Ian Hodson | 2ee91b4 | 2012-05-14 12:29:36 +0100 | [diff] [blame] | 1 | // Copyright 2006 The RE2 Authors. All Rights Reserved. |
| 2 | // Use of this source code is governed by a BSD-style |
| 3 | // license that can be found in the LICENSE file. |
| 4 | |
| 5 | // DESCRIPTION |
| 6 | // |
| 7 | // SparseArray<T>(m) is a map from integers in [0, m) to T values. |
| 8 | // It requires (sizeof(T)+sizeof(int))*m memory, but it provides |
| 9 | // fast iteration through the elements in the array and fast clearing |
| 10 | // of the array. The array has a concept of certain elements being |
| 11 | // uninitialized (having no value). |
| 12 | // |
| 13 | // Insertion and deletion are constant time operations. |
| 14 | // |
| 15 | // Allocating the array is a constant time operation |
| 16 | // when memory allocation is a constant time operation. |
| 17 | // |
| 18 | // Clearing the array is a constant time operation (unusual!). |
| 19 | // |
| 20 | // Iterating through the array is an O(n) operation, where n |
| 21 | // is the number of items in the array (not O(m)). |
| 22 | // |
| 23 | // The array iterator visits entries in the order they were first |
| 24 | // inserted into the array. It is safe to add items to the array while |
| 25 | // using an iterator: the iterator will visit indices added to the array |
| 26 | // during the iteration, but will not re-visit indices whose values |
| 27 | // change after visiting. Thus SparseArray can be a convenient |
| 28 | // implementation of a work queue. |
| 29 | // |
| 30 | // The SparseArray implementation is NOT thread-safe. It is up to the |
| 31 | // caller to make sure only one thread is accessing the array. (Typically |
| 32 | // these arrays are temporary values and used in situations where speed is |
| 33 | // important.) |
| 34 | // |
| 35 | // The SparseArray interface does not present all the usual STL bells and |
| 36 | // whistles. |
| 37 | // |
| 38 | // Implemented with reference to Briggs & Torczon, An Efficient |
| 39 | // Representation for Sparse Sets, ACM Letters on Programming Languages |
| 40 | // and Systems, Volume 2, Issue 1-4 (March-Dec. 1993), pp. 59-69. |
| 41 | // |
| 42 | // Briggs & Torczon popularized this technique, but it had been known |
| 43 | // long before their paper. They point out that Aho, Hopcroft, and |
| 44 | // Ullman's 1974 Design and Analysis of Computer Algorithms and Bentley's |
| 45 | // 1986 Programming Pearls both hint at the technique in exercises to the |
| 46 | // reader (in Aho & Hopcroft, exercise 2.12; in Bentley, column 1 |
| 47 | // exercise 8). |
| 48 | // |
| 49 | // Briggs & Torczon describe a sparse set implementation. I have |
| 50 | // trivially generalized it to create a sparse array (actually the original |
| 51 | // target of the AHU and Bentley exercises). |
| 52 | |
| 53 | // IMPLEMENTATION |
| 54 | // |
| 55 | // SparseArray uses a vector dense_ and an array sparse_to_dense_, both of |
| 56 | // size max_size_. At any point, the number of elements in the sparse array is |
| 57 | // size_. |
| 58 | // |
| 59 | // The vector dense_ contains the size_ elements in the sparse array (with |
| 60 | // their indices), |
| 61 | // in the order that the elements were first inserted. This array is dense: |
| 62 | // the size_ pairs are dense_[0] through dense_[size_-1]. |
| 63 | // |
| 64 | // The array sparse_to_dense_ maps from indices in [0,m) to indices in |
| 65 | // [0,size_). |
| 66 | // For indices present in the array, dense_[sparse_to_dense_[i]].index_ == i. |
| 67 | // For indices not present in the array, sparse_to_dense_ can contain |
| 68 | // any value at all, perhaps outside the range [0, size_) but perhaps not. |
| 69 | // |
| 70 | // The lax requirement on sparse_to_dense_ values makes clearing |
| 71 | // the array very easy: set size_ to 0. Lookups are slightly more |
| 72 | // complicated. An index i has a value in the array if and only if: |
| 73 | // sparse_to_dense_[i] is in [0, size_) AND |
| 74 | // dense_[sparse_to_dense_[i]].index_ == i. |
| 75 | // If both these properties hold, only then it is safe to refer to |
| 76 | // dense_[sparse_to_dense_[i]].value_ |
| 77 | // as the value associated with index i. |
| 78 | // |
| 79 | // To insert a new entry, set sparse_to_dense_[i] to size_, |
| 80 | // initialize dense_[size_], and then increment size_. |
| 81 | // |
| 82 | // Deletion of specific values from the array is implemented by |
| 83 | // swapping dense_[size_-1] and the dense_ being deleted and then |
| 84 | // updating the appropriate sparse_to_dense_ entries. |
| 85 | // |
| 86 | // To make the sparse array as efficient as possible for non-primitive types, |
| 87 | // elements may or may not be destroyed when they are deleted from the sparse |
| 88 | // array through a call to erase(), erase_existing() or resize(). They |
| 89 | // immediately become inaccessible, but they are only guaranteed to be |
| 90 | // destroyed when the SparseArray destructor is called. |
| 91 | |
| 92 | #ifndef RE2_UTIL_SPARSE_ARRAY_H__ |
| 93 | #define RE2_UTIL_SPARSE_ARRAY_H__ |
| 94 | |
| 95 | #include "util/util.h" |
| 96 | |
| 97 | namespace re2 { |
| 98 | |
| 99 | template<typename Value> |
| 100 | class SparseArray { |
| 101 | public: |
| 102 | SparseArray(); |
| 103 | SparseArray(int max_size); |
| 104 | ~SparseArray(); |
| 105 | |
| 106 | // IndexValue pairs: exposed in SparseArray::iterator. |
| 107 | class IndexValue; |
| 108 | |
| 109 | typedef IndexValue value_type; |
| 110 | typedef typename vector<IndexValue>::iterator iterator; |
| 111 | typedef typename vector<IndexValue>::const_iterator const_iterator; |
| 112 | |
| 113 | inline const IndexValue& iv(int i) const; |
| 114 | |
| 115 | // Return the number of entries in the array. |
| 116 | int size() const { |
| 117 | return size_; |
| 118 | } |
| 119 | |
| 120 | // Iterate over the array. |
| 121 | iterator begin() { |
| 122 | return dense_.begin(); |
| 123 | } |
| 124 | iterator end() { |
| 125 | return dense_.begin() + size_; |
| 126 | } |
| 127 | |
| 128 | const_iterator begin() const { |
| 129 | return dense_.begin(); |
| 130 | } |
| 131 | const_iterator end() const { |
| 132 | return dense_.begin() + size_; |
| 133 | } |
| 134 | |
| 135 | // Change the maximum size of the array. |
| 136 | // Invalidates all iterators. |
| 137 | void resize(int max_size); |
| 138 | |
| 139 | // Return the maximum size of the array. |
| 140 | // Indices can be in the range [0, max_size). |
| 141 | int max_size() const { |
| 142 | return max_size_; |
| 143 | } |
| 144 | |
| 145 | // Clear the array. |
| 146 | void clear() { |
| 147 | size_ = 0; |
| 148 | } |
| 149 | |
| 150 | // Check whether index i is in the array. |
| 151 | inline bool has_index(int i) const; |
| 152 | |
| 153 | // Comparison function for sorting. |
| 154 | // Can sort the sparse array so that future iterations |
| 155 | // will visit indices in increasing order using |
| 156 | // sort(arr.begin(), arr.end(), arr.less); |
| 157 | static bool less(const IndexValue& a, const IndexValue& b); |
| 158 | |
| 159 | public: |
| 160 | // Set the value at index i to v. |
| 161 | inline iterator set(int i, Value v); |
| 162 | |
| 163 | pair<iterator, bool> insert(const value_type& new_value); |
| 164 | |
| 165 | // Returns the value at index i |
| 166 | // or defaultv if index i is not initialized in the array. |
| 167 | inline Value get(int i, Value defaultv) const; |
| 168 | |
| 169 | iterator find(int i); |
| 170 | |
| 171 | const_iterator find(int i) const; |
| 172 | |
| 173 | // Change the value at index i to v. |
| 174 | // Fast but unsafe: only use if has_index(i) is true. |
| 175 | inline iterator set_existing(int i, Value v); |
| 176 | |
| 177 | // Set the value at the new index i to v. |
| 178 | // Fast but unsafe: only use if has_index(i) is false. |
| 179 | inline iterator set_new(int i, Value v); |
| 180 | |
| 181 | // Get the value at index i from the array.. |
| 182 | // Fast but unsafe: only use if has_index(i) is true. |
| 183 | inline Value get_existing(int i) const; |
| 184 | |
| 185 | // Erasing items from the array during iteration is in general |
| 186 | // NOT safe. There is one special case, which is that the current |
| 187 | // index-value pair can be erased as long as the iterator is then |
| 188 | // checked for being at the end before being incremented. |
| 189 | // For example: |
| 190 | // |
| 191 | // for (i = m.begin(); i != m.end(); ++i) { |
| 192 | // if (ShouldErase(i->index(), i->value())) { |
| 193 | // m.erase(i->index()); |
| 194 | // --i; |
| 195 | // } |
| 196 | // } |
| 197 | // |
| 198 | // Except in the specific case just described, elements must |
| 199 | // not be erased from the array (including clearing the array) |
| 200 | // while iterators are walking over the array. Otherwise, |
| 201 | // the iterators could walk past the end of the array. |
| 202 | |
| 203 | // Erases the element at index i from the array. |
| 204 | inline void erase(int i); |
| 205 | |
| 206 | // Erases the element at index i from the array. |
| 207 | // Fast but unsafe: only use if has_index(i) is true. |
| 208 | inline void erase_existing(int i); |
| 209 | |
| 210 | private: |
| 211 | // Add the index i to the array. |
| 212 | // Only use if has_index(i) is known to be false. |
| 213 | // Since it doesn't set the value associated with i, |
| 214 | // this function is private, only intended as a helper |
| 215 | // for other methods. |
| 216 | inline void create_index(int i); |
| 217 | |
| 218 | // In debug mode, verify that some invariant properties of the class |
| 219 | // are being maintained. This is called at the end of the constructor |
| 220 | // and at the beginning and end of all public non-const member functions. |
| 221 | inline void DebugCheckInvariants() const; |
| 222 | |
| 223 | int size_; |
| 224 | int max_size_; |
| 225 | int* sparse_to_dense_; |
| 226 | vector<IndexValue> dense_; |
Alexander Gutkin | 0d4c523 | 2013-02-28 13:47:27 +0000 | [diff] [blame] | 227 | bool valgrind_; |
Ian Hodson | 2ee91b4 | 2012-05-14 12:29:36 +0100 | [diff] [blame] | 228 | |
| 229 | DISALLOW_EVIL_CONSTRUCTORS(SparseArray); |
| 230 | }; |
| 231 | |
| 232 | template<typename Value> |
| 233 | SparseArray<Value>::SparseArray() |
Alexander Gutkin | 0d4c523 | 2013-02-28 13:47:27 +0000 | [diff] [blame] | 234 | : size_(0), max_size_(0), sparse_to_dense_(NULL), dense_(), valgrind_(RunningOnValgrind()) {} |
Ian Hodson | 2ee91b4 | 2012-05-14 12:29:36 +0100 | [diff] [blame] | 235 | |
| 236 | // IndexValue pairs: exposed in SparseArray::iterator. |
| 237 | template<typename Value> |
| 238 | class SparseArray<Value>::IndexValue { |
| 239 | friend class SparseArray; |
| 240 | public: |
| 241 | typedef int first_type; |
| 242 | typedef Value second_type; |
| 243 | |
| 244 | IndexValue() {} |
| 245 | IndexValue(int index, const Value& value) : second(value), index_(index) {} |
| 246 | |
| 247 | int index() const { return index_; } |
| 248 | Value value() const { return second; } |
| 249 | |
| 250 | // Provide the data in the 'second' member so that the utilities |
| 251 | // in map-util work. |
| 252 | Value second; |
| 253 | |
| 254 | private: |
| 255 | int index_; |
| 256 | }; |
| 257 | |
| 258 | template<typename Value> |
| 259 | const typename SparseArray<Value>::IndexValue& |
| 260 | SparseArray<Value>::iv(int i) const { |
| 261 | DCHECK_GE(i, 0); |
| 262 | DCHECK_LT(i, size_); |
| 263 | return dense_[i]; |
| 264 | } |
| 265 | |
| 266 | // Change the maximum size of the array. |
| 267 | // Invalidates all iterators. |
| 268 | template<typename Value> |
| 269 | void SparseArray<Value>::resize(int new_max_size) { |
| 270 | DebugCheckInvariants(); |
| 271 | if (new_max_size > max_size_) { |
| 272 | int* a = new int[new_max_size]; |
| 273 | if (sparse_to_dense_) { |
| 274 | memmove(a, sparse_to_dense_, max_size_*sizeof a[0]); |
| 275 | // Don't need to zero the memory but appease Valgrind. |
Alexander Gutkin | 0d4c523 | 2013-02-28 13:47:27 +0000 | [diff] [blame] | 276 | if (valgrind_) { |
Ian Hodson | 2ee91b4 | 2012-05-14 12:29:36 +0100 | [diff] [blame] | 277 | for (int i = max_size_; i < new_max_size; i++) |
| 278 | a[i] = 0xababababU; |
| 279 | } |
| 280 | delete[] sparse_to_dense_; |
| 281 | } |
| 282 | sparse_to_dense_ = a; |
| 283 | |
| 284 | dense_.resize(new_max_size); |
| 285 | } |
| 286 | max_size_ = new_max_size; |
| 287 | if (size_ > max_size_) |
| 288 | size_ = max_size_; |
| 289 | DebugCheckInvariants(); |
| 290 | } |
| 291 | |
| 292 | // Check whether index i is in the array. |
| 293 | template<typename Value> |
| 294 | bool SparseArray<Value>::has_index(int i) const { |
| 295 | DCHECK_GE(i, 0); |
| 296 | DCHECK_LT(i, max_size_); |
| 297 | if (static_cast<uint>(i) >= max_size_) { |
| 298 | return false; |
| 299 | } |
| 300 | // Unsigned comparison avoids checking sparse_to_dense_[i] < 0. |
| 301 | return (uint)sparse_to_dense_[i] < (uint)size_ && |
| 302 | dense_[sparse_to_dense_[i]].index_ == i; |
| 303 | } |
| 304 | |
| 305 | // Set the value at index i to v. |
| 306 | template<typename Value> |
| 307 | typename SparseArray<Value>::iterator SparseArray<Value>::set(int i, Value v) { |
| 308 | DebugCheckInvariants(); |
| 309 | if (static_cast<uint>(i) >= max_size_) { |
| 310 | // Semantically, end() would be better here, but we already know |
| 311 | // the user did something stupid, so begin() insulates them from |
| 312 | // dereferencing an invalid pointer. |
| 313 | return begin(); |
| 314 | } |
| 315 | if (!has_index(i)) |
| 316 | create_index(i); |
| 317 | return set_existing(i, v); |
| 318 | } |
| 319 | |
| 320 | template<typename Value> |
| 321 | pair<typename SparseArray<Value>::iterator, bool> SparseArray<Value>::insert( |
| 322 | const value_type& new_value) { |
| 323 | DebugCheckInvariants(); |
| 324 | pair<typename SparseArray<Value>::iterator, bool> p; |
| 325 | if (has_index(new_value.index_)) { |
| 326 | p = make_pair(dense_.begin() + sparse_to_dense_[new_value.index_], false); |
| 327 | } else { |
| 328 | p = make_pair(set_new(new_value.index_, new_value.second), true); |
| 329 | } |
| 330 | DebugCheckInvariants(); |
| 331 | return p; |
| 332 | } |
| 333 | |
| 334 | template<typename Value> |
| 335 | Value SparseArray<Value>::get(int i, Value defaultv) const { |
| 336 | if (!has_index(i)) |
| 337 | return defaultv; |
| 338 | return get_existing(i); |
| 339 | } |
| 340 | |
| 341 | template<typename Value> |
| 342 | typename SparseArray<Value>::iterator SparseArray<Value>::find(int i) { |
| 343 | if (has_index(i)) |
| 344 | return dense_.begin() + sparse_to_dense_[i]; |
| 345 | return end(); |
| 346 | } |
| 347 | |
| 348 | template<typename Value> |
| 349 | typename SparseArray<Value>::const_iterator |
| 350 | SparseArray<Value>::find(int i) const { |
| 351 | if (has_index(i)) { |
| 352 | return dense_.begin() + sparse_to_dense_[i]; |
| 353 | } |
| 354 | return end(); |
| 355 | } |
| 356 | |
| 357 | template<typename Value> |
| 358 | typename SparseArray<Value>::iterator |
| 359 | SparseArray<Value>::set_existing(int i, Value v) { |
| 360 | DebugCheckInvariants(); |
| 361 | DCHECK(has_index(i)); |
| 362 | dense_[sparse_to_dense_[i]].second = v; |
| 363 | DebugCheckInvariants(); |
| 364 | return dense_.begin() + sparse_to_dense_[i]; |
| 365 | } |
| 366 | |
| 367 | template<typename Value> |
| 368 | typename SparseArray<Value>::iterator |
| 369 | SparseArray<Value>::set_new(int i, Value v) { |
| 370 | DebugCheckInvariants(); |
| 371 | if (static_cast<uint>(i) >= max_size_) { |
| 372 | // Semantically, end() would be better here, but we already know |
| 373 | // the user did something stupid, so begin() insulates them from |
| 374 | // dereferencing an invalid pointer. |
| 375 | return begin(); |
| 376 | } |
| 377 | DCHECK(!has_index(i)); |
| 378 | create_index(i); |
| 379 | return set_existing(i, v); |
| 380 | } |
| 381 | |
| 382 | template<typename Value> |
| 383 | Value SparseArray<Value>::get_existing(int i) const { |
| 384 | DCHECK(has_index(i)); |
| 385 | return dense_[sparse_to_dense_[i]].second; |
| 386 | } |
| 387 | |
| 388 | template<typename Value> |
| 389 | void SparseArray<Value>::erase(int i) { |
| 390 | DebugCheckInvariants(); |
| 391 | if (has_index(i)) |
| 392 | erase_existing(i); |
| 393 | DebugCheckInvariants(); |
| 394 | } |
| 395 | |
| 396 | template<typename Value> |
| 397 | void SparseArray<Value>::erase_existing(int i) { |
| 398 | DebugCheckInvariants(); |
| 399 | DCHECK(has_index(i)); |
| 400 | int di = sparse_to_dense_[i]; |
| 401 | if (di < size_ - 1) { |
| 402 | dense_[di] = dense_[size_ - 1]; |
| 403 | sparse_to_dense_[dense_[di].index_] = di; |
| 404 | } |
| 405 | size_--; |
| 406 | DebugCheckInvariants(); |
| 407 | } |
| 408 | |
| 409 | template<typename Value> |
| 410 | void SparseArray<Value>::create_index(int i) { |
| 411 | DCHECK(!has_index(i)); |
| 412 | DCHECK_LT(size_, max_size_); |
| 413 | sparse_to_dense_[i] = size_; |
| 414 | dense_[size_].index_ = i; |
| 415 | size_++; |
| 416 | } |
| 417 | |
| 418 | template<typename Value> SparseArray<Value>::SparseArray(int max_size) { |
| 419 | max_size_ = max_size; |
| 420 | sparse_to_dense_ = new int[max_size]; |
Alexander Gutkin | 0d4c523 | 2013-02-28 13:47:27 +0000 | [diff] [blame] | 421 | valgrind_ = RunningOnValgrind(); |
Ian Hodson | 2ee91b4 | 2012-05-14 12:29:36 +0100 | [diff] [blame] | 422 | dense_.resize(max_size); |
| 423 | // Don't need to zero the new memory, but appease Valgrind. |
Alexander Gutkin | 0d4c523 | 2013-02-28 13:47:27 +0000 | [diff] [blame] | 424 | if (valgrind_) { |
Ian Hodson | 2ee91b4 | 2012-05-14 12:29:36 +0100 | [diff] [blame] | 425 | for (int i = 0; i < max_size; i++) { |
| 426 | sparse_to_dense_[i] = 0xababababU; |
| 427 | dense_[i].index_ = 0xababababU; |
| 428 | } |
| 429 | } |
| 430 | size_ = 0; |
| 431 | DebugCheckInvariants(); |
| 432 | } |
| 433 | |
| 434 | template<typename Value> SparseArray<Value>::~SparseArray() { |
| 435 | DebugCheckInvariants(); |
| 436 | delete[] sparse_to_dense_; |
| 437 | } |
| 438 | |
| 439 | template<typename Value> void SparseArray<Value>::DebugCheckInvariants() const { |
| 440 | DCHECK_LE(0, size_); |
| 441 | DCHECK_LE(size_, max_size_); |
| 442 | DCHECK(size_ == 0 || sparse_to_dense_ != NULL); |
| 443 | } |
| 444 | |
| 445 | // Comparison function for sorting. |
| 446 | template<typename Value> bool SparseArray<Value>::less(const IndexValue& a, |
| 447 | const IndexValue& b) { |
| 448 | return a.index_ < b.index_; |
| 449 | } |
| 450 | |
| 451 | } // namespace re2 |
| 452 | |
| 453 | #endif // RE2_UTIL_SPARSE_ARRAY_H__ |