Mathieu Chartier | c2e2062 | 2014-11-03 11:41:47 -0800 | [diff] [blame] | 1 | /* |
| 2 | * Copyright (C) 2014 The Android Open Source Project |
| 3 | * |
| 4 | * Licensed under the Apache License, Version 2.0 (the "License"); |
| 5 | * you may not use this file except in compliance with the License. |
| 6 | * You may obtain a copy of the License at |
| 7 | * |
| 8 | * http://www.apache.org/licenses/LICENSE-2.0 |
| 9 | * |
| 10 | * Unless required by applicable law or agreed to in writing, software |
| 11 | * distributed under the License is distributed on an "AS IS" BASIS, |
| 12 | * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| 13 | * See the License for the specific language governing permissions and |
| 14 | * limitations under the License. |
| 15 | */ |
| 16 | |
| 17 | #ifndef ART_RUNTIME_BASE_HASH_SET_H_ |
| 18 | #define ART_RUNTIME_BASE_HASH_SET_H_ |
| 19 | |
| 20 | #include <functional> |
| 21 | #include <memory> |
| 22 | #include <stdint.h> |
| 23 | #include <utility> |
| 24 | |
| 25 | #include "logging.h" |
| 26 | |
| 27 | namespace art { |
| 28 | |
| 29 | // Returns true if an item is empty. |
| 30 | template <class T> |
| 31 | class DefaultEmptyFn { |
| 32 | public: |
| 33 | void MakeEmpty(T& item) const { |
| 34 | item = T(); |
| 35 | } |
| 36 | bool IsEmpty(const T& item) const { |
| 37 | return item == T(); |
| 38 | } |
| 39 | }; |
| 40 | |
| 41 | template <class T> |
| 42 | class DefaultEmptyFn<T*> { |
| 43 | public: |
| 44 | void MakeEmpty(T*& item) const { |
| 45 | item = nullptr; |
| 46 | } |
| 47 | bool IsEmpty(const T*& item) const { |
| 48 | return item == nullptr; |
| 49 | } |
| 50 | }; |
| 51 | |
| 52 | // Low memory version of a hash set, uses less memory than std::unordered_set since elements aren't |
| 53 | // boxed. Uses linear probing. |
| 54 | // EmptyFn needs to implement two functions MakeEmpty(T& item) and IsEmpty(const T& item) |
| 55 | template <class T, class EmptyFn = DefaultEmptyFn<T>, class HashFn = std::hash<T>, |
| 56 | class Pred = std::equal_to<T>, class Alloc = std::allocator<T>> |
| 57 | class HashSet { |
| 58 | public: |
| 59 | static constexpr double kDefaultMinLoadFactor = 0.5; |
| 60 | static constexpr double kDefaultMaxLoadFactor = 0.9; |
| 61 | static constexpr size_t kMinBuckets = 1000; |
| 62 | |
| 63 | class Iterator { |
| 64 | public: |
| 65 | Iterator(const Iterator&) = default; |
| 66 | Iterator(HashSet* hash_set, size_t index) : hash_set_(hash_set), index_(index) { |
| 67 | } |
| 68 | Iterator& operator=(const Iterator&) = default; |
| 69 | bool operator==(const Iterator& other) const { |
| 70 | return hash_set_ == other.hash_set_ && index_ == other.index_; |
| 71 | } |
| 72 | bool operator!=(const Iterator& other) const { |
| 73 | return !(*this == other); |
| 74 | } |
| 75 | Iterator operator++() { // Value after modification. |
| 76 | index_ = NextNonEmptySlot(index_); |
| 77 | return *this; |
| 78 | } |
| 79 | Iterator operator++(int) { |
| 80 | Iterator temp = *this; |
| 81 | index_ = NextNonEmptySlot(index_); |
| 82 | return temp; |
| 83 | } |
| 84 | T& operator*() { |
| 85 | DCHECK(!hash_set_->IsFreeSlot(GetIndex())); |
| 86 | return hash_set_->ElementForIndex(index_); |
| 87 | } |
| 88 | const T& operator*() const { |
| 89 | DCHECK(!hash_set_->IsFreeSlot(GetIndex())); |
| 90 | return hash_set_->ElementForIndex(index_); |
| 91 | } |
| 92 | T* operator->() { |
| 93 | return &**this; |
| 94 | } |
| 95 | const T* operator->() const { |
| 96 | return &**this; |
| 97 | } |
| 98 | // TODO: Operator -- --(int) |
| 99 | |
| 100 | private: |
| 101 | HashSet* hash_set_; |
| 102 | size_t index_; |
| 103 | |
| 104 | size_t GetIndex() const { |
| 105 | return index_; |
| 106 | } |
| 107 | size_t NextNonEmptySlot(size_t index) const { |
| 108 | const size_t num_buckets = hash_set_->NumBuckets(); |
| 109 | DCHECK_LT(index, num_buckets); |
| 110 | do { |
| 111 | ++index; |
| 112 | } while (index < num_buckets && hash_set_->IsFreeSlot(index)); |
| 113 | return index; |
| 114 | } |
| 115 | |
| 116 | friend class HashSet; |
| 117 | }; |
| 118 | |
| 119 | void Clear() { |
| 120 | DeallocateStorage(); |
| 121 | AllocateStorage(1); |
| 122 | num_elements_ = 0; |
| 123 | elements_until_expand_ = 0; |
| 124 | } |
| 125 | HashSet() : num_elements_(0), num_buckets_(0), data_(nullptr), |
| 126 | min_load_factor_(kDefaultMinLoadFactor), max_load_factor_(kDefaultMaxLoadFactor) { |
| 127 | Clear(); |
| 128 | } |
| 129 | HashSet(const HashSet& other) : num_elements_(0), num_buckets_(0), data_(nullptr) { |
| 130 | *this = other; |
| 131 | } |
| 132 | HashSet(HashSet&& other) : num_elements_(0), num_buckets_(0), data_(nullptr) { |
| 133 | *this = std::move(other); |
| 134 | } |
| 135 | ~HashSet() { |
| 136 | DeallocateStorage(); |
| 137 | } |
| 138 | HashSet& operator=(HashSet&& other) { |
| 139 | std::swap(data_, other.data_); |
| 140 | std::swap(num_buckets_, other.num_buckets_); |
| 141 | std::swap(num_elements_, other.num_elements_); |
| 142 | std::swap(elements_until_expand_, other.elements_until_expand_); |
| 143 | std::swap(min_load_factor_, other.min_load_factor_); |
| 144 | std::swap(max_load_factor_, other.max_load_factor_); |
| 145 | return *this; |
| 146 | } |
| 147 | HashSet& operator=(const HashSet& other) { |
| 148 | DeallocateStorage(); |
| 149 | AllocateStorage(other.NumBuckets()); |
| 150 | for (size_t i = 0; i < num_buckets_; ++i) { |
| 151 | ElementForIndex(i) = other.data_[i]; |
| 152 | } |
| 153 | num_elements_ = other.num_elements_; |
| 154 | elements_until_expand_ = other.elements_until_expand_; |
| 155 | min_load_factor_ = other.min_load_factor_; |
| 156 | max_load_factor_ = other.max_load_factor_; |
| 157 | return *this; |
| 158 | } |
| 159 | // Lower case for c++11 for each. |
| 160 | Iterator begin() { |
| 161 | Iterator ret(this, 0); |
| 162 | if (num_buckets_ != 0 && IsFreeSlot(ret.GetIndex())) { |
| 163 | ++ret; // Skip all the empty slots. |
| 164 | } |
| 165 | return ret; |
| 166 | } |
| 167 | // Lower case for c++11 for each. |
| 168 | Iterator end() { |
| 169 | return Iterator(this, NumBuckets()); |
| 170 | } |
| 171 | bool Empty() { |
| 172 | return begin() == end(); |
| 173 | } |
| 174 | // Erase algorithm: |
| 175 | // Make an empty slot where the iterator is pointing. |
| 176 | // Scan fowards until we hit another empty slot. |
| 177 | // If an element inbetween doesn't rehash to the range from the current empty slot to the |
| 178 | // iterator. It must be before the empty slot, in that case we can move it to the empty slot |
| 179 | // and set the empty slot to be the location we just moved from. |
| 180 | // Relies on maintaining the invariant that there's no empty slots from the 'ideal' index of an |
| 181 | // element to its actual location/index. |
| 182 | Iterator Erase(Iterator it) { |
| 183 | // empty_index is the index that will become empty. |
| 184 | size_t empty_index = it.GetIndex(); |
| 185 | DCHECK(!IsFreeSlot(empty_index)); |
| 186 | size_t next_index = empty_index; |
| 187 | bool filled = false; // True if we filled the empty index. |
| 188 | while (true) { |
| 189 | next_index = NextIndex(next_index); |
| 190 | T& next_element = ElementForIndex(next_index); |
| 191 | // If the next element is empty, we are done. Make sure to clear the current empty index. |
| 192 | if (emptyfn_.IsEmpty(next_element)) { |
| 193 | emptyfn_.MakeEmpty(ElementForIndex(empty_index)); |
| 194 | break; |
| 195 | } |
| 196 | // Otherwise try to see if the next element can fill the current empty index. |
| 197 | const size_t next_hash = hashfn_(next_element); |
| 198 | // Calculate the ideal index, if it is within empty_index + 1 to next_index then there is |
| 199 | // nothing we can do. |
| 200 | size_t next_ideal_index = IndexForHash(next_hash); |
| 201 | // Loop around if needed for our check. |
| 202 | size_t unwrapped_next_index = next_index; |
| 203 | if (unwrapped_next_index < empty_index) { |
| 204 | unwrapped_next_index += NumBuckets(); |
| 205 | } |
| 206 | // Loop around if needed for our check. |
| 207 | size_t unwrapped_next_ideal_index = next_ideal_index; |
| 208 | if (unwrapped_next_ideal_index < empty_index) { |
| 209 | unwrapped_next_ideal_index += NumBuckets(); |
| 210 | } |
| 211 | if (unwrapped_next_ideal_index <= empty_index || |
| 212 | unwrapped_next_ideal_index > unwrapped_next_index) { |
| 213 | // If the target index isn't within our current range it must have been probed from before |
| 214 | // the empty index. |
| 215 | ElementForIndex(empty_index) = std::move(next_element); |
| 216 | filled = true; // TODO: Optimize |
| 217 | empty_index = next_index; |
| 218 | } |
| 219 | } |
| 220 | --num_elements_; |
| 221 | // If we didn't fill the slot then we need go to the next non free slot. |
| 222 | if (!filled) { |
| 223 | ++it; |
| 224 | } |
| 225 | return it; |
| 226 | } |
| 227 | // Find an element, returns end() if not found. |
| 228 | // Allows custom K types, example of when this is useful. |
| 229 | // Set of Class* sorted by name, want to find a class with a name but can't allocate a dummy |
| 230 | // object in the heap for performance solution. |
| 231 | template <typename K> |
| 232 | Iterator Find(const K& element) { |
| 233 | return FindWithHash(element, hashfn_(element)); |
| 234 | } |
| 235 | template <typename K> |
| 236 | Iterator FindWithHash(const K& element, size_t hash) { |
| 237 | DCHECK_EQ(hashfn_(element), hash); |
| 238 | size_t index = IndexForHash(hash); |
| 239 | while (true) { |
| 240 | T& slot = ElementForIndex(index); |
| 241 | if (emptyfn_.IsEmpty(slot)) { |
| 242 | return end(); |
| 243 | } |
| 244 | if (pred_(slot, element)) { |
| 245 | return Iterator(this, index); |
| 246 | } |
| 247 | index = NextIndex(index); |
| 248 | } |
| 249 | } |
| 250 | // Insert an element, allows duplicates. |
| 251 | void Insert(const T& element) { |
| 252 | InsertWithHash(element, hashfn_(element)); |
| 253 | } |
| 254 | void InsertWithHash(const T& element, size_t hash) { |
| 255 | DCHECK_EQ(hash, hashfn_(element)); |
| 256 | if (num_elements_ >= elements_until_expand_) { |
| 257 | Expand(); |
| 258 | DCHECK_LT(num_elements_, elements_until_expand_); |
| 259 | } |
| 260 | const size_t index = FirstAvailableSlot(IndexForHash(hash)); |
| 261 | data_[index] = element; |
| 262 | ++num_elements_; |
| 263 | } |
| 264 | size_t Size() const { |
| 265 | return num_elements_; |
| 266 | } |
| 267 | void ShrinkToMaximumLoad() { |
| 268 | Resize(Size() / max_load_factor_); |
| 269 | } |
| 270 | // To distance that inserted elements were probed. Used for measuring how good hash functions |
| 271 | // are. |
| 272 | size_t TotalProbeDistance() const { |
| 273 | size_t total = 0; |
| 274 | for (size_t i = 0; i < NumBuckets(); ++i) { |
| 275 | const T& element = ElementForIndex(i); |
| 276 | if (!emptyfn_.IsEmpty(element)) { |
| 277 | size_t ideal_location = IndexForHash(hashfn_(element)); |
| 278 | if (ideal_location > i) { |
| 279 | total += i + NumBuckets() - ideal_location; |
| 280 | } else { |
| 281 | total += i - ideal_location; |
| 282 | } |
| 283 | } |
| 284 | } |
| 285 | return total; |
| 286 | } |
| 287 | // Calculate the current load factor and return it. |
| 288 | double CalculateLoadFactor() const { |
| 289 | return static_cast<double>(Size()) / static_cast<double>(NumBuckets()); |
| 290 | } |
| 291 | // Make sure that everything reinserts in the right spot. Returns the number of errors. |
| 292 | size_t Verify() { |
| 293 | size_t errors = 0; |
| 294 | for (size_t i = 0; i < num_buckets_; ++i) { |
| 295 | T& element = data_[i]; |
| 296 | if (!emptyfn_.IsEmpty(element)) { |
| 297 | T temp; |
| 298 | emptyfn_.MakeEmpty(temp); |
| 299 | std::swap(temp, element); |
| 300 | size_t first_slot = FirstAvailableSlot(IndexForHash(hashfn_(temp))); |
| 301 | if (i != first_slot) { |
| 302 | LOG(ERROR) << "Element " << i << " should be in slot " << first_slot; |
| 303 | ++errors; |
| 304 | } |
| 305 | std::swap(temp, element); |
| 306 | } |
| 307 | } |
| 308 | return errors; |
| 309 | } |
| 310 | |
| 311 | private: |
| 312 | T& ElementForIndex(size_t index) { |
| 313 | DCHECK_LT(index, NumBuckets()); |
| 314 | DCHECK(data_ != nullptr); |
| 315 | return data_[index]; |
| 316 | } |
| 317 | const T& ElementForIndex(size_t index) const { |
| 318 | DCHECK_LT(index, NumBuckets()); |
| 319 | DCHECK(data_ != nullptr); |
| 320 | return data_[index]; |
| 321 | } |
| 322 | size_t IndexForHash(size_t hash) const { |
| 323 | return hash % num_buckets_; |
| 324 | } |
| 325 | size_t NextIndex(size_t index) const { |
| 326 | if (UNLIKELY(++index >= num_buckets_)) { |
| 327 | DCHECK_EQ(index, NumBuckets()); |
| 328 | return 0; |
| 329 | } |
| 330 | return index; |
| 331 | } |
| 332 | bool IsFreeSlot(size_t index) const { |
| 333 | return emptyfn_.IsEmpty(ElementForIndex(index)); |
| 334 | } |
| 335 | size_t NumBuckets() const { |
| 336 | return num_buckets_; |
| 337 | } |
| 338 | // Allocate a number of buckets. |
| 339 | void AllocateStorage(size_t num_buckets) { |
| 340 | num_buckets_ = num_buckets; |
| 341 | data_ = allocfn_.allocate(num_buckets_); |
| 342 | for (size_t i = 0; i < num_buckets_; ++i) { |
| 343 | allocfn_.construct(allocfn_.address(data_[i])); |
| 344 | emptyfn_.MakeEmpty(data_[i]); |
| 345 | } |
| 346 | } |
| 347 | void DeallocateStorage() { |
| 348 | if (num_buckets_ != 0) { |
| 349 | for (size_t i = 0; i < NumBuckets(); ++i) { |
| 350 | allocfn_.destroy(allocfn_.address(data_[i])); |
| 351 | } |
| 352 | allocfn_.deallocate(data_, NumBuckets()); |
| 353 | data_ = nullptr; |
| 354 | num_buckets_ = 0; |
| 355 | } |
| 356 | } |
| 357 | // Expand the set based on the load factors. |
| 358 | void Expand() { |
| 359 | size_t min_index = static_cast<size_t>(Size() / min_load_factor_); |
| 360 | if (min_index < kMinBuckets) { |
| 361 | min_index = kMinBuckets; |
| 362 | } |
| 363 | // Resize based on the minimum load factor. |
| 364 | Resize(min_index); |
| 365 | // When we hit elements_until_expand_, we are at the max load factor and must expand again. |
| 366 | elements_until_expand_ = NumBuckets() * max_load_factor_; |
| 367 | } |
| 368 | // Expand / shrink the table to the new specified size. |
| 369 | void Resize(size_t new_size) { |
| 370 | DCHECK_GE(new_size, Size()); |
| 371 | T* old_data = data_; |
| 372 | size_t old_num_buckets = num_buckets_; |
| 373 | // Reinsert all of the old elements. |
| 374 | AllocateStorage(new_size); |
| 375 | for (size_t i = 0; i < old_num_buckets; ++i) { |
| 376 | T& element = old_data[i]; |
| 377 | if (!emptyfn_.IsEmpty(element)) { |
| 378 | data_[FirstAvailableSlot(IndexForHash(hashfn_(element)))] = std::move(element); |
| 379 | } |
| 380 | allocfn_.destroy(allocfn_.address(element)); |
| 381 | } |
| 382 | allocfn_.deallocate(old_data, old_num_buckets); |
| 383 | } |
| 384 | ALWAYS_INLINE size_t FirstAvailableSlot(size_t index) const { |
| 385 | while (!emptyfn_.IsEmpty(data_[index])) { |
| 386 | index = NextIndex(index); |
| 387 | } |
| 388 | return index; |
| 389 | } |
| 390 | |
| 391 | Alloc allocfn_; // Allocator function. |
| 392 | HashFn hashfn_; // Hashing function. |
| 393 | EmptyFn emptyfn_; // IsEmpty/SetEmpty function. |
| 394 | Pred pred_; // Equals function. |
| 395 | size_t num_elements_; // Number of inserted elements. |
| 396 | size_t num_buckets_; // Number of hash table buckets. |
| 397 | size_t elements_until_expand_; // Maxmimum number of elements until we expand the table. |
| 398 | T* data_; // Backing storage. |
| 399 | double min_load_factor_; |
| 400 | double max_load_factor_; |
| 401 | |
| 402 | friend class Iterator; |
| 403 | }; |
| 404 | |
| 405 | } // namespace art |
| 406 | |
| 407 | #endif // ART_RUNTIME_BASE_HASH_SET_H_ |