 1146 // both, so it's just subtraction). We provide it as part of the 1147 // iterator to avoid callers having to linear walk to get the block. 1148 // If the operation becomes constant time on MemoryPHI's, this bit of 1149 // abstraction breaking should be removed. 1150 BasicBlock *getPhiArgBlock() const { 1151 MemoryPhi *MP = dyn_cast<MemoryPhi>(Access); 1152 assert(MP && "Tried to get phi arg block when not iterating over a PHI"); 1153 return MP->getIncomingBlock(ArgNo); 1154 } 1155 1156 typename std::iterator_traits<BaseT>::pointer operator*() const { 1157 assert(Access && "Tried to access past the end of our iterator"); 1158 // Go to the first argument for phis, and the defining access for everything 1159 // else. 1160 if (const MemoryPhi *MP = dyn_cast<MemoryPhi>(Access)) 1161 return MP->getIncomingValue(ArgNo); 1162 return cast<MemoryUseOrDef>(Access)->getDefiningAccess(); 1163 } 1164 1165 using BaseT::operator++; 1166 memoryaccess_def_iterator_base &operator++() { 1167 assert(Access && "Hit end of iterator"); 1168 if (const MemoryPhi *MP = dyn_cast<MemoryPhi>(Access)) { 1169 if (++ArgNo >= MP->getNumIncomingValues()) { 1170 ArgNo = 0; 1171 Access = nullptr; 1172 } 1173 } else { 1174 Access = nullptr; 1175 } 1176 return *this; 1177 } 1178 1179 private: 1180 T *Access = nullptr; 1181 unsigned ArgNo = 0; 1182 }; 1183 1184 inline memoryaccess_def_iterator MemoryAccess::defs_begin() { 1185 return memoryaccess_def_iterator(this); 1186 } 1187 1188 inline const_memoryaccess_def_iterator MemoryAccess::defs_begin() const { 1189 return const_memoryaccess_def_iterator(this); 1190 } 1191 1192 inline memoryaccess_def_iterator MemoryAccess::defs_end() { 1193 return memoryaccess_def_iterator(); 1194 } 1195 1196 inline const_memoryaccess_def_iterator MemoryAccess::defs_end() const { 1197 return const_memoryaccess_def_iterator(); 1198 } 1199 1200 /// GraphTraits for a MemoryAccess, which walks defs in the normal case, 1201 /// and uses in the inverse case. 1202 template <> struct GraphTraits<MemoryAccess *> { 1203 using NodeRef = MemoryAccess *; 1204 using ChildIteratorType = memoryaccess_def_iterator; 1205 1206 static NodeRef getEntryNode(NodeRef N) { return N; } 1207 static ChildIteratorType child_begin(NodeRef N) { return N->defs_begin(); } 1208 static ChildIteratorType child_end(NodeRef N) { return N->defs_end(); } 1209 }; 1210 1211 template <> struct GraphTraits<Inverse<MemoryAccess *>> { 1212 using NodeRef = MemoryAccess *; 1213 using ChildIteratorType = MemoryAccess::iterator; 1214 1215 static NodeRef getEntryNode(NodeRef N) { return N; } 1216 static ChildIteratorType child_begin(NodeRef N) { return N->user_begin(); } 1217 static ChildIteratorType child_end(NodeRef N) { return N->user_end(); } 1218 }; 1219 1220 /// Provide an iterator that walks defs, giving both the memory access, 1221 /// and the current pointer location, updating the pointer location as it 1222 /// changes due to phi node translation. 1223 /// 1224 /// This iterator, while somewhat specialized, is what most clients actually 1225 /// want when walking upwards through MemorySSA def chains. It takes a pair of 1226 /// <MemoryAccess,MemoryLocation>, and walks defs, properly translating the 1227 /// memory location through phi nodes for the user. 1228 class upward_defs_iterator 1229 : public iterator_facade_base<upward_defs_iterator, 1230 std::forward_iterator_tag, 1231 const MemoryAccessPair> { 1232 using BaseT = upward_defs_iterator::iterator_facade_base; 1233 1234 public: 1235 upward_defs_iterator(const MemoryAccessPair &Info, DominatorTree *DT) 1236 : DefIterator(Info.first), Location(Info.second), 1237 OriginalAccess(Info.first), DT(DT) { 1238 CurrentPair.first = nullptr; 1239 1240 WalkingPhi = Info.first && isa<MemoryPhi>(Info.first); 1241 fillInCurrentPair(); 1242 } 1243 1244 upward_defs_iterator() { CurrentPair.first = nullptr; } 1245 1246 bool operator==(const upward_defs_iterator &Other) const { 1247 return DefIterator == Other.DefIterator; 1248 } 1249 1250 typename std::iterator_traits<BaseT>::reference operator*() const { 1251 assert(DefIterator != OriginalAccess->defs_end() && 1252 "Tried to access past the end of our iterator"); 1253 return CurrentPair; 1254 } 1255 1256 using BaseT::operator++; 1257 upward_defs_iterator &operator++() { 1258 assert(DefIterator != OriginalAccess->defs_end() && 1259 "Tried to access past the end of the iterator"); 1260 ++DefIterator; 1261 if (DefIterator != OriginalAccess->defs_end()) 1262 fillInCurrentPair(); 1263 return *this; 1264 } 1265 1266 BasicBlock *getPhiArgBlock() const { return DefIterator.getPhiArgBlock(); } 1267 1268 private: 1269 /// Returns true if \p Ptr is guaranteed to be loop invariant for any possible 1270 /// loop. In particular, this guarantees that it only references a single 1271 /// MemoryLocation during execution of the containing function. 1272 LLVM_ABI bool IsGuaranteedLoopInvariant(const Value *Ptr) const; 1273 1274 void fillInCurrentPair() { 1275 CurrentPair.first = *DefIterator; 1276 CurrentPair.second = Location; 1277 if (WalkingPhi && Location.Ptr) { 1278 PHITransAddr Translator( 1279 const_cast<Value *>(Location.Ptr), 1280 OriginalAccess->getBlock()->getDataLayout(), nullptr); 1281 1282 if (Value *Addr = 1283 Translator.translateValue(OriginalAccess->getBlock(), 1284 DefIterator.getPhiArgBlock(), DT, true)) 1285 if (Addr != CurrentPair.second.Ptr) 1286 CurrentPair.second = CurrentPair.second.getWithNewPtr(Addr); 1287 1288 // Mark size as unknown, if the location is not guaranteed to be 1289 // loop-invariant for any possible loop in the function. Setting the size 1290 // to unknown guarantees that any memory accesses that access locations 1291 // after the pointer are considered as clobbers, which is important to 1292 // catch loop carried dependences. 1293 if (!IsGuaranteedLoopInvariant(CurrentPair.second.Ptr)) 1294 CurrentPair.second = CurrentPair.second.getWithNewSize( 1295 LocationSize::beforeOrAfterPointer()); 1296 } 1297 } 1298 1299 MemoryAccessPair CurrentPair; 1300 memoryaccess_def_iterator DefIterator; 1301 MemoryLocation Location; 1302 MemoryAccess *OriginalAccess = nullptr; 1303 DominatorTree *DT = nullptr; 1304 bool WalkingPhi = false; 1305 }; 1306 1307 inline upward_defs_iterator 1308 upward_defs_begin(const MemoryAccessPair &Pair, DominatorTree &DT) { 1309 return upward_defs_iterator(Pair, &DT); 1310 } 1311 1312 inline upward_defs_iterator upward_defs_end() { return upward_defs_iterator(); } 1313 1314 inline iterator_range<upward_defs_iterator> 1315 upward_defs(const MemoryAccessPair &Pair, DominatorTree &DT) { 1316 return make_range(upward_defs_begin(Pair, DT), upward_defs_end()); 1317 } 1318 1319 /// Walks the defining accesses of MemoryDefs. Stops after we hit something that 1320 /// has no defining use (e.g. a MemoryPhi or liveOnEntry). Note that, when 1321 /// comparing against a null def_chain_iterator, this will compare equal only 1322 /// after walking said Phi/liveOnEntry. 1323 /// 1324 /// The UseOptimizedChain flag specifies whether to walk the clobbering 1325 /// access chain, or all the accesses. 1326 /// 1327 /// Normally, MemoryDef are all just def/use linked together, so a def_chain on 1328 /// a MemoryDef will walk all MemoryDefs above it in the program until it hits 1329 /// a phi node. The optimized chain walks the clobbering access of a store. 1330 /// So if you are just trying to find, given a store, what the next 1331 /// thing that would clobber the same memory is, you want the optimized chain. 1332 template <class T, bool UseOptimizedChain = false> 1333 struct def_chain_iterator 1334 : public iterator_facade_base<def_chain_iterator<T, UseOptimizedChain>, 1335 std::forward_iterator_tag, MemoryAccess *> { 1336 def_chain_iterator() : MA(nullptr) {} 1337 def_chain_iterator(T MA) : MA(MA) {} 1338 1339 T operator*() const { return MA; } 1340 1341 def_chain_iterator &operator++() { 1342 // N.B. liveOnEntry has a null defining access. 1343 if (auto *MUD = dyn_cast<MemoryUseOrDef>(MA)) { 1344 if (UseOptimizedChain && MUD->isOptimized()) 1345 MA = MUD->getOptimized(); 1346 else 1347 MA = MUD->getDefiningAccess(); 1348 } else { 1349 MA = nullptr; 1350 } 1351 1352 return *this; 1353 } 1354 1355 bool operator==(const def_chain_iterator &O) const { return MA == O.MA; } 1356 1357 private: 1358 T MA; 1359 }; 1360 1361 template <class T> 1362 inline iterator_range<def_chain_iterator<T>> 1363 def_chain(T MA, MemoryAccess *UpTo = nullptr) { 1364 #ifdef EXPENSIVE_CHECKS 1365 assert((!UpTo || find(def_chain(MA), UpTo) != def_chain_iterator<T>()) && 1366 "UpTo isn't in the def chain!"); 1367 #endif 1368 return make_range(def_chain_iterator<T>(MA), def_chain_iterator<T>(UpTo)); 1369 } 1370 1371 template <class T> 1372 inline iterator_range<def_chain_iterator<T, true>> optimized_def_chain(T MA) { 1373 return make_range(def_chain_iterator<T, true>(MA), 1374 def_chain_iterator<T, true>(nullptr)); 1375 } 1376 1377 } // end namespace llvm 1378 1379 #endif // LLVM_ANALYSIS_MEMORYSSA_H 1380