FoldingSet.h revision dbac071050010effb95febae0ddf72102551323d
1//===-- llvm/ADT/FoldingSet.h - Uniquing Hash Set ---------------*- C++ -*-===// 2// 3// The LLVM Compiler Infrastructure 4// 5// This file is distributed under the University of Illinois Open Source 6// License. See LICENSE.TXT for details. 7// 8//===----------------------------------------------------------------------===// 9// 10// This file defines a hash set that can be used to remove duplication of nodes 11// in a graph. This code was originally created by Chris Lattner for use with 12// SelectionDAGCSEMap, but was isolated to provide use across the llvm code set. 13// 14//===----------------------------------------------------------------------===// 15 16#ifndef LLVM_ADT_FOLDINGSET_H 17#define LLVM_ADT_FOLDINGSET_H 18 19#include "llvm/System/DataTypes.h" 20#include "llvm/ADT/SmallVector.h" 21#include "llvm/ADT/StringRef.h" 22 23namespace llvm { 24 class APFloat; 25 class APInt; 26 class BumpPtrAllocator; 27 28/// This folding set used for two purposes: 29/// 1. Given information about a node we want to create, look up the unique 30/// instance of the node in the set. If the node already exists, return 31/// it, otherwise return the bucket it should be inserted into. 32/// 2. Given a node that has already been created, remove it from the set. 33/// 34/// This class is implemented as a single-link chained hash table, where the 35/// "buckets" are actually the nodes themselves (the next pointer is in the 36/// node). The last node points back to the bucket to simplify node removal. 37/// 38/// Any node that is to be included in the folding set must be a subclass of 39/// FoldingSetNode. The node class must also define a Profile method used to 40/// establish the unique bits of data for the node. The Profile method is 41/// passed a FoldingSetNodeID object which is used to gather the bits. Just 42/// call one of the Add* functions defined in the FoldingSetImpl::NodeID class. 43/// NOTE: That the folding set does not own the nodes and it is the 44/// responsibility of the user to dispose of the nodes. 45/// 46/// Eg. 47/// class MyNode : public FoldingSetNode { 48/// private: 49/// std::string Name; 50/// unsigned Value; 51/// public: 52/// MyNode(const char *N, unsigned V) : Name(N), Value(V) {} 53/// ... 54/// void Profile(FoldingSetNodeID &ID) const { 55/// ID.AddString(Name); 56/// ID.AddInteger(Value); 57/// } 58/// ... 59/// }; 60/// 61/// To define the folding set itself use the FoldingSet template; 62/// 63/// Eg. 64/// FoldingSet<MyNode> MyFoldingSet; 65/// 66/// Four public methods are available to manipulate the folding set; 67/// 68/// 1) If you have an existing node that you want add to the set but unsure 69/// that the node might already exist then call; 70/// 71/// MyNode *M = MyFoldingSet.GetOrInsertNode(N); 72/// 73/// If The result is equal to the input then the node has been inserted. 74/// Otherwise, the result is the node existing in the folding set, and the 75/// input can be discarded (use the result instead.) 76/// 77/// 2) If you are ready to construct a node but want to check if it already 78/// exists, then call FindNodeOrInsertPos with a FoldingSetNodeID of the bits to 79/// check; 80/// 81/// FoldingSetNodeID ID; 82/// ID.AddString(Name); 83/// ID.AddInteger(Value); 84/// void *InsertPoint; 85/// 86/// MyNode *M = MyFoldingSet.FindNodeOrInsertPos(ID, InsertPoint); 87/// 88/// If found then M with be non-NULL, else InsertPoint will point to where it 89/// should be inserted using InsertNode. 90/// 91/// 3) If you get a NULL result from FindNodeOrInsertPos then you can as a new 92/// node with FindNodeOrInsertPos; 93/// 94/// InsertNode(N, InsertPoint); 95/// 96/// 4) Finally, if you want to remove a node from the folding set call; 97/// 98/// bool WasRemoved = RemoveNode(N); 99/// 100/// The result indicates whether the node existed in the folding set. 101 102class FoldingSetNodeID; 103 104//===----------------------------------------------------------------------===// 105/// FoldingSetImpl - Implements the folding set functionality. The main 106/// structure is an array of buckets. Each bucket is indexed by the hash of 107/// the nodes it contains. The bucket itself points to the nodes contained 108/// in the bucket via a singly linked list. The last node in the list points 109/// back to the bucket to facilitate node removal. 110/// 111class FoldingSetImpl { 112protected: 113 /// Buckets - Array of bucket chains. 114 /// 115 void **Buckets; 116 117 /// NumBuckets - Length of the Buckets array. Always a power of 2. 118 /// 119 unsigned NumBuckets; 120 121 /// NumNodes - Number of nodes in the folding set. Growth occurs when NumNodes 122 /// is greater than twice the number of buckets. 123 unsigned NumNodes; 124 125public: 126 explicit FoldingSetImpl(unsigned Log2InitSize = 6); 127 virtual ~FoldingSetImpl(); 128 129 //===--------------------------------------------------------------------===// 130 /// Node - This class is used to maintain the singly linked bucket list in 131 /// a folding set. 132 /// 133 class Node { 134 private: 135 // NextInFoldingSetBucket - next link in the bucket list. 136 void *NextInFoldingSetBucket; 137 138 public: 139 140 Node() : NextInFoldingSetBucket(0) {} 141 142 // Accessors 143 void *getNextInBucket() const { return NextInFoldingSetBucket; } 144 void SetNextInBucket(void *N) { NextInFoldingSetBucket = N; } 145 }; 146 147 /// clear - Remove all nodes from the folding set. 148 void clear(); 149 150 /// RemoveNode - Remove a node from the folding set, returning true if one 151 /// was removed or false if the node was not in the folding set. 152 bool RemoveNode(Node *N); 153 154 /// GetOrInsertNode - If there is an existing simple Node exactly 155 /// equal to the specified node, return it. Otherwise, insert 'N' and return 156 /// it instead. 157 Node *GetOrInsertNode(Node *N); 158 159 /// FindNodeOrInsertPos - Look up the node specified by ID. If it exists, 160 /// return it. If not, return the insertion token that will make insertion 161 /// faster. 162 Node *FindNodeOrInsertPos(const FoldingSetNodeID &ID, void *&InsertPos); 163 164 /// InsertNode - Insert the specified node into the folding set, knowing that 165 /// it is not already in the folding set. InsertPos must be obtained from 166 /// FindNodeOrInsertPos. 167 void InsertNode(Node *N, void *InsertPos); 168 169 /// InsertNode - Insert the specified node into the folding set, knowing that 170 /// it is not already in the folding set. 171 void InsertNode(Node *N) { 172 Node *Inserted = GetOrInsertNode(N); 173 (void)Inserted; 174 assert(Inserted == N && "Node already inserted!"); 175 } 176 177 /// size - Returns the number of nodes in the folding set. 178 unsigned size() const { return NumNodes; } 179 180 /// empty - Returns true if there are no nodes in the folding set. 181 bool empty() const { return NumNodes == 0; } 182 183private: 184 185 /// GrowHashTable - Double the size of the hash table and rehash everything. 186 /// 187 void GrowHashTable(); 188 189protected: 190 191 /// GetNodeProfile - Instantiations of the FoldingSet template implement 192 /// this function to gather data bits for the given node. 193 virtual void GetNodeProfile(FoldingSetNodeID &ID, Node *N) const = 0; 194}; 195 196//===----------------------------------------------------------------------===// 197/// FoldingSetTrait - This trait class is used to define behavior of how 198/// to "profile" (in the FoldingSet parlance) an object of a given type. 199/// The default behavior is to invoke a 'Profile' method on an object, but 200/// through template specialization the behavior can be tailored for specific 201/// types. Combined with the FoldingSetNodeWrapper class, one can add objects 202/// to FoldingSets that were not originally designed to have that behavior. 203/// 204template<typename T> struct FoldingSetTrait { 205 static inline void Profile(const T& X, FoldingSetNodeID& ID) { X.Profile(ID);} 206 static inline void Profile(T& X, FoldingSetNodeID& ID) { X.Profile(ID); } 207 template <typename Ctx> 208 static inline void Profile(T &X, FoldingSetNodeID &ID, Ctx Context) { 209 X.Profile(ID, Context); 210 } 211}; 212 213//===--------------------------------------------------------------------===// 214/// FoldingSetNodeIDRef - This class describes a reference to an interned 215/// FoldingSetNodeID, which can be a useful to store node id data rather 216/// than using plain FoldingSetNodeIDs, since the 32-element SmallVector 217/// is often much larger than necessary, and the possibility of heap 218/// allocation means it requires a non-trivial destructor call. 219class FoldingSetNodeIDRef { 220 unsigned* Data; 221 size_t Size; 222public: 223 FoldingSetNodeIDRef() : Data(0), Size(0) {} 224 FoldingSetNodeIDRef(unsigned *D, size_t S) : Data(D), Size(S) {} 225 226 unsigned *getData() const { return Data; } 227 size_t getSize() const { return Size; } 228}; 229 230//===--------------------------------------------------------------------===// 231/// FoldingSetNodeID - This class is used to gather all the unique data bits of 232/// a node. When all the bits are gathered this class is used to produce a 233/// hash value for the node. 234/// 235class FoldingSetNodeID { 236 /// Bits - Vector of all the data bits that make the node unique. 237 /// Use a SmallVector to avoid a heap allocation in the common case. 238 SmallVector<unsigned, 32> Bits; 239 240public: 241 FoldingSetNodeID() {} 242 243 FoldingSetNodeID(FoldingSetNodeIDRef Ref) 244 : Bits(Ref.getData(), Ref.getData() + Ref.getSize()) {} 245 246 /// Add* - Add various data types to Bit data. 247 /// 248 void AddPointer(const void *Ptr); 249 void AddInteger(signed I); 250 void AddInteger(unsigned I); 251 void AddInteger(long I); 252 void AddInteger(unsigned long I); 253 void AddInteger(long long I); 254 void AddInteger(unsigned long long I); 255 void AddBoolean(bool B) { AddInteger(B ? 1U : 0U); } 256 void AddString(StringRef String); 257 258 template <typename T> 259 inline void Add(const T& x) { FoldingSetTrait<T>::Profile(x, *this); } 260 261 /// clear - Clear the accumulated profile, allowing this FoldingSetNodeID 262 /// object to be used to compute a new profile. 263 inline void clear() { Bits.clear(); } 264 265 /// ComputeHash - Compute a strong hash value for this FoldingSetNodeID, used 266 /// to lookup the node in the FoldingSetImpl. 267 unsigned ComputeHash() const; 268 269 /// operator== - Used to compare two nodes to each other. 270 /// 271 bool operator==(const FoldingSetNodeID &RHS) const; 272 273 /// Intern - Copy this node's data to a memory region allocated from the 274 /// given allocator and return a FoldingSetNodeIDRef describing the 275 /// interned data. 276 FoldingSetNodeIDRef Intern(BumpPtrAllocator &Allocator) const; 277}; 278 279// Convenience type to hide the implementation of the folding set. 280typedef FoldingSetImpl::Node FoldingSetNode; 281template<class T> class FoldingSetIterator; 282template<class T> class FoldingSetBucketIterator; 283 284//===----------------------------------------------------------------------===// 285/// FoldingSet - This template class is used to instantiate a specialized 286/// implementation of the folding set to the node class T. T must be a 287/// subclass of FoldingSetNode and implement a Profile function. 288/// 289template<class T> class FoldingSet : public FoldingSetImpl { 290private: 291 /// GetNodeProfile - Each instantiatation of the FoldingSet needs to provide a 292 /// way to convert nodes into a unique specifier. 293 virtual void GetNodeProfile(FoldingSetNodeID &ID, Node *N) const { 294 T *TN = static_cast<T *>(N); 295 FoldingSetTrait<T>::Profile(*TN,ID); 296 } 297 298public: 299 explicit FoldingSet(unsigned Log2InitSize = 6) 300 : FoldingSetImpl(Log2InitSize) 301 {} 302 303 typedef FoldingSetIterator<T> iterator; 304 iterator begin() { return iterator(Buckets); } 305 iterator end() { return iterator(Buckets+NumBuckets); } 306 307 typedef FoldingSetIterator<const T> const_iterator; 308 const_iterator begin() const { return const_iterator(Buckets); } 309 const_iterator end() const { return const_iterator(Buckets+NumBuckets); } 310 311 typedef FoldingSetBucketIterator<T> bucket_iterator; 312 313 bucket_iterator bucket_begin(unsigned hash) { 314 return bucket_iterator(Buckets + (hash & (NumBuckets-1))); 315 } 316 317 bucket_iterator bucket_end(unsigned hash) { 318 return bucket_iterator(Buckets + (hash & (NumBuckets-1)), true); 319 } 320 321 /// GetOrInsertNode - If there is an existing simple Node exactly 322 /// equal to the specified node, return it. Otherwise, insert 'N' and 323 /// return it instead. 324 T *GetOrInsertNode(Node *N) { 325 return static_cast<T *>(FoldingSetImpl::GetOrInsertNode(N)); 326 } 327 328 /// FindNodeOrInsertPos - Look up the node specified by ID. If it exists, 329 /// return it. If not, return the insertion token that will make insertion 330 /// faster. 331 T *FindNodeOrInsertPos(const FoldingSetNodeID &ID, void *&InsertPos) { 332 return static_cast<T *>(FoldingSetImpl::FindNodeOrInsertPos(ID, InsertPos)); 333 } 334}; 335 336//===----------------------------------------------------------------------===// 337/// ContextualFoldingSet - This template class is a further refinement 338/// of FoldingSet which provides a context argument when calling 339/// Profile on its nodes. Currently, that argument is fixed at 340/// initialization time. 341/// 342/// T must be a subclass of FoldingSetNode and implement a Profile 343/// function with signature 344/// void Profile(llvm::FoldingSetNodeID &, Ctx); 345template <class T, class Ctx> 346class ContextualFoldingSet : public FoldingSetImpl { 347 // Unfortunately, this can't derive from FoldingSet<T> because the 348 // construction vtable for FoldingSet<T> requires 349 // FoldingSet<T>::GetNodeProfile to be instantiated, which in turn 350 // requires a single-argument T::Profile(). 351 352private: 353 Ctx Context; 354 355 /// GetNodeProfile - Each instantiatation of the FoldingSet needs to provide a 356 /// way to convert nodes into a unique specifier. 357 virtual void GetNodeProfile(FoldingSetNodeID &ID, 358 FoldingSetImpl::Node *N) const { 359 T *TN = static_cast<T *>(N); 360 361 // We must use explicit template arguments in case Ctx is a 362 // reference type. 363 FoldingSetTrait<T>::template Profile<Ctx>(*TN, ID, Context); 364 } 365 366public: 367 explicit ContextualFoldingSet(Ctx Context, unsigned Log2InitSize = 6) 368 : FoldingSetImpl(Log2InitSize), Context(Context) 369 {} 370 371 Ctx getContext() const { return Context; } 372 373 374 typedef FoldingSetIterator<T> iterator; 375 iterator begin() { return iterator(Buckets); } 376 iterator end() { return iterator(Buckets+NumBuckets); } 377 378 typedef FoldingSetIterator<const T> const_iterator; 379 const_iterator begin() const { return const_iterator(Buckets); } 380 const_iterator end() const { return const_iterator(Buckets+NumBuckets); } 381 382 typedef FoldingSetBucketIterator<T> bucket_iterator; 383 384 bucket_iterator bucket_begin(unsigned hash) { 385 return bucket_iterator(Buckets + (hash & (NumBuckets-1))); 386 } 387 388 bucket_iterator bucket_end(unsigned hash) { 389 return bucket_iterator(Buckets + (hash & (NumBuckets-1)), true); 390 } 391 392 /// GetOrInsertNode - If there is an existing simple Node exactly 393 /// equal to the specified node, return it. Otherwise, insert 'N' 394 /// and return it instead. 395 T *GetOrInsertNode(Node *N) { 396 return static_cast<T *>(FoldingSetImpl::GetOrInsertNode(N)); 397 } 398 399 /// FindNodeOrInsertPos - Look up the node specified by ID. If it 400 /// exists, return it. If not, return the insertion token that will 401 /// make insertion faster. 402 T *FindNodeOrInsertPos(const FoldingSetNodeID &ID, void *&InsertPos) { 403 return static_cast<T *>(FoldingSetImpl::FindNodeOrInsertPos(ID, InsertPos)); 404 } 405}; 406 407//===----------------------------------------------------------------------===// 408/// FoldingSetIteratorImpl - This is the common iterator support shared by all 409/// folding sets, which knows how to walk the folding set hash table. 410class FoldingSetIteratorImpl { 411protected: 412 FoldingSetNode *NodePtr; 413 FoldingSetIteratorImpl(void **Bucket); 414 void advance(); 415 416public: 417 bool operator==(const FoldingSetIteratorImpl &RHS) const { 418 return NodePtr == RHS.NodePtr; 419 } 420 bool operator!=(const FoldingSetIteratorImpl &RHS) const { 421 return NodePtr != RHS.NodePtr; 422 } 423}; 424 425 426template<class T> 427class FoldingSetIterator : public FoldingSetIteratorImpl { 428public: 429 explicit FoldingSetIterator(void **Bucket) : FoldingSetIteratorImpl(Bucket) {} 430 431 T &operator*() const { 432 return *static_cast<T*>(NodePtr); 433 } 434 435 T *operator->() const { 436 return static_cast<T*>(NodePtr); 437 } 438 439 inline FoldingSetIterator& operator++() { // Preincrement 440 advance(); 441 return *this; 442 } 443 FoldingSetIterator operator++(int) { // Postincrement 444 FoldingSetIterator tmp = *this; ++*this; return tmp; 445 } 446}; 447 448//===----------------------------------------------------------------------===// 449/// FoldingSetBucketIteratorImpl - This is the common bucket iterator support 450/// shared by all folding sets, which knows how to walk a particular bucket 451/// of a folding set hash table. 452 453class FoldingSetBucketIteratorImpl { 454protected: 455 void *Ptr; 456 457 explicit FoldingSetBucketIteratorImpl(void **Bucket); 458 459 FoldingSetBucketIteratorImpl(void **Bucket, bool) 460 : Ptr(Bucket) {} 461 462 void advance() { 463 void *Probe = static_cast<FoldingSetNode*>(Ptr)->getNextInBucket(); 464 uintptr_t x = reinterpret_cast<uintptr_t>(Probe) & ~0x1; 465 Ptr = reinterpret_cast<void*>(x); 466 } 467 468public: 469 bool operator==(const FoldingSetBucketIteratorImpl &RHS) const { 470 return Ptr == RHS.Ptr; 471 } 472 bool operator!=(const FoldingSetBucketIteratorImpl &RHS) const { 473 return Ptr != RHS.Ptr; 474 } 475}; 476 477 478template<class T> 479class FoldingSetBucketIterator : public FoldingSetBucketIteratorImpl { 480public: 481 explicit FoldingSetBucketIterator(void **Bucket) : 482 FoldingSetBucketIteratorImpl(Bucket) {} 483 484 FoldingSetBucketIterator(void **Bucket, bool) : 485 FoldingSetBucketIteratorImpl(Bucket, true) {} 486 487 T& operator*() const { return *static_cast<T*>(Ptr); } 488 T* operator->() const { return static_cast<T*>(Ptr); } 489 490 inline FoldingSetBucketIterator& operator++() { // Preincrement 491 advance(); 492 return *this; 493 } 494 FoldingSetBucketIterator operator++(int) { // Postincrement 495 FoldingSetBucketIterator tmp = *this; ++*this; return tmp; 496 } 497}; 498 499//===----------------------------------------------------------------------===// 500/// FoldingSetNodeWrapper - This template class is used to "wrap" arbitrary 501/// types in an enclosing object so that they can be inserted into FoldingSets. 502template <typename T> 503class FoldingSetNodeWrapper : public FoldingSetNode { 504 T data; 505public: 506 explicit FoldingSetNodeWrapper(const T& x) : data(x) {} 507 virtual ~FoldingSetNodeWrapper() {} 508 509 template<typename A1> 510 explicit FoldingSetNodeWrapper(const A1& a1) 511 : data(a1) {} 512 513 template <typename A1, typename A2> 514 explicit FoldingSetNodeWrapper(const A1& a1, const A2& a2) 515 : data(a1,a2) {} 516 517 template <typename A1, typename A2, typename A3> 518 explicit FoldingSetNodeWrapper(const A1& a1, const A2& a2, const A3& a3) 519 : data(a1,a2,a3) {} 520 521 template <typename A1, typename A2, typename A3, typename A4> 522 explicit FoldingSetNodeWrapper(const A1& a1, const A2& a2, const A3& a3, 523 const A4& a4) 524 : data(a1,a2,a3,a4) {} 525 526 template <typename A1, typename A2, typename A3, typename A4, typename A5> 527 explicit FoldingSetNodeWrapper(const A1& a1, const A2& a2, const A3& a3, 528 const A4& a4, const A5& a5) 529 : data(a1,a2,a3,a4,a5) {} 530 531 532 void Profile(FoldingSetNodeID& ID) { FoldingSetTrait<T>::Profile(data, ID); } 533 534 T& getValue() { return data; } 535 const T& getValue() const { return data; } 536 537 operator T&() { return data; } 538 operator const T&() const { return data; } 539}; 540 541//===----------------------------------------------------------------------===// 542/// FastFoldingSetNode - This is a subclass of FoldingSetNode which stores 543/// a FoldingSetNodeID value rather than requiring the node to recompute it 544/// each time it is needed. This trades space for speed (which can be 545/// significant if the ID is long), and it also permits nodes to drop 546/// information that would otherwise only be required for recomputing an ID. 547class FastFoldingSetNode : public FoldingSetNode { 548 FoldingSetNodeID FastID; 549protected: 550 explicit FastFoldingSetNode(const FoldingSetNodeID &ID) : FastID(ID) {} 551public: 552 void Profile(FoldingSetNodeID& ID) { ID = FastID; } 553}; 554 555//===----------------------------------------------------------------------===// 556// Partial specializations of FoldingSetTrait. 557 558template<typename T> struct FoldingSetTrait<T*> { 559 static inline void Profile(const T* X, FoldingSetNodeID& ID) { 560 ID.AddPointer(X); 561 } 562 static inline void Profile(T* X, FoldingSetNodeID& ID) { 563 ID.AddPointer(X); 564 } 565}; 566 567template<typename T> struct FoldingSetTrait<const T*> { 568 static inline void Profile(const T* X, FoldingSetNodeID& ID) { 569 ID.AddPointer(X); 570 } 571}; 572 573} // End of namespace llvm. 574 575#endif 576