| 1 | /* |
| 2 | * Copyright 2015 Google Inc. |
| 3 | * |
| 4 | * Use of this source code is governed by a BSD-style license that can be |
| 5 | * found in the LICENSE file. |
| 6 | */ |
| 7 | |
| 8 | #ifndef SkTHash_DEFINED |
| 9 | #define SkTHash_DEFINED |
| 10 | |
| 11 | #include "include/core/SkTypes.h" |
| 12 | #include "include/private/SkChecksum.h" |
| 13 | #include "include/private/SkTemplates.h" |
| 14 | #include <new> |
| 15 | |
| 16 | // Before trying to use SkTHashTable, look below to see if SkTHashMap or SkTHashSet works for you. |
| 17 | // They're easier to use, usually perform the same, and have fewer sharp edges. |
| 18 | |
| 19 | // T and K are treated as ordinary copyable C++ types. |
| 20 | // Traits must have: |
| 21 | // - static K GetKey(T) |
| 22 | // - static uint32_t Hash(K) |
| 23 | // If the key is large and stored inside T, you may want to make K a const&. |
| 24 | // Similarly, if T is large you might want it to be a pointer. |
| 25 | template <typename T, typename K, typename Traits = T> |
| 26 | class SkTHashTable { |
| 27 | public: |
| 28 | SkTHashTable() : fCount(0), fCapacity(0) {} |
| 29 | SkTHashTable(SkTHashTable&& other) |
| 30 | : fCount(other.fCount) |
| 31 | , fCapacity(other.fCapacity) |
| 32 | , fSlots(std::move(other.fSlots)) { other.fCount = other.fCapacity = 0; } |
| 33 | |
| 34 | SkTHashTable& operator=(SkTHashTable&& other) { |
| 35 | if (this != &other) { |
| 36 | this->~SkTHashTable(); |
| 37 | new (this) SkTHashTable(std::move(other)); |
| 38 | } |
| 39 | return *this; |
| 40 | } |
| 41 | |
| 42 | // Clear the table. |
| 43 | void reset() { *this = SkTHashTable(); } |
| 44 | |
| 45 | // How many entries are in the table? |
| 46 | int count() const { return fCount; } |
| 47 | |
| 48 | // Approximately how many bytes of memory do we use beyond sizeof(*this)? |
| 49 | size_t approxBytesUsed() const { return fCapacity * sizeof(Slot); } |
| 50 | |
| 51 | // !!!!!!!!!!!!!!!!! CAUTION !!!!!!!!!!!!!!!!! |
| 52 | // set(), find() and foreach() all allow mutable access to table entries. |
| 53 | // If you change an entry so that it no longer has the same key, all hell |
| 54 | // will break loose. Do not do that! |
| 55 | // |
| 56 | // Please prefer to use SkTHashMap or SkTHashSet, which do not have this danger. |
| 57 | |
| 58 | // The pointers returned by set() and find() are valid only until the next call to set(). |
| 59 | // The pointers you receive in foreach() are only valid for its duration. |
| 60 | |
| 61 | // Copy val into the hash table, returning a pointer to the copy now in the table. |
| 62 | // If there already is an entry in the table with the same key, we overwrite it. |
| 63 | T* set(T val) { |
| 64 | if (4 * fCount >= 3 * fCapacity) { |
| 65 | this->resize(fCapacity > 0 ? fCapacity * 2 : 4); |
| 66 | } |
| 67 | return this->uncheckedSet(std::move(val)); |
| 68 | } |
| 69 | |
| 70 | // If there is an entry in the table with this key, return a pointer to it. If not, null. |
| 71 | T* find(const K& key) const { |
| 72 | uint32_t hash = Hash(key); |
| 73 | int index = hash & (fCapacity-1); |
| 74 | for (int n = 0; n < fCapacity; n++) { |
| 75 | Slot& s = fSlots[index]; |
| 76 | if (s.empty()) { |
| 77 | return nullptr; |
| 78 | } |
| 79 | if (hash == s.hash && key == Traits::GetKey(s.val)) { |
| 80 | return &s.val; |
| 81 | } |
| 82 | index = this->next(index); |
| 83 | } |
| 84 | SkASSERT(fCapacity == 0); |
| 85 | return nullptr; |
| 86 | } |
| 87 | |
| 88 | // If there is an entry in the table with this key, return it. If not, null. |
| 89 | // This only works for pointer type T, and cannot be used to find an nullptr entry. |
| 90 | T findOrNull(const K& key) const { |
| 91 | if (T* p = this->find(key)) { |
| 92 | return *p; |
| 93 | } |
| 94 | return nullptr; |
| 95 | } |
| 96 | |
| 97 | // Remove the value with this key from the hash table. |
| 98 | void remove(const K& key) { |
| 99 | SkASSERT(this->find(key)); |
| 100 | |
| 101 | uint32_t hash = Hash(key); |
| 102 | int index = hash & (fCapacity-1); |
| 103 | for (int n = 0; n < fCapacity; n++) { |
| 104 | Slot& s = fSlots[index]; |
| 105 | SkASSERT(!s.empty()); |
| 106 | if (hash == s.hash && key == Traits::GetKey(s.val)) { |
| 107 | this->removeSlot(index); |
| 108 | if (4 * fCount <= fCapacity && fCapacity > 4) { |
| 109 | this->resize(fCapacity / 2); |
| 110 | } |
| 111 | return; |
| 112 | } |
| 113 | index = this->next(index); |
| 114 | } |
| 115 | } |
| 116 | |
| 117 | // Call fn on every entry in the table. You may mutate the entries, but be very careful. |
| 118 | template <typename Fn> // f(T*) |
| 119 | void foreach(Fn&& fn) { |
| 120 | for (int i = 0; i < fCapacity; i++) { |
| 121 | if (!fSlots[i].empty()) { |
| 122 | fn(&fSlots[i].val); |
| 123 | } |
| 124 | } |
| 125 | } |
| 126 | |
| 127 | // Call fn on every entry in the table. You may not mutate anything. |
| 128 | template <typename Fn> // f(T) or f(const T&) |
| 129 | void foreach(Fn&& fn) const { |
| 130 | for (int i = 0; i < fCapacity; i++) { |
| 131 | if (!fSlots[i].empty()) { |
| 132 | fn(fSlots[i].val); |
| 133 | } |
| 134 | } |
| 135 | } |
| 136 | |
| 137 | private: |
| 138 | T* uncheckedSet(T&& val) { |
| 139 | const K& key = Traits::GetKey(val); |
| 140 | uint32_t hash = Hash(key); |
| 141 | int index = hash & (fCapacity-1); |
| 142 | for (int n = 0; n < fCapacity; n++) { |
| 143 | Slot& s = fSlots[index]; |
| 144 | if (s.empty()) { |
| 145 | // New entry. |
| 146 | s.val = std::move(val); |
| 147 | s.hash = hash; |
| 148 | fCount++; |
| 149 | return &s.val; |
| 150 | } |
| 151 | if (hash == s.hash && key == Traits::GetKey(s.val)) { |
| 152 | // Overwrite previous entry. |
| 153 | // Note: this triggers extra copies when adding the same value repeatedly. |
| 154 | s.val = std::move(val); |
| 155 | return &s.val; |
| 156 | } |
| 157 | |
| 158 | index = this->next(index); |
| 159 | } |
| 160 | SkASSERT(false); |
| 161 | return nullptr; |
| 162 | } |
| 163 | |
| 164 | void resize(int capacity) { |
| 165 | int oldCapacity = fCapacity; |
| 166 | SkDEBUGCODE(int oldCount = fCount); |
| 167 | |
| 168 | fCount = 0; |
| 169 | fCapacity = capacity; |
| 170 | SkAutoTArray<Slot> oldSlots = std::move(fSlots); |
| 171 | fSlots = SkAutoTArray<Slot>(capacity); |
| 172 | |
| 173 | for (int i = 0; i < oldCapacity; i++) { |
| 174 | Slot& s = oldSlots[i]; |
| 175 | if (!s.empty()) { |
| 176 | this->uncheckedSet(std::move(s.val)); |
| 177 | } |
| 178 | } |
| 179 | SkASSERT(fCount == oldCount); |
| 180 | } |
| 181 | |
| 182 | void removeSlot(int index) { |
| 183 | fCount--; |
| 184 | |
| 185 | // Rearrange elements to restore the invariants for linear probing. |
| 186 | for (;;) { |
| 187 | Slot& emptySlot = fSlots[index]; |
| 188 | int emptyIndex = index; |
| 189 | int originalIndex; |
| 190 | // Look for an element that can be moved into the empty slot. |
| 191 | // If the empty slot is in between where an element landed, and its native slot, then |
| 192 | // move it to the empty slot. Don't move it if its native slot is in between where |
| 193 | // the element landed and the empty slot. |
| 194 | // [native] <= [empty] < [candidate] == GOOD, can move candidate to empty slot |
| 195 | // [empty] < [native] < [candidate] == BAD, need to leave candidate where it is |
| 196 | do { |
| 197 | index = this->next(index); |
| 198 | Slot& s = fSlots[index]; |
| 199 | if (s.empty()) { |
| 200 | // We're done shuffling elements around. Clear the last empty slot. |
| 201 | emptySlot = Slot(); |
| 202 | return; |
| 203 | } |
| 204 | originalIndex = s.hash & (fCapacity - 1); |
| 205 | } while ((index <= originalIndex && originalIndex < emptyIndex) |
| 206 | || (originalIndex < emptyIndex && emptyIndex < index) |
| 207 | || (emptyIndex < index && index <= originalIndex)); |
| 208 | // Move the element to the empty slot. |
| 209 | Slot& moveFrom = fSlots[index]; |
| 210 | emptySlot = std::move(moveFrom); |
| 211 | } |
| 212 | } |
| 213 | |
| 214 | int next(int index) const { |
| 215 | index--; |
| 216 | if (index < 0) { index += fCapacity; } |
| 217 | return index; |
| 218 | } |
| 219 | |
| 220 | static uint32_t Hash(const K& key) { |
| 221 | uint32_t hash = Traits::Hash(key) & 0xffffffff; |
| 222 | return hash ? hash : 1; // We reserve hash 0 to mark empty. |
| 223 | } |
| 224 | |
| 225 | struct Slot { |
| 226 | Slot() : val{}, hash(0) {} |
| 227 | Slot(T&& v, uint32_t h) : val(std::move(v)), hash(h) {} |
| 228 | Slot(Slot&& o) { *this = std::move(o); } |
| 229 | Slot& operator=(Slot&& o) { |
| 230 | val = std::move(o.val); |
| 231 | hash = o.hash; |
| 232 | return *this; |
| 233 | } |
| 234 | |
| 235 | bool empty() const { return this->hash == 0; } |
| 236 | |
| 237 | T val; |
| 238 | uint32_t hash; |
| 239 | }; |
| 240 | |
| 241 | int fCount, fCapacity; |
| 242 | SkAutoTArray<Slot> fSlots; |
| 243 | |
| 244 | SkTHashTable(const SkTHashTable&) = delete; |
| 245 | SkTHashTable& operator=(const SkTHashTable&) = delete; |
| 246 | }; |
| 247 | |
| 248 | // Maps K->V. A more user-friendly wrapper around SkTHashTable, suitable for most use cases. |
| 249 | // K and V are treated as ordinary copyable C++ types, with no assumed relationship between the two. |
| 250 | template <typename K, typename V, typename HashK = SkGoodHash> |
| 251 | class SkTHashMap { |
| 252 | public: |
| 253 | SkTHashMap() {} |
| 254 | SkTHashMap(SkTHashMap&&) = default; |
| 255 | SkTHashMap& operator=(SkTHashMap&&) = default; |
| 256 | |
| 257 | // Clear the map. |
| 258 | void reset() { fTable.reset(); } |
| 259 | |
| 260 | // How many key/value pairs are in the table? |
| 261 | int count() const { return fTable.count(); } |
| 262 | |
| 263 | // Approximately how many bytes of memory do we use beyond sizeof(*this)? |
| 264 | size_t approxBytesUsed() const { return fTable.approxBytesUsed(); } |
| 265 | |
| 266 | // N.B. The pointers returned by set() and find() are valid only until the next call to set(). |
| 267 | |
| 268 | // Set key to val in the table, replacing any previous value with the same key. |
| 269 | // We copy both key and val, and return a pointer to the value copy now in the table. |
| 270 | V* set(K key, V val) { |
| 271 | Pair* out = fTable.set({std::move(key), std::move(val)}); |
| 272 | return &out->val; |
| 273 | } |
| 274 | |
| 275 | // If there is key/value entry in the table with this key, return a pointer to the value. |
| 276 | // If not, return null. |
| 277 | V* find(const K& key) const { |
| 278 | if (Pair* p = fTable.find(key)) { |
| 279 | return &p->val; |
| 280 | } |
| 281 | return nullptr; |
| 282 | } |
| 283 | |
| 284 | V& operator[](const K& key) { |
| 285 | if (V* val = this->find(key)) { |
| 286 | return *val; |
| 287 | } |
| 288 | return *this->set(key, V{}); |
| 289 | } |
| 290 | |
| 291 | // Remove the key/value entry in the table with this key. |
| 292 | void remove(const K& key) { |
| 293 | SkASSERT(this->find(key)); |
| 294 | fTable.remove(key); |
| 295 | } |
| 296 | |
| 297 | // Call fn on every key/value pair in the table. You may mutate the value but not the key. |
| 298 | template <typename Fn> // f(K, V*) or f(const K&, V*) |
| 299 | void foreach(Fn&& fn) { |
| 300 | fTable.foreach([&fn](Pair* p){ fn(p->key, &p->val); }); |
| 301 | } |
| 302 | |
| 303 | // Call fn on every key/value pair in the table. You may not mutate anything. |
| 304 | template <typename Fn> // f(K, V), f(const K&, V), f(K, const V&) or f(const K&, const V&). |
| 305 | void foreach(Fn&& fn) const { |
| 306 | fTable.foreach([&fn](const Pair& p){ fn(p.key, p.val); }); |
| 307 | } |
| 308 | |
| 309 | private: |
| 310 | struct Pair { |
| 311 | K key; |
| 312 | V val; |
| 313 | static const K& GetKey(const Pair& p) { return p.key; } |
| 314 | static auto Hash(const K& key) { return HashK()(key); } |
| 315 | }; |
| 316 | |
| 317 | SkTHashTable<Pair, K> fTable; |
| 318 | |
| 319 | SkTHashMap(const SkTHashMap&) = delete; |
| 320 | SkTHashMap& operator=(const SkTHashMap&) = delete; |
| 321 | }; |
| 322 | |
| 323 | // A set of T. T is treated as an ordinary copyable C++ type. |
| 324 | template <typename T, typename HashT = SkGoodHash> |
| 325 | class SkTHashSet { |
| 326 | public: |
| 327 | SkTHashSet() {} |
| 328 | SkTHashSet(SkTHashSet&&) = default; |
| 329 | SkTHashSet& operator=(SkTHashSet&&) = default; |
| 330 | |
| 331 | // Clear the set. |
| 332 | void reset() { fTable.reset(); } |
| 333 | |
| 334 | // How many items are in the set? |
| 335 | int count() const { return fTable.count(); } |
| 336 | |
| 337 | // Is empty? |
| 338 | bool empty() const { return fTable.count() == 0; } |
| 339 | |
| 340 | // Approximately how many bytes of memory do we use beyond sizeof(*this)? |
| 341 | size_t approxBytesUsed() const { return fTable.approxBytesUsed(); } |
| 342 | |
| 343 | // Copy an item into the set. |
| 344 | void add(T item) { fTable.set(std::move(item)); } |
| 345 | |
| 346 | // Is this item in the set? |
| 347 | bool contains(const T& item) const { return SkToBool(this->find(item)); } |
| 348 | |
| 349 | // If an item equal to this is in the set, return a pointer to it, otherwise null. |
| 350 | // This pointer remains valid until the next call to add(). |
| 351 | const T* find(const T& item) const { return fTable.find(item); } |
| 352 | |
| 353 | // Remove the item in the set equal to this. |
| 354 | void remove(const T& item) { |
| 355 | SkASSERT(this->contains(item)); |
| 356 | fTable.remove(item); |
| 357 | } |
| 358 | |
| 359 | // Call fn on every item in the set. You may not mutate anything. |
| 360 | template <typename Fn> // f(T), f(const T&) |
| 361 | void foreach (Fn&& fn) const { |
| 362 | fTable.foreach(fn); |
| 363 | } |
| 364 | |
| 365 | private: |
| 366 | struct Traits { |
| 367 | static const T& GetKey(const T& item) { return item; } |
| 368 | static auto Hash(const T& item) { return HashT()(item); } |
| 369 | }; |
| 370 | SkTHashTable<T, T, Traits> fTable; |
| 371 | |
| 372 | SkTHashSet(const SkTHashSet&) = delete; |
| 373 | SkTHashSet& operator=(const SkTHashSet&) = delete; |
| 374 | }; |
| 375 | |
| 376 | #endif//SkTHash_DEFINED |
| 377 | |