| 1 | // Copyright (c) 2005, Google Inc. |
|---|---|
| 2 | // All rights reserved. |
| 3 | // |
| 4 | // Redistribution and use in source and binary forms, with or without |
| 5 | // modification, are permitted provided that the following conditions are |
| 6 | // met: |
| 7 | // |
| 8 | // * Redistributions of source code must retain the above copyright |
| 9 | // notice, this list of conditions and the following disclaimer. |
| 10 | // * Redistributions in binary form must reproduce the above |
| 11 | // copyright notice, this list of conditions and the following disclaimer |
| 12 | // in the documentation and/or other materials provided with the |
| 13 | // distribution. |
| 14 | // * Neither the name of Google Inc. nor the names of its |
| 15 | // contributors may be used to endorse or promote products derived from |
| 16 | // this software without specific prior written permission. |
| 17 | // |
| 18 | // THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS |
| 19 | // "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT |
| 20 | // LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR |
| 21 | // A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT |
| 22 | // OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, |
| 23 | // SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT |
| 24 | // LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, |
| 25 | // DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY |
| 26 | // THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT |
| 27 | // (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE |
| 28 | // OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. |
| 29 | |
| 30 | // --- |
| 31 | // |
| 32 | // |
| 33 | // A sparsetable is a random container that implements a sparse array, |
| 34 | // that is, an array that uses very little memory to store unassigned |
| 35 | // indices (in this case, between 1-2 bits per unassigned index). For |
| 36 | // instance, if you allocate an array of size 5 and assign a[2] = <big |
| 37 | // struct>, then a[2] will take up a lot of memory but a[0], a[1], |
| 38 | // a[3], and a[4] will not. Array elements that have a value are |
| 39 | // called "assigned". Array elements that have no value yet, or have |
| 40 | // had their value cleared using erase() or clear(), are called |
| 41 | // "unassigned". |
| 42 | // |
| 43 | // Unassigned values seem to have the default value of T (see below). |
| 44 | // Nevertheless, there is a difference between an unassigned index and |
| 45 | // one explicitly assigned the value of T(). The latter is considered |
| 46 | // assigned. |
| 47 | // |
| 48 | // Access to an array element is constant time, as is insertion and |
| 49 | // deletion. Insertion and deletion may be fairly slow, however: |
| 50 | // because of this container's memory economy, each insert and delete |
| 51 | // causes a memory reallocation. |
| 52 | // |
| 53 | // NOTE: You should not test(), get(), or set() any index that is |
| 54 | // greater than sparsetable.size(). If you need to do that, call |
| 55 | // resize() first. |
| 56 | // |
| 57 | // --- Template parameters |
| 58 | // PARAMETER DESCRIPTION DEFAULT |
| 59 | // T The value of the array: the type of -- |
| 60 | // object that is stored in the array. |
| 61 | // |
| 62 | // GROUP_SIZE How large each "group" in the table 48 |
| 63 | // is (see below). Larger values use |
| 64 | // a little less memory but cause most |
| 65 | // operations to be a little slower |
| 66 | // |
| 67 | // Alloc: Allocator to use to allocate memory. libc_allocator_with_realloc |
| 68 | // |
| 69 | // --- Model of |
| 70 | // Random Access Container |
| 71 | // |
| 72 | // --- Type requirements |
| 73 | // T must be Copy Constructible. It need not be Assignable. |
| 74 | // |
| 75 | // --- Public base classes |
| 76 | // None. |
| 77 | // |
| 78 | // --- Members |
| 79 | // Type members |
| 80 | // |
| 81 | // MEMBER WHERE DEFINED DESCRIPTION |
| 82 | // value_type container The type of object, T, stored in the array |
| 83 | // allocator_type container Allocator to use |
| 84 | // pointer container Pointer to p |
| 85 | // const_pointer container Const pointer to p |
| 86 | // reference container Reference to t |
| 87 | // const_reference container Const reference to t |
| 88 | // size_type container An unsigned integral type |
| 89 | // difference_type container A signed integral type |
| 90 | // iterator [*] container Iterator used to iterate over a sparsetable |
| 91 | // const_iterator container Const iterator used to iterate over a table |
| 92 | // reverse_iterator reversible Iterator used to iterate backwards over |
| 93 | // container a sparsetable |
| 94 | // const_reverse_iterator reversible container Guess |
| 95 | // nonempty_iterator [+] sparsetable Iterates over assigned |
| 96 | // array elements only |
| 97 | // const_nonempty_iterator sparsetable Iterates over assigned |
| 98 | // array elements only |
| 99 | // reverse_nonempty_iterator sparsetable Iterates backwards over |
| 100 | // assigned array elements only |
| 101 | // const_reverse_nonempty_iterator sparsetable Iterates backwards over |
| 102 | // assigned array elements only |
| 103 | // |
| 104 | // [*] All iterators are const in a sparsetable (though nonempty_iterators |
| 105 | // may not be). Use get() and set() to assign values, not iterators. |
| 106 | // |
| 107 | // [+] iterators are random-access iterators. nonempty_iterators are |
| 108 | // bidirectional iterators. |
| 109 | |
| 110 | // Iterator members |
| 111 | // MEMBER WHERE DEFINED DESCRIPTION |
| 112 | // |
| 113 | // iterator begin() container An iterator to the beginning of the table |
| 114 | // iterator end() container An iterator to the end of the table |
| 115 | // const_iterator container A const_iterator pointing to the |
| 116 | // begin() const beginning of a sparsetable |
| 117 | // const_iterator container A const_iterator pointing to the |
| 118 | // end() const end of a sparsetable |
| 119 | // |
| 120 | // reverse_iterator reversable Points to beginning of a reversed |
| 121 | // rbegin() container sparsetable |
| 122 | // reverse_iterator reversable Points to end of a reversed table |
| 123 | // rend() container |
| 124 | // const_reverse_iterator reversable Points to beginning of a |
| 125 | // rbegin() const container reversed sparsetable |
| 126 | // const_reverse_iterator reversable Points to end of a reversed table |
| 127 | // rend() const container |
| 128 | // |
| 129 | // nonempty_iterator sparsetable Points to first assigned element |
| 130 | // begin() of a sparsetable |
| 131 | // nonempty_iterator sparsetable Points past last assigned element |
| 132 | // end() of a sparsetable |
| 133 | // const_nonempty_iterator sparsetable Points to first assigned element |
| 134 | // begin() const of a sparsetable |
| 135 | // const_nonempty_iterator sparsetable Points past last assigned element |
| 136 | // end() const of a sparsetable |
| 137 | // |
| 138 | // reverse_nonempty_iterator sparsetable Points to first assigned element |
| 139 | // begin() of a reversed sparsetable |
| 140 | // reverse_nonempty_iterator sparsetable Points past last assigned element |
| 141 | // end() of a reversed sparsetable |
| 142 | // const_reverse_nonempty_iterator sparsetable Points to first assigned |
| 143 | // begin() const elt of a reversed sparsetable |
| 144 | // const_reverse_nonempty_iterator sparsetable Points past last assigned |
| 145 | // end() const elt of a reversed sparsetable |
| 146 | // |
| 147 | // |
| 148 | // Other members |
| 149 | // MEMBER WHERE DEFINED DESCRIPTION |
| 150 | // sparsetable() sparsetable A table of size 0; must resize() |
| 151 | // before using. |
| 152 | // sparsetable(size_type size) sparsetable A table of size size. All |
| 153 | // indices are unassigned. |
| 154 | // sparsetable( |
| 155 | // const sparsetable &tbl) sparsetable Copy constructor |
| 156 | // ~sparsetable() sparsetable The destructor |
| 157 | // sparsetable &operator=( sparsetable The assignment operator |
| 158 | // const sparsetable &tbl) |
| 159 | // |
| 160 | // void resize(size_type size) sparsetable Grow or shrink a table to |
| 161 | // have size indices [*] |
| 162 | // |
| 163 | // void swap(sparsetable &x) sparsetable Swap two sparsetables |
| 164 | // void swap(sparsetable &x, sparsetable Swap two sparsetables |
| 165 | // sparsetable &y) (global, not member, function) |
| 166 | // |
| 167 | // size_type size() const sparsetable Number of "buckets" in the table |
| 168 | // size_type max_size() const sparsetable Max allowed size of a sparsetable |
| 169 | // bool empty() const sparsetable true if size() == 0 |
| 170 | // size_type num_nonempty() const sparsetable Number of assigned "buckets" |
| 171 | // |
| 172 | // const_reference get( sparsetable Value at index i, or default |
| 173 | // size_type i) const value if i is unassigned |
| 174 | // const_reference operator[]( sparsetable Identical to get(i) [+] |
| 175 | // difference_type i) const |
| 176 | // reference set(size_type i, sparsetable Set element at index i to |
| 177 | // const_reference val) be a copy of val |
| 178 | // bool test(size_type i) sparsetable True if element at index i |
| 179 | // const has been assigned to |
| 180 | // bool test(iterator pos) sparsetable True if element pointed to |
| 181 | // const by pos has been assigned to |
| 182 | // void erase(iterator pos) sparsetable Set element pointed to by |
| 183 | // pos to be unassigned [!] |
| 184 | // void erase(size_type i) sparsetable Set element i to be unassigned |
| 185 | // void erase(iterator start, sparsetable Erases all elements between |
| 186 | // iterator end) start and end |
| 187 | // void clear() sparsetable Erases all elements in the table |
| 188 | // |
| 189 | // I/O versions exist for both FILE* and for File* (Google2-style files): |
| 190 | // bool write_metadata(FILE *fp) sparsetable Writes a sparsetable to the |
| 191 | // bool write_metadata(File *fp) given file. true if write |
| 192 | // completes successfully |
| 193 | // bool read_metadata(FILE *fp) sparsetable Replaces sparsetable with |
| 194 | // bool read_metadata(File *fp) version read from fp. true |
| 195 | // if read completes sucessfully |
| 196 | // bool write_nopointer_data(FILE *fp) Read/write the data stored in |
| 197 | // bool read_nopointer_data(FILE*fp) the table, if it's simple |
| 198 | // |
| 199 | // bool operator==( forward Tests two tables for equality. |
| 200 | // const sparsetable &t1, container This is a global function, |
| 201 | // const sparsetable &t2) not a member function. |
| 202 | // bool operator<( forward Lexicographical comparison. |
| 203 | // const sparsetable &t1, container This is a global function, |
| 204 | // const sparsetable &t2) not a member function. |
| 205 | // |
| 206 | // [*] If you shrink a sparsetable using resize(), assigned elements |
| 207 | // past the end of the table are removed using erase(). If you grow |
| 208 | // a sparsetable, new unassigned indices are created. |
| 209 | // |
| 210 | // [+] Note that operator[] returns a const reference. You must use |
| 211 | // set() to change the value of a table element. |
| 212 | // |
| 213 | // [!] Unassignment also calls the destructor. |
| 214 | // |
| 215 | // Iterators are invalidated whenever an item is inserted or |
| 216 | // deleted (ie set() or erase() is used) or when the size of |
| 217 | // the table changes (ie resize() or clear() is used). |
| 218 | // |
| 219 | // See doc/sparsetable.html for more information about how to use this class. |
| 220 | |
| 221 | // Note: this uses STL style for naming, rather than Google naming. |
| 222 | // That's because this is an STL-y container |
| 223 | |
| 224 | #pragma once |
| 225 | |
| 226 | #include <cstdlib> // for malloc/free |
| 227 | #include <cstdio> // to read/write tables |
| 228 | #include <cstring> // for memcpy |
| 229 | #include <cstdint> // the normal place uint16_t is defined |
| 230 | #include <cassert> // for bounds checking |
| 231 | #include <iterator> // to define reverse_iterator for me |
| 232 | #include <algorithm> // equal, lexicographical_compare, swap,... |
| 233 | #include <memory> // uninitialized_copy, uninitialized_fill |
| 234 | #include <vector> // a sparsetable is a vector of groups |
| 235 | #include <type_traits> |
| 236 | #include <sparsehash/internal/hashtable-common.h> |
| 237 | #include <sparsehash/internal/libc_allocator_with_realloc.h> |
| 238 | #include <sparsehash/traits> |
| 239 | |
| 240 | namespace google { |
| 241 | // The smaller this is, the faster lookup is (because the group bitmap is |
| 242 | // smaller) and the faster insert is, because there's less to move. |
| 243 | // On the other hand, there are more groups. Since group::size_type is |
| 244 | // a short, this number should be of the form 32*x + 16 to avoid waste. |
| 245 | static const uint16_t DEFAULT_SPARSEGROUP_SIZE = 48; // fits in 1.5 words |
| 246 | |
| 247 | // Our iterator as simple as iterators can be: basically it's just |
| 248 | // the index into our table. Dereference, the only complicated |
| 249 | // thing, we punt to the table class. This just goes to show how |
| 250 | // much machinery STL requires to do even the most trivial tasks. |
| 251 | // |
| 252 | // A NOTE ON ASSIGNING: |
| 253 | // A sparse table does not actually allocate memory for entries |
| 254 | // that are not filled. Because of this, it becomes complicated |
| 255 | // to have a non-const iterator: we don't know, if the iterator points |
| 256 | // to a not-filled bucket, whether you plan to fill it with something |
| 257 | // or whether you plan to read its value (in which case you'll get |
| 258 | // the default bucket value). Therefore, while we can define const |
| 259 | // operations in a pretty 'normal' way, for non-const operations, we |
| 260 | // define something that returns a helper object with operator= and |
| 261 | // operator& that allocate a bucket lazily. We use this for table[] |
| 262 | // and also for regular table iterators. |
| 263 | |
| 264 | template <class tabletype> |
| 265 | class table_element_adaptor { |
| 266 | public: |
| 267 | typedef typename tabletype::value_type value_type; |
| 268 | typedef typename tabletype::size_type size_type; |
| 269 | typedef typename tabletype::reference reference; |
| 270 | typedef typename tabletype::pointer pointer; |
| 271 | |
| 272 | table_element_adaptor(tabletype* tbl, size_type p) : table(tbl), pos(p) {} |
| 273 | table_element_adaptor& operator=(const value_type& val) { |
| 274 | table->set(pos, val); |
| 275 | return *this; |
| 276 | } |
| 277 | operator value_type() { return table->get(pos); } // we look like a value |
| 278 | pointer operator&() { return &table->mutating_get(pos); } |
| 279 | |
| 280 | private: |
| 281 | tabletype* table; |
| 282 | size_type pos; |
| 283 | }; |
| 284 | |
| 285 | // Our iterator as simple as iterators can be: basically it's just |
| 286 | // the index into our table. Dereference, the only complicated |
| 287 | // thing, we punt to the table class. This just goes to show how |
| 288 | // much machinery STL requires to do even the most trivial tasks. |
| 289 | // |
| 290 | // By templatizing over tabletype, we have one iterator type which |
| 291 | // we can use for both sparsetables and sparsebins. In fact it |
| 292 | // works on any class that allows size() and operator[] (eg vector), |
| 293 | // as long as it does the standard STL typedefs too (eg value_type). |
| 294 | |
| 295 | template <class tabletype> |
| 296 | class table_iterator { |
| 297 | public: |
| 298 | typedef table_iterator iterator; |
| 299 | |
| 300 | typedef std::random_access_iterator_tag iterator_category; |
| 301 | typedef typename tabletype::value_type value_type; |
| 302 | typedef typename tabletype::difference_type difference_type; |
| 303 | typedef typename tabletype::size_type size_type; |
| 304 | typedef table_element_adaptor<tabletype> reference; |
| 305 | typedef table_element_adaptor<tabletype>* pointer; |
| 306 | typedef typename tabletype::const_reference const_reference; // we're const-only |
| 307 | |
| 308 | // The "real" constructor |
| 309 | table_iterator(tabletype* tbl, size_type p) : table(tbl), pos(p) {} |
| 310 | // The default constructor, used when I define vars of type table::iterator |
| 311 | table_iterator() : table(NULL), pos(0) {} |
| 312 | // The copy constructor, for when I say table::iterator foo = tbl.begin() |
| 313 | // The default destructor is fine; we don't define one |
| 314 | // The default operator= is fine; we don't define one |
| 315 | |
| 316 | // The main thing our iterator does is dereference. If the table entry |
| 317 | // we point to is empty, we return the default value type. |
| 318 | // This is the big different function from the const iterator. |
| 319 | reference operator*() { return table_element_adaptor<tabletype>(table, pos); } |
| 320 | const_reference operator*() const { return table_element_adaptor<tabletype>(table, pos); } |
| 321 | pointer operator->() { return &(operator*()); } |
| 322 | |
| 323 | // Helper function to assert things are ok; eg pos is still in range |
| 324 | void check() const { |
| 325 | assert(table); |
| 326 | assert(pos <= table->size()); |
| 327 | } |
| 328 | |
| 329 | // Arithmetic: we just do arithmetic on pos. We don't even need to |
| 330 | // do bounds checking, since STL doesn't consider that its job. :-) |
| 331 | iterator& operator+=(size_type t) { |
| 332 | pos += t; |
| 333 | check(); |
| 334 | return *this; |
| 335 | } |
| 336 | iterator& operator-=(size_type t) { |
| 337 | pos -= t; |
| 338 | check(); |
| 339 | return *this; |
| 340 | } |
| 341 | iterator& operator++() { |
| 342 | ++pos; |
| 343 | check(); |
| 344 | return *this; |
| 345 | } |
| 346 | iterator& operator--() { |
| 347 | --pos; |
| 348 | check(); |
| 349 | return *this; |
| 350 | } |
| 351 | iterator operator++(int) { |
| 352 | iterator tmp(*this); // for x++ |
| 353 | ++pos; |
| 354 | check(); |
| 355 | return tmp; |
| 356 | } |
| 357 | iterator operator--(int) { |
| 358 | iterator tmp(*this); // for x-- |
| 359 | --pos; |
| 360 | check(); |
| 361 | return tmp; |
| 362 | } |
| 363 | iterator operator+(difference_type i) const { |
| 364 | iterator tmp(*this); |
| 365 | tmp += i; |
| 366 | return tmp; |
| 367 | } |
| 368 | iterator operator-(difference_type i) const { |
| 369 | iterator tmp(*this); |
| 370 | tmp -= i; |
| 371 | return tmp; |
| 372 | } |
| 373 | difference_type operator-(iterator it) const { // for "x = it2 - it" |
| 374 | assert(table == it.table); |
| 375 | return pos - it.pos; |
| 376 | } |
| 377 | reference operator[](difference_type n) const { |
| 378 | return *(*this + n); // simple though not totally efficient |
| 379 | } |
| 380 | |
| 381 | // Comparisons. |
| 382 | bool operator==(const iterator& it) const { |
| 383 | return table == it.table && pos == it.pos; |
| 384 | } |
| 385 | bool operator<(const iterator& it) const { |
| 386 | assert(table == it.table); // life is bad bad bad otherwise |
| 387 | return pos < it.pos; |
| 388 | } |
| 389 | bool operator!=(const iterator& it) const { return !(*this == it); } |
| 390 | bool operator<=(const iterator& it) const { return !(it < *this); } |
| 391 | bool operator>(const iterator& it) const { return it < *this; } |
| 392 | bool operator>=(const iterator& it) const { return !(*this < it); } |
| 393 | |
| 394 | // Here's the info we actually need to be an iterator |
| 395 | tabletype* table; // so we can dereference and bounds-check |
| 396 | size_type pos; // index into the table |
| 397 | }; |
| 398 | |
| 399 | // support for "3 + iterator" has to be defined outside the class, alas |
| 400 | template <class T> |
| 401 | table_iterator<T> operator+(typename table_iterator<T>::difference_type i, |
| 402 | table_iterator<T> it) { |
| 403 | return it + i; // so people can say it2 = 3 + it |
| 404 | } |
| 405 | |
| 406 | template <class tabletype> |
| 407 | class const_table_iterator { |
| 408 | public: |
| 409 | typedef table_iterator<tabletype> iterator; |
| 410 | typedef const_table_iterator const_iterator; |
| 411 | |
| 412 | typedef std::random_access_iterator_tag iterator_category; |
| 413 | typedef typename tabletype::value_type value_type; |
| 414 | typedef typename tabletype::difference_type difference_type; |
| 415 | typedef typename tabletype::size_type size_type; |
| 416 | typedef typename tabletype::const_reference reference; // we're const-only |
| 417 | typedef typename tabletype::const_pointer pointer; |
| 418 | |
| 419 | // The "real" constructor |
| 420 | const_table_iterator(const tabletype* tbl, size_type p) |
| 421 | : table(tbl), pos(p) {} |
| 422 | // The default constructor, used when I define vars of type table::iterator |
| 423 | const_table_iterator() : table(NULL), pos(0) {} |
| 424 | // The copy constructor, for when I say table::iterator foo = tbl.begin() |
| 425 | // Also converts normal iterators to const iterators |
| 426 | const_table_iterator(const iterator& from) |
| 427 | : table(from.table), pos(from.pos) {} |
| 428 | // The default destructor is fine; we don't define one |
| 429 | // The default operator= is fine; we don't define one |
| 430 | |
| 431 | // The main thing our iterator does is dereference. If the table entry |
| 432 | // we point to is empty, we return the default value type. |
| 433 | reference operator*() const { return (*table)[pos]; } |
| 434 | pointer operator->() const { return &(operator*()); } |
| 435 | |
| 436 | // Helper function to assert things are ok; eg pos is still in range |
| 437 | void check() const { |
| 438 | assert(table); |
| 439 | assert(pos <= table->size()); |
| 440 | } |
| 441 | |
| 442 | // Arithmetic: we just do arithmetic on pos. We don't even need to |
| 443 | // do bounds checking, since STL doesn't consider that its job. :-) |
| 444 | const_iterator& operator+=(size_type t) { |
| 445 | pos += t; |
| 446 | check(); |
| 447 | return *this; |
| 448 | } |
| 449 | const_iterator& operator-=(size_type t) { |
| 450 | pos -= t; |
| 451 | check(); |
| 452 | return *this; |
| 453 | } |
| 454 | const_iterator& operator++() { |
| 455 | ++pos; |
| 456 | check(); |
| 457 | return *this; |
| 458 | } |
| 459 | const_iterator& operator--() { |
| 460 | --pos; |
| 461 | check(); |
| 462 | return *this; |
| 463 | } |
| 464 | const_iterator operator++(int) { |
| 465 | const_iterator tmp(*this); // for x++ |
| 466 | ++pos; |
| 467 | check(); |
| 468 | return tmp; |
| 469 | } |
| 470 | const_iterator operator--(int) { |
| 471 | const_iterator tmp(*this); // for x-- |
| 472 | --pos; |
| 473 | check(); |
| 474 | return tmp; |
| 475 | } |
| 476 | const_iterator operator+(difference_type i) const { |
| 477 | const_iterator tmp(*this); |
| 478 | tmp += i; |
| 479 | return tmp; |
| 480 | } |
| 481 | const_iterator operator-(difference_type i) const { |
| 482 | const_iterator tmp(*this); |
| 483 | tmp -= i; |
| 484 | return tmp; |
| 485 | } |
| 486 | difference_type operator-(const_iterator it) const { // for "x = it2 - it" |
| 487 | assert(table == it.table); |
| 488 | return pos - it.pos; |
| 489 | } |
| 490 | reference operator[](difference_type n) const { |
| 491 | return *(*this + n); // simple though not totally efficient |
| 492 | } |
| 493 | |
| 494 | // Comparisons. |
| 495 | bool operator==(const const_iterator& it) const { |
| 496 | return table == it.table && pos == it.pos; |
| 497 | } |
| 498 | bool operator<(const const_iterator& it) const { |
| 499 | assert(table == it.table); // life is bad bad bad otherwise |
| 500 | return pos < it.pos; |
| 501 | } |
| 502 | bool operator!=(const const_iterator& it) const { return !(*this == it); } |
| 503 | bool operator<=(const const_iterator& it) const { return !(it < *this); } |
| 504 | bool operator>(const const_iterator& it) const { return it < *this; } |
| 505 | bool operator>=(const const_iterator& it) const { return !(*this < it); } |
| 506 | |
| 507 | // Here's the info we actually need to be an iterator |
| 508 | const tabletype* table; // so we can dereference and bounds-check |
| 509 | size_type pos; // index into the table |
| 510 | }; |
| 511 | |
| 512 | // support for "3 + iterator" has to be defined outside the class, alas |
| 513 | template <class T> |
| 514 | const_table_iterator<T> operator+( |
| 515 | typename const_table_iterator<T>::difference_type i, |
| 516 | const_table_iterator<T> it) { |
| 517 | return it + i; // so people can say it2 = 3 + it |
| 518 | } |
| 519 | |
| 520 | // --------------------------------------------------------------------------- |
| 521 | |
| 522 | /* |
| 523 | // This is a 2-D iterator. You specify a begin and end over a list |
| 524 | // of *containers*. We iterate over each container by iterating over |
| 525 | // it. It's actually simple: |
| 526 | // VECTOR.begin() VECTOR[0].begin() --------> VECTOR[0].end() ---, |
| 527 | // | ________________________________________________/ |
| 528 | // | \_> VECTOR[1].begin() --------> VECTOR[1].end() -, |
| 529 | // | ___________________________________________________/ |
| 530 | // v \_> ...... |
| 531 | // VECTOR.end() |
| 532 | // |
| 533 | // It's impossible to do random access on one of these things in constant |
| 534 | // time, so it's just a bidirectional iterator. |
| 535 | // |
| 536 | // Unfortunately, because we need to use this for a non-empty iterator, |
| 537 | // we use nonempty_begin() and nonempty_end() instead of begin() and end() |
| 538 | // (though only going across, not down). |
| 539 | */ |
| 540 | |
| 541 | #define TWOD_BEGIN_ nonempty_begin |
| 542 | #define TWOD_END_ nonempty_end |
| 543 | #define TWOD_ITER_ nonempty_iterator |
| 544 | #define TWOD_CONST_ITER_ const_nonempty_iterator |
| 545 | |
| 546 | template <class containertype> |
| 547 | class two_d_iterator { |
| 548 | public: |
| 549 | typedef two_d_iterator iterator; |
| 550 | |
| 551 | typedef std::bidirectional_iterator_tag iterator_category; |
| 552 | // apparently some versions of VC++ have trouble with two ::'s in a typename |
| 553 | typedef typename containertype::value_type _tmp_vt; |
| 554 | typedef typename _tmp_vt::value_type value_type; |
| 555 | typedef typename _tmp_vt::difference_type difference_type; |
| 556 | typedef typename _tmp_vt::reference reference; |
| 557 | typedef typename _tmp_vt::pointer pointer; |
| 558 | |
| 559 | // The "real" constructor. begin and end specify how many rows we have |
| 560 | // (in the diagram above); we always iterate over each row completely. |
| 561 | two_d_iterator(typename containertype::iterator begin, |
| 562 | typename containertype::iterator end, |
| 563 | typename containertype::iterator curr) |
| 564 | : row_begin(begin), row_end(end), row_current(curr), col_current() { |
| 565 | if (row_current != row_end) { |
| 566 | col_current = row_current->TWOD_BEGIN_(); |
| 567 | advance_past_end(); // in case cur->begin() == cur->end() |
| 568 | } |
| 569 | } |
| 570 | // If you want to start at an arbitrary place, you can, I guess |
| 571 | two_d_iterator(typename containertype::iterator begin, |
| 572 | typename containertype::iterator end, |
| 573 | typename containertype::iterator curr, |
| 574 | typename containertype::value_type::TWOD_ITER_ col) |
| 575 | : row_begin(begin), row_end(end), row_current(curr), col_current(col) { |
| 576 | advance_past_end(); // in case cur->begin() == cur->end() |
| 577 | } |
| 578 | // The default constructor, used when I define vars of type table::iterator |
| 579 | two_d_iterator() : row_begin(), row_end(), row_current(), col_current() {} |
| 580 | // The default destructor is fine; we don't define one |
| 581 | // The default operator= is fine; we don't define one |
| 582 | |
| 583 | // Happy dereferencer |
| 584 | reference operator*() const { return *col_current; } |
| 585 | pointer operator->() const { return &(operator*()); } |
| 586 | |
| 587 | // Arithmetic: we just do arithmetic on pos. We don't even need to |
| 588 | // do bounds checking, since STL doesn't consider that its job. :-) |
| 589 | // NOTE: this is not amortized constant time! What do we do about it? |
| 590 | void advance_past_end() { // used when col_current points to end() |
| 591 | while (col_current == row_current->TWOD_END_()) { // end of current row |
| 592 | ++row_current; // go to beginning of next |
| 593 | if (row_current != row_end) // col is irrelevant at end |
| 594 | col_current = row_current->TWOD_BEGIN_(); |
| 595 | else |
| 596 | break; // don't go past row_end |
| 597 | } |
| 598 | } |
| 599 | |
| 600 | iterator& operator++() { |
| 601 | assert(row_current != row_end); // how to ++ from there? |
| 602 | ++col_current; |
| 603 | advance_past_end(); // in case col_current is at end() |
| 604 | return *this; |
| 605 | } |
| 606 | iterator& operator--() { |
| 607 | while (row_current == row_end || |
| 608 | col_current == row_current->TWOD_BEGIN_()) { |
| 609 | assert(row_current != row_begin); |
| 610 | --row_current; |
| 611 | col_current = row_current->TWOD_END_(); // this is 1 too far |
| 612 | } |
| 613 | --col_current; |
| 614 | return *this; |
| 615 | } |
| 616 | iterator operator++(int) { |
| 617 | iterator tmp(*this); |
| 618 | ++*this; |
| 619 | return tmp; |
| 620 | } |
| 621 | iterator operator--(int) { |
| 622 | iterator tmp(*this); |
| 623 | --*this; |
| 624 | return tmp; |
| 625 | } |
| 626 | |
| 627 | // Comparisons. |
| 628 | bool operator==(const iterator& it) const { |
| 629 | return (row_begin == it.row_begin && row_end == it.row_end && |
| 630 | row_current == it.row_current && |
| 631 | (row_current == row_end || col_current == it.col_current)); |
| 632 | } |
| 633 | bool operator!=(const iterator& it) const { return !(*this == it); } |
| 634 | |
| 635 | // Here's the info we actually need to be an iterator |
| 636 | // These need to be public so we convert from iterator to const_iterator |
| 637 | typename containertype::iterator row_begin, row_end, row_current; |
| 638 | typename containertype::value_type::TWOD_ITER_ col_current; |
| 639 | }; |
| 640 | |
| 641 | // The same thing again, but this time const. :-( |
| 642 | template <class containertype> |
| 643 | class const_two_d_iterator { |
| 644 | public: |
| 645 | typedef const_two_d_iterator iterator; |
| 646 | |
| 647 | typedef std::bidirectional_iterator_tag iterator_category; |
| 648 | // apparently some versions of VC++ have trouble with two ::'s in a typename |
| 649 | typedef typename containertype::value_type _tmp_vt; |
| 650 | typedef typename _tmp_vt::value_type value_type; |
| 651 | typedef typename _tmp_vt::difference_type difference_type; |
| 652 | typedef typename _tmp_vt::const_reference reference; |
| 653 | typedef typename _tmp_vt::const_pointer pointer; |
| 654 | |
| 655 | const_two_d_iterator(typename containertype::const_iterator begin, |
| 656 | typename containertype::const_iterator end, |
| 657 | typename containertype::const_iterator curr) |
| 658 | : row_begin(begin), row_end(end), row_current(curr), col_current() { |
| 659 | if (curr != end) { |
| 660 | col_current = curr->TWOD_BEGIN_(); |
| 661 | advance_past_end(); // in case cur->begin() == cur->end() |
| 662 | } |
| 663 | } |
| 664 | const_two_d_iterator(typename containertype::const_iterator begin, |
| 665 | typename containertype::const_iterator end, |
| 666 | typename containertype::const_iterator curr, |
| 667 | typename containertype::value_type::TWOD_CONST_ITER_ col) |
| 668 | : row_begin(begin), row_end(end), row_current(curr), col_current(col) { |
| 669 | advance_past_end(); // in case cur->begin() == cur->end() |
| 670 | } |
| 671 | const_two_d_iterator() |
| 672 | : row_begin(), row_end(), row_current(), col_current() {} |
| 673 | // Need this explicitly so we can convert normal iterators to const |
| 674 | // iterators |
| 675 | const_two_d_iterator(const two_d_iterator<containertype>& it) |
| 676 | : row_begin(it.row_begin), |
| 677 | row_end(it.row_end), |
| 678 | row_current(it.row_current), |
| 679 | col_current(it.col_current) {} |
| 680 | |
| 681 | typename containertype::const_iterator row_begin, row_end, row_current; |
| 682 | typename containertype::value_type::TWOD_CONST_ITER_ col_current; |
| 683 | |
| 684 | // EVERYTHING FROM HERE DOWN IS THE SAME AS THE NON-CONST ITERATOR |
| 685 | reference operator*() const { return *col_current; } |
| 686 | pointer operator->() const { return &(operator*()); } |
| 687 | |
| 688 | void advance_past_end() { // used when col_current points to end() |
| 689 | while (col_current == row_current->TWOD_END_()) { // end of current row |
| 690 | ++row_current; // go to beginning of next |
| 691 | if (row_current != row_end) // col is irrelevant at end |
| 692 | col_current = row_current->TWOD_BEGIN_(); |
| 693 | else |
| 694 | break; // don't go past row_end |
| 695 | } |
| 696 | } |
| 697 | iterator& operator++() { |
| 698 | assert(row_current != row_end); // how to ++ from there? |
| 699 | ++col_current; |
| 700 | advance_past_end(); // in case col_current is at end() |
| 701 | return *this; |
| 702 | } |
| 703 | iterator& operator--() { |
| 704 | while (row_current == row_end || |
| 705 | col_current == row_current->TWOD_BEGIN_()) { |
| 706 | assert(row_current != row_begin); |
| 707 | --row_current; |
| 708 | col_current = row_current->TWOD_END_(); // this is 1 too far |
| 709 | } |
| 710 | --col_current; |
| 711 | return *this; |
| 712 | } |
| 713 | iterator operator++(int) { |
| 714 | iterator tmp(*this); |
| 715 | ++*this; |
| 716 | return tmp; |
| 717 | } |
| 718 | iterator operator--(int) { |
| 719 | iterator tmp(*this); |
| 720 | --*this; |
| 721 | return tmp; |
| 722 | } |
| 723 | |
| 724 | bool operator==(const iterator& it) const { |
| 725 | return (row_begin == it.row_begin && row_end == it.row_end && |
| 726 | row_current == it.row_current && |
| 727 | (row_current == row_end || col_current == it.col_current)); |
| 728 | } |
| 729 | bool operator!=(const iterator& it) const { return !(*this == it); } |
| 730 | }; |
| 731 | |
| 732 | // We provide yet another version, to be as frugal with memory as |
| 733 | // possible. This one frees each block of memory as it finishes |
| 734 | // iterating over it. By the end, the entire table is freed. |
| 735 | // For understandable reasons, you can only iterate over it once, |
| 736 | // which is why it's an input iterator |
| 737 | template <class containertype> |
| 738 | class destructive_two_d_iterator { |
| 739 | public: |
| 740 | typedef destructive_two_d_iterator iterator; |
| 741 | |
| 742 | typedef std::input_iterator_tag iterator_category; |
| 743 | // apparently some versions of VC++ have trouble with two ::'s in a typename |
| 744 | typedef typename containertype::value_type _tmp_vt; |
| 745 | typedef typename _tmp_vt::value_type value_type; |
| 746 | typedef typename _tmp_vt::difference_type difference_type; |
| 747 | typedef typename _tmp_vt::reference reference; |
| 748 | typedef typename _tmp_vt::pointer pointer; |
| 749 | |
| 750 | destructive_two_d_iterator(typename containertype::iterator begin, |
| 751 | typename containertype::iterator end, |
| 752 | typename containertype::iterator curr) |
| 753 | : row_begin(begin), row_end(end), row_current(curr), col_current() { |
| 754 | if (curr != end) { |
| 755 | col_current = curr->TWOD_BEGIN_(); |
| 756 | advance_past_end(); // in case cur->begin() == cur->end() |
| 757 | } |
| 758 | } |
| 759 | destructive_two_d_iterator(typename containertype::iterator begin, |
| 760 | typename containertype::iterator end, |
| 761 | typename containertype::iterator curr, |
| 762 | typename containertype::value_type::TWOD_ITER_ col) |
| 763 | : row_begin(begin), row_end(end), row_current(curr), col_current(col) { |
| 764 | advance_past_end(); // in case cur->begin() == cur->end() |
| 765 | } |
| 766 | destructive_two_d_iterator() |
| 767 | : row_begin(), row_end(), row_current(), col_current() {} |
| 768 | |
| 769 | typename containertype::iterator row_begin, row_end, row_current; |
| 770 | typename containertype::value_type::TWOD_ITER_ col_current; |
| 771 | |
| 772 | // This is the part that destroys |
| 773 | void advance_past_end() { // used when col_current points to end() |
| 774 | while (col_current == row_current->TWOD_END_()) { // end of current row |
| 775 | row_current->clear(); // the destructive part |
| 776 | // It would be nice if we could decrement sparsetable->num_buckets |
| 777 | // here |
| 778 | ++row_current; // go to beginning of next |
| 779 | if (row_current != row_end) // col is irrelevant at end |
| 780 | col_current = row_current->TWOD_BEGIN_(); |
| 781 | else |
| 782 | break; // don't go past row_end |
| 783 | } |
| 784 | } |
| 785 | |
| 786 | // EVERYTHING FROM HERE DOWN IS THE SAME AS THE REGULAR ITERATOR |
| 787 | reference operator*() const { return *col_current; } |
| 788 | pointer operator->() const { return &(operator*()); } |
| 789 | |
| 790 | iterator& operator++() { |
| 791 | assert(row_current != row_end); // how to ++ from there? |
| 792 | ++col_current; |
| 793 | advance_past_end(); // in case col_current is at end() |
| 794 | return *this; |
| 795 | } |
| 796 | iterator operator++(int) { |
| 797 | iterator tmp(*this); |
| 798 | ++*this; |
| 799 | return tmp; |
| 800 | } |
| 801 | |
| 802 | bool operator==(const iterator& it) const { |
| 803 | return (row_begin == it.row_begin && row_end == it.row_end && |
| 804 | row_current == it.row_current && |
| 805 | (row_current == row_end || col_current == it.col_current)); |
| 806 | } |
| 807 | bool operator!=(const iterator& it) const { return !(*this == it); } |
| 808 | }; |
| 809 | |
| 810 | #undef TWOD_BEGIN_ |
| 811 | #undef TWOD_END_ |
| 812 | #undef TWOD_ITER_ |
| 813 | #undef TWOD_CONST_ITER_ |
| 814 | |
| 815 | // SPARSE-TABLE |
| 816 | // ------------ |
| 817 | // The idea is that a table with (logically) t buckets is divided |
| 818 | // into t/M *groups* of M buckets each. (M is a constant set in |
| 819 | // GROUP_SIZE for efficiency.) Each group is stored sparsely. |
| 820 | // Thus, inserting into the table causes some array to grow, which is |
| 821 | // slow but still constant time. Lookup involves doing a |
| 822 | // logical-position-to-sparse-position lookup, which is also slow but |
| 823 | // constant time. The larger M is, the slower these operations are |
| 824 | // but the less overhead (slightly). |
| 825 | // |
| 826 | // To store the sparse array, we store a bitmap B, where B[i] = 1 iff |
| 827 | // bucket i is non-empty. Then to look up bucket i we really look up |
| 828 | // array[# of 1s before i in B]. This is constant time for fixed M. |
| 829 | // |
| 830 | // Terminology: the position of an item in the overall table (from |
| 831 | // 1 .. t) is called its "location." The logical position in a group |
| 832 | // (from 1 .. M ) is called its "position." The actual location in |
| 833 | // the array (from 1 .. # of non-empty buckets in the group) is |
| 834 | // called its "offset." |
| 835 | |
| 836 | template <class T, uint16_t GROUP_SIZE, class Alloc> |
| 837 | class sparsegroup { |
| 838 | public: |
| 839 | typedef T value_type; |
| 840 | typedef Alloc allocator_type; |
| 841 | |
| 842 | private: |
| 843 | using value_alloc_type = |
| 844 | typename std::allocator_traits<Alloc>::template rebind_alloc<T>; |
| 845 | typedef std::integral_constant< |
| 846 | bool, (is_relocatable<value_type>::value && |
| 847 | std::is_same<allocator_type, |
| 848 | libc_allocator_with_realloc<value_type>>::value)> |
| 849 | realloc_and_memmove_ok; // we pretend mv(x,y) == "x.~T(); |
| 850 | // new(x) T(y)" |
| 851 | public: |
| 852 | // Basic types |
| 853 | typedef typename value_alloc_type::reference reference; |
| 854 | typedef typename value_alloc_type::const_reference const_reference; |
| 855 | typedef typename value_alloc_type::pointer pointer; |
| 856 | typedef typename value_alloc_type::const_pointer const_pointer; |
| 857 | |
| 858 | typedef table_iterator<sparsegroup<T, GROUP_SIZE, Alloc>> iterator; |
| 859 | typedef const_table_iterator<sparsegroup<T, GROUP_SIZE, Alloc>> |
| 860 | const_iterator; |
| 861 | typedef table_element_adaptor<sparsegroup<T, GROUP_SIZE, Alloc>> |
| 862 | element_adaptor; |
| 863 | typedef uint16_t size_type; // max # of buckets |
| 864 | typedef int16_t difference_type; |
| 865 | typedef std::reverse_iterator<const_iterator> const_reverse_iterator; |
| 866 | typedef std::reverse_iterator<iterator> reverse_iterator; // from iterator.h |
| 867 | |
| 868 | // These are our special iterators, that go over non-empty buckets in a |
| 869 | // group. These aren't const-only because you can change non-empty bcks. |
| 870 | typedef pointer nonempty_iterator; |
| 871 | typedef const_pointer const_nonempty_iterator; |
| 872 | typedef std::reverse_iterator<nonempty_iterator> reverse_nonempty_iterator; |
| 873 | typedef std::reverse_iterator<const_nonempty_iterator> |
| 874 | const_reverse_nonempty_iterator; |
| 875 | |
| 876 | // Iterator functions |
| 877 | iterator begin() { return iterator(this, 0); } |
| 878 | const_iterator begin() const { return const_iterator(this, 0); } |
| 879 | iterator end() { return iterator(this, size()); } |
| 880 | const_iterator end() const { return const_iterator(this, size()); } |
| 881 | reverse_iterator rbegin() { return reverse_iterator(end()); } |
| 882 | const_reverse_iterator rbegin() const { |
| 883 | return const_reverse_iterator(end()); |
| 884 | } |
| 885 | reverse_iterator rend() { return reverse_iterator(begin()); } |
| 886 | const_reverse_iterator rend() const { |
| 887 | return const_reverse_iterator(begin()); |
| 888 | } |
| 889 | |
| 890 | // We'll have versions for our special non-empty iterator too |
| 891 | nonempty_iterator nonempty_begin() { return group; } |
| 892 | const_nonempty_iterator nonempty_begin() const { return group; } |
| 893 | nonempty_iterator nonempty_end() { return group + settings.num_buckets; } |
| 894 | const_nonempty_iterator nonempty_end() const { |
| 895 | return group + settings.num_buckets; |
| 896 | } |
| 897 | reverse_nonempty_iterator nonempty_rbegin() { |
| 898 | return reverse_nonempty_iterator(nonempty_end()); |
| 899 | } |
| 900 | const_reverse_nonempty_iterator nonempty_rbegin() const { |
| 901 | return const_reverse_nonempty_iterator(nonempty_end()); |
| 902 | } |
| 903 | reverse_nonempty_iterator nonempty_rend() { |
| 904 | return reverse_nonempty_iterator(nonempty_begin()); |
| 905 | } |
| 906 | const_reverse_nonempty_iterator nonempty_rend() const { |
| 907 | return const_reverse_nonempty_iterator(nonempty_begin()); |
| 908 | } |
| 909 | |
| 910 | // This gives us the "default" value to return for an empty bucket. |
| 911 | // We just use the default constructor on T, the template type |
| 912 | const_reference default_value() const { |
| 913 | static value_type defaultval = value_type(); |
| 914 | return defaultval; |
| 915 | } |
| 916 | |
| 917 | private: |
| 918 | // We need to do all this bit manipulation, of course. ick |
| 919 | static size_type charbit(size_type i) { return i >> 3; } |
| 920 | static size_type modbit(size_type i) { return 1 << (i & 7); } |
| 921 | int bmtest(size_type i) const { return bitmap[charbit(i)] & modbit(i); } |
| 922 | void bmset(size_type i) { bitmap[charbit(i)] |= modbit(i); } |
| 923 | void bmclear(size_type i) { bitmap[charbit(i)] &= ~modbit(i); } |
| 924 | |
| 925 | pointer allocate_group(size_type n) { |
| 926 | pointer retval = settings.allocate(n); |
| 927 | if (retval == NULL) { |
| 928 | // We really should use PRIuS here, but I don't want to have to add |
| 929 | // a whole new configure option, with concomitant macro namespace |
| 930 | // pollution, just to print this (unlikely) error message. So I |
| 931 | // cast. |
| 932 | fprintf(stderr, "sparsehash FATAL ERROR: failed to allocate %lu groups\n", |
| 933 | static_cast<unsigned long>(n)); |
| 934 | exit(1); |
| 935 | } |
| 936 | return retval; |
| 937 | } |
| 938 | |
| 939 | void free_group() { |
| 940 | if (!group) return; |
| 941 | pointer end_it = group + settings.num_buckets; |
| 942 | for (pointer p = group; p != end_it; ++p) p->~value_type(); |
| 943 | settings.deallocate(group, settings.num_buckets); |
| 944 | group = NULL; |
| 945 | } |
| 946 | |
| 947 | static size_type bits_in_char(unsigned char c) { |
| 948 | // We could make these ints. The tradeoff is size (eg does it overwhelm |
| 949 | // the cache?) vs efficiency in referencing sub-word-sized array |
| 950 | // elements. |
| 951 | static const char bits_in[256] = { |
| 952 | 0, 1, 1, 2, 1, 2, 2, 3, 1, 2, 2, 3, 2, 3, 3, 4, 1, 2, 2, 3, 2, 3, 3, 4, |
| 953 | 2, 3, 3, 4, 3, 4, 4, 5, 1, 2, 2, 3, 2, 3, 3, 4, 2, 3, 3, 4, 3, 4, 4, 5, |
| 954 | 2, 3, 3, 4, 3, 4, 4, 5, 3, 4, 4, 5, 4, 5, 5, 6, 1, 2, 2, 3, 2, 3, 3, 4, |
| 955 | 2, 3, 3, 4, 3, 4, 4, 5, 2, 3, 3, 4, 3, 4, 4, 5, 3, 4, 4, 5, 4, 5, 5, 6, |
| 956 | 2, 3, 3, 4, 3, 4, 4, 5, 3, 4, 4, 5, 4, 5, 5, 6, 3, 4, 4, 5, 4, 5, 5, 6, |
| 957 | 4, 5, 5, 6, 5, 6, 6, 7, 1, 2, 2, 3, 2, 3, 3, 4, 2, 3, 3, 4, 3, 4, 4, 5, |
| 958 | 2, 3, 3, 4, 3, 4, 4, 5, 3, 4, 4, 5, 4, 5, 5, 6, 2, 3, 3, 4, 3, 4, 4, 5, |
| 959 | 3, 4, 4, 5, 4, 5, 5, 6, 3, 4, 4, 5, 4, 5, 5, 6, 4, 5, 5, 6, 5, 6, 6, 7, |
| 960 | 2, 3, 3, 4, 3, 4, 4, 5, 3, 4, 4, 5, 4, 5, 5, 6, 3, 4, 4, 5, 4, 5, 5, 6, |
| 961 | 4, 5, 5, 6, 5, 6, 6, 7, 3, 4, 4, 5, 4, 5, 5, 6, 4, 5, 5, 6, 5, 6, 6, 7, |
| 962 | 4, 5, 5, 6, 5, 6, 6, 7, 5, 6, 6, 7, 6, 7, 7, 8, |
| 963 | }; |
| 964 | return bits_in[c]; |
| 965 | } |
| 966 | |
| 967 | public: // get_iter() in sparsetable needs it |
| 968 | // We need a small function that tells us how many set bits there are |
| 969 | // in positions 0..i-1 of the bitmap. It uses a big table. |
| 970 | // We make it static so templates don't allocate lots of these tables. |
| 971 | // There are lots of ways to do this calculation (called 'popcount'). |
| 972 | // The 8-bit table lookup is one of the fastest, though this |
| 973 | // implementation suffers from not doing any loop unrolling. See, eg, |
| 974 | // http://www.dalkescientific.com/writings/diary/archive/2008/07/03/hakmem_and_other_popcounts.html |
| 975 | // http://gurmeetsingh.wordpress.com/2008/08/05/fast-bit-counting-routines/ |
| 976 | static size_type pos_to_offset(const unsigned char* bm, size_type pos) { |
| 977 | size_type retval = 0; |
| 978 | |
| 979 | // [Note: condition pos > 8 is an optimization; convince yourself we |
| 980 | // give exactly the same result as if we had pos >= 8 here instead.] |
| 981 | for (; pos > 8; pos -= 8) // bm[0..pos/8-1] |
| 982 | retval += bits_in_char(*bm++); // chars we want *all* bits in |
| 983 | return retval + bits_in_char(*bm & ((1 << pos) - 1)); // char including pos |
| 984 | } |
| 985 | |
| 986 | size_type pos_to_offset(size_type pos) const { // not static but still const |
| 987 | return pos_to_offset(bitmap, pos); |
| 988 | } |
| 989 | |
| 990 | // Returns the (logical) position in the bm[] array, i, such that |
| 991 | // bm[i] is the offset-th set bit in the array. It is the inverse |
| 992 | // of pos_to_offset. get_pos() uses this function to find the index |
| 993 | // of an nonempty_iterator in the table. Bit-twiddling from |
| 994 | // http://hackersdelight.org/basics.pdf |
| 995 | static size_type offset_to_pos(const unsigned char* bm, size_type offset) { |
| 996 | size_type retval = 0; |
| 997 | // This is sizeof(this->bitmap). |
| 998 | const size_type group_size = (GROUP_SIZE - 1) / 8 + 1; |
| 999 | for (size_type i = 0; i < group_size; i++) { // forward scan |
| 1000 | const size_type pop_count = bits_in_char(*bm); |
| 1001 | if (pop_count > offset) { |
| 1002 | unsigned char last_bm = *bm; |
| 1003 | for (; offset > 0; offset--) { |
| 1004 | last_bm &= (last_bm - 1); // remove right-most set bit |
| 1005 | } |
| 1006 | // Clear all bits to the left of the rightmost bit (the &), |
| 1007 | // and then clear the rightmost bit but set all bits to the |
| 1008 | // right of it (the -1). |
| 1009 | last_bm = (last_bm & -last_bm) - 1; |
| 1010 | retval += bits_in_char(last_bm); |
| 1011 | return retval; |
| 1012 | } |
| 1013 | offset -= pop_count; |
| 1014 | retval += 8; |
| 1015 | bm++; |
| 1016 | } |
| 1017 | return retval; |
| 1018 | } |
| 1019 | |
| 1020 | size_type offset_to_pos(size_type offset) const { |
| 1021 | return offset_to_pos(bitmap, offset); |
| 1022 | } |
| 1023 | |
| 1024 | public: |
| 1025 | // Constructors -- default and copy -- and destructor |
| 1026 | explicit sparsegroup(allocator_type& a) |
| 1027 | : group(0), settings(alloc_impl<value_alloc_type>(a)) { |
| 1028 | memset(bitmap, 0, sizeof(bitmap)); |
| 1029 | } |
| 1030 | sparsegroup(const sparsegroup& x) : group(0), settings(x.settings) { |
| 1031 | if (settings.num_buckets) { |
| 1032 | group = allocate_group(x.settings.num_buckets); |
| 1033 | std::uninitialized_copy(x.group, x.group + x.settings.num_buckets, group); |
| 1034 | } |
| 1035 | memcpy(bitmap, x.bitmap, sizeof(bitmap)); |
| 1036 | } |
| 1037 | ~sparsegroup() { free_group(); } |
| 1038 | |
| 1039 | // Operator= is just like the copy constructor, I guess |
| 1040 | // TODO(austern): Make this exception safe. Handle exceptions in |
| 1041 | // value_type's |
| 1042 | // copy constructor. |
| 1043 | sparsegroup& operator=(const sparsegroup& x) { |
| 1044 | if (&x == this) return *this; // x = x |
| 1045 | if (x.settings.num_buckets == 0) { |
| 1046 | free_group(); |
| 1047 | } else { |
| 1048 | pointer p = allocate_group(x.settings.num_buckets); |
| 1049 | std::uninitialized_copy(x.group, x.group + x.settings.num_buckets, p); |
| 1050 | free_group(); |
| 1051 | group = p; |
| 1052 | } |
| 1053 | memcpy(bitmap, x.bitmap, sizeof(bitmap)); |
| 1054 | settings.num_buckets = x.settings.num_buckets; |
| 1055 | return *this; |
| 1056 | } |
| 1057 | |
| 1058 | // Many STL algorithms use swap instead of copy constructors |
| 1059 | void swap(sparsegroup& x) { |
| 1060 | std::swap(group, x.group); // defined in <algorithm> |
| 1061 | for (int i = 0; i < sizeof(bitmap) / sizeof(*bitmap); ++i) |
| 1062 | std::swap(bitmap[i], x.bitmap[i]); // swap not defined on arrays |
| 1063 | std::swap(settings.num_buckets, x.settings.num_buckets); |
| 1064 | // we purposefully don't swap the allocator, which may not be swap-able |
| 1065 | } |
| 1066 | |
| 1067 | // It's always nice to be able to clear a table without deallocating it |
| 1068 | void clear() { |
| 1069 | free_group(); |
| 1070 | memset(bitmap, 0, sizeof(bitmap)); |
| 1071 | settings.num_buckets = 0; |
| 1072 | } |
| 1073 | |
| 1074 | // Functions that tell you about size. Alas, these aren't so useful |
| 1075 | // because our table is always fixed size. |
| 1076 | size_type size() const { return GROUP_SIZE; } |
| 1077 | size_type max_size() const { return GROUP_SIZE; } |
| 1078 | bool empty() const { return false; } |
| 1079 | // We also may want to know how many *used* buckets there are |
| 1080 | size_type num_nonempty() const { return settings.num_buckets; } |
| 1081 | |
| 1082 | // get()/set() are explicitly const/non-const. You can use [] if |
| 1083 | // you want something that can be either (potentially more expensive). |
| 1084 | const_reference get(size_type i) const { |
| 1085 | if (bmtest(i)) // bucket i is occupied |
| 1086 | return group[pos_to_offset(bitmap, i)]; |
| 1087 | else |
| 1088 | return default_value(); // return the default reference |
| 1089 | } |
| 1090 | |
| 1091 | // TODO(csilvers): make protected + friend |
| 1092 | // This is used by sparse_hashtable to get an element from the table |
| 1093 | // when we know it exists. |
| 1094 | const_reference unsafe_get(size_type i) const { |
| 1095 | assert(bmtest(i)); |
| 1096 | return group[pos_to_offset(bitmap, i)]; |
| 1097 | } |
| 1098 | |
| 1099 | // TODO(csilvers): make protected + friend |
| 1100 | reference mutating_get(size_type i) { // fills bucket i before getting |
| 1101 | if (!bmtest(i)) set(i, default_value()); |
| 1102 | return group[pos_to_offset(bitmap, i)]; |
| 1103 | } |
| 1104 | |
| 1105 | // Syntactic sugar. It's easy to return a const reference. To |
| 1106 | // return a non-const reference, we need to use the assigner adaptor. |
| 1107 | const_reference operator[](size_type i) const { return get(i); } |
| 1108 | |
| 1109 | element_adaptor operator[](size_type i) { return element_adaptor(this, i); } |
| 1110 | |
| 1111 | private: |
| 1112 | // Create space at group[offset], assuming value_type has trivial |
| 1113 | // copy constructor and destructor, and the allocator_type is |
| 1114 | // the default libc_allocator_with_alloc. (Really, we want it to have |
| 1115 | // "trivial move", because that's what realloc and memmove both do. |
| 1116 | // But there's no way to capture that using type_traits, so we |
| 1117 | // pretend that move(x, y) is equivalent to "x.~T(); new(x) T(y);" |
| 1118 | // which is pretty much correct, if a bit conservative.) |
| 1119 | void set_aux(size_type offset, std::true_type) { |
| 1120 | group = settings.realloc_or_die(group, settings.num_buckets + 1); |
| 1121 | // This is equivalent to memmove(), but faster on my Intel P4, |
| 1122 | // at least with gcc4.1 -O2 / glibc 2.3.6. |
| 1123 | for (size_type i = settings.num_buckets; i > offset; --i) |
| 1124 | // cast to void* to prevent compiler warnings about writing to an object |
| 1125 | // with no trivial copy-assignment |
| 1126 | memcpy(static_cast<void*>(group + i), group + i - 1, sizeof(*group)); |
| 1127 | } |
| 1128 | |
| 1129 | // Create space at group[offset], without special assumptions about |
| 1130 | // value_type |
| 1131 | // and allocator_type. |
| 1132 | void set_aux(size_type offset, std::false_type) { |
| 1133 | // This is valid because 0 <= offset <= num_buckets |
| 1134 | pointer p = allocate_group(settings.num_buckets + 1); |
| 1135 | std::uninitialized_copy(group, group + offset, p); |
| 1136 | std::uninitialized_copy(group + offset, group + settings.num_buckets, |
| 1137 | p + offset + 1); |
| 1138 | free_group(); |
| 1139 | group = p; |
| 1140 | } |
| 1141 | |
| 1142 | public: |
| 1143 | // This returns a reference to the inserted item (which is a copy of val). |
| 1144 | // TODO(austern): Make this exception safe: handle exceptions from |
| 1145 | // value_type's copy constructor. |
| 1146 | reference set(size_type i, const_reference val) { |
| 1147 | size_type offset = |
| 1148 | pos_to_offset(bitmap, i); // where we'll find (or insert) |
| 1149 | if (bmtest(i)) { |
| 1150 | // Delete the old value, which we're replacing with the new one |
| 1151 | group[offset].~value_type(); |
| 1152 | } else { |
| 1153 | set_aux(offset, realloc_and_memmove_ok()); |
| 1154 | ++settings.num_buckets; |
| 1155 | bmset(i); |
| 1156 | } |
| 1157 | // This does the actual inserting. Since we made the array using |
| 1158 | // malloc, we use "placement new" to just call the constructor. |
| 1159 | new (&group[offset]) value_type(val); |
| 1160 | return group[offset]; |
| 1161 | } |
| 1162 | |
| 1163 | // We let you see if a bucket is non-empty without retrieving it |
| 1164 | bool test(size_type i) const { return bmtest(i) != 0; } |
| 1165 | bool test(iterator pos) const { return bmtest(pos.pos) != 0; } |
| 1166 | |
| 1167 | private: |
| 1168 | // Shrink the array, assuming value_type has trivial copy |
| 1169 | // constructor and destructor, and the allocator_type is the default |
| 1170 | // libc_allocator_with_alloc. (Really, we want it to have "trivial |
| 1171 | // move", because that's what realloc and memmove both do. But |
| 1172 | // there's no way to capture that using type_traits, so we pretend |
| 1173 | // that move(x, y) is equivalent to ""x.~T(); new(x) T(y);" |
| 1174 | // which is pretty much correct, if a bit conservative.) |
| 1175 | void erase_aux(size_type offset, std::true_type) { |
| 1176 | // This isn't technically necessary, since we know we have a |
| 1177 | // trivial destructor, but is a cheap way to get a bit more safety. |
| 1178 | group[offset].~value_type(); |
| 1179 | // This is equivalent to memmove(), but faster on my Intel P4, |
| 1180 | // at lesat with gcc4.1 -O2 / glibc 2.3.6. |
| 1181 | assert(settings.num_buckets > 0); |
| 1182 | for (size_type i = offset; i < settings.num_buckets - 1; ++i) |
| 1183 | // cast to void* to prevent compiler warnings about writing to an object |
| 1184 | // with no trivial copy-assignment |
| 1185 | // hopefully inlined! |
| 1186 | memcpy(static_cast<void*>(group + i), group + i + 1, sizeof(*group)); |
| 1187 | group = settings.realloc_or_die(group, settings.num_buckets - 1); |
| 1188 | } |
| 1189 | |
| 1190 | // Shrink the array, without any special assumptions about value_type and |
| 1191 | // allocator_type. |
| 1192 | void erase_aux(size_type offset, std::false_type) { |
| 1193 | // This is valid because 0 <= offset < num_buckets. Note the inequality. |
| 1194 | pointer p = allocate_group(settings.num_buckets - 1); |
| 1195 | std::uninitialized_copy(group, group + offset, p); |
| 1196 | std::uninitialized_copy(group + offset + 1, group + settings.num_buckets, |
| 1197 | p + offset); |
| 1198 | free_group(); |
| 1199 | group = p; |
| 1200 | } |
| 1201 | |
| 1202 | public: |
| 1203 | // This takes the specified elements out of the group. This is |
| 1204 | // "undefining", rather than "clearing". |
| 1205 | // TODO(austern): Make this exception safe: handle exceptions from |
| 1206 | // value_type's copy constructor. |
| 1207 | void erase(size_type i) { |
| 1208 | if (bmtest(i)) { // trivial to erase empty bucket |
| 1209 | size_type offset = |
| 1210 | pos_to_offset(bitmap, i); // where we'll find (or insert) |
| 1211 | if (settings.num_buckets == 1) { |
| 1212 | free_group(); |
| 1213 | group = NULL; |
| 1214 | } else { |
| 1215 | erase_aux(offset, realloc_and_memmove_ok()); |
| 1216 | } |
| 1217 | --settings.num_buckets; |
| 1218 | bmclear(i); |
| 1219 | } |
| 1220 | } |
| 1221 | |
| 1222 | void erase(iterator pos) { erase(pos.pos); } |
| 1223 | |
| 1224 | void erase(iterator start_it, iterator end_it) { |
| 1225 | // This could be more efficient, but to do so we'd need to make |
| 1226 | // bmclear() clear a range of indices. Doesn't seem worth it. |
| 1227 | for (; start_it != end_it; ++start_it) erase(start_it); |
| 1228 | } |
| 1229 | |
| 1230 | // I/O |
| 1231 | // We support reading and writing groups to disk. We don't store |
| 1232 | // the actual array contents (which we don't know how to store), |
| 1233 | // just the bitmap and size. Meant to be used with table I/O. |
| 1234 | |
| 1235 | template <typename OUTPUT> |
| 1236 | bool write_metadata(OUTPUT* fp) const { |
| 1237 | // we explicitly set to uint16_t |
| 1238 | assert(sizeof(settings.num_buckets) == 2); |
| 1239 | if (!sparsehash_internal::write_bigendian_number(fp, settings.num_buckets, |
| 1240 | 2)) |
| 1241 | return false; |
| 1242 | if (!sparsehash_internal::write_data(fp, bitmap, sizeof(bitmap))) |
| 1243 | return false; |
| 1244 | return true; |
| 1245 | } |
| 1246 | |
| 1247 | // Reading destroys the old group contents! Returns true if all was ok. |
| 1248 | template <typename INPUT> |
| 1249 | bool read_metadata(INPUT* fp) { |
| 1250 | clear(); |
| 1251 | if (!sparsehash_internal::read_bigendian_number(fp, &settings.num_buckets, |
| 1252 | 2)) |
| 1253 | return false; |
| 1254 | if (!sparsehash_internal::read_data(fp, bitmap, sizeof(bitmap))) |
| 1255 | return false; |
| 1256 | // We'll allocate the space, but we won't fill it: it will be |
| 1257 | // left as uninitialized raw memory. |
| 1258 | group = allocate_group(settings.num_buckets); |
| 1259 | return true; |
| 1260 | } |
| 1261 | |
| 1262 | // Again, only meaningful if value_type is a POD. |
| 1263 | template <typename INPUT> |
| 1264 | bool read_nopointer_data(INPUT* fp) { |
| 1265 | for (nonempty_iterator it = nonempty_begin(); it != nonempty_end(); ++it) { |
| 1266 | if (!sparsehash_internal::read_data(fp, &(*it), sizeof(*it))) |
| 1267 | return false; |
| 1268 | } |
| 1269 | return true; |
| 1270 | } |
| 1271 | |
| 1272 | // If your keys and values are simple enough, we can write them |
| 1273 | // to disk for you. "simple enough" means POD and no pointers. |
| 1274 | // However, we don't try to normalize endianness. |
| 1275 | template <typename OUTPUT> |
| 1276 | bool write_nopointer_data(OUTPUT* fp) const { |
| 1277 | for (const_nonempty_iterator it = nonempty_begin(); it != nonempty_end(); |
| 1278 | ++it) { |
| 1279 | if (!sparsehash_internal::write_data(fp, &(*it), sizeof(*it))) |
| 1280 | return false; |
| 1281 | } |
| 1282 | return true; |
| 1283 | } |
| 1284 | |
| 1285 | // Comparisons. We only need to define == and < -- we get |
| 1286 | // != > <= >= via relops.h (which we happily included above). |
| 1287 | // Note the comparisons are pretty arbitrary: we compare |
| 1288 | // values of the first index that isn't equal (using default |
| 1289 | // value for empty buckets). |
| 1290 | bool operator==(const sparsegroup& x) const { |
| 1291 | return (settings.num_buckets == x.settings.num_buckets && |
| 1292 | memcmp(bitmap, x.bitmap, sizeof(bitmap)) == 0 && |
| 1293 | std::equal(begin(), end(), x.begin())); // from <algorithm> |
| 1294 | } |
| 1295 | |
| 1296 | bool operator<(const sparsegroup& x) const { // also from <algorithm> |
| 1297 | return std::lexicographical_compare(begin(), end(), x.begin(), x.end()); |
| 1298 | } |
| 1299 | bool operator!=(const sparsegroup& x) const { return !(*this == x); } |
| 1300 | bool operator<=(const sparsegroup& x) const { return !(x < *this); } |
| 1301 | bool operator>(const sparsegroup& x) const { return x < *this; } |
| 1302 | bool operator>=(const sparsegroup& x) const { return !(*this < x); } |
| 1303 | |
| 1304 | private: |
| 1305 | template <class A> |
| 1306 | class alloc_impl : public A { |
| 1307 | public: |
| 1308 | typedef typename A::pointer pointer; |
| 1309 | typedef typename A::size_type size_type; |
| 1310 | |
| 1311 | // Convert a normal allocator to one that has realloc_or_die() |
| 1312 | alloc_impl(const A& a) : A(a) {} |
| 1313 | |
| 1314 | // realloc_or_die should only be used when using the default |
| 1315 | // allocator (libc_allocator_with_realloc). |
| 1316 | pointer realloc_or_die(pointer /*ptr*/, size_type /*n*/) { |
| 1317 | fprintf(stderr, |
| 1318 | "realloc_or_die is only supported for " |
| 1319 | "libc_allocator_with_realloc\n"); |
| 1320 | exit(1); |
| 1321 | return NULL; |
| 1322 | } |
| 1323 | }; |
| 1324 | |
| 1325 | // A template specialization of alloc_impl for |
| 1326 | // libc_allocator_with_realloc that can handle realloc_or_die. |
| 1327 | template <class A> |
| 1328 | class alloc_impl<libc_allocator_with_realloc<A>> |
| 1329 | : public libc_allocator_with_realloc<A> { |
| 1330 | public: |
| 1331 | typedef typename libc_allocator_with_realloc<A>::pointer pointer; |
| 1332 | typedef typename libc_allocator_with_realloc<A>::size_type size_type; |
| 1333 | |
| 1334 | alloc_impl(const libc_allocator_with_realloc<A>& a) |
| 1335 | : libc_allocator_with_realloc<A>(a) {} |
| 1336 | |
| 1337 | pointer realloc_or_die(pointer ptr, size_type n) { |
| 1338 | pointer retval = this->reallocate(ptr, n); |
| 1339 | if (retval == NULL) { |
| 1340 | fprintf(stderr, |
| 1341 | "sparsehash: FATAL ERROR: failed to reallocate " |
| 1342 | "%lu elements for ptr %p", |
| 1343 | static_cast<unsigned long>(n), static_cast<void*>(ptr)); |
| 1344 | exit(1); |
| 1345 | } |
| 1346 | return retval; |
| 1347 | } |
| 1348 | }; |
| 1349 | |
| 1350 | // Package allocator with num_buckets to eliminate memory needed for the |
| 1351 | // zero-size allocator. |
| 1352 | // If new fields are added to this class, we should add them to |
| 1353 | // operator= and swap. |
| 1354 | class Settings : public alloc_impl<value_alloc_type> { |
| 1355 | public: |
| 1356 | Settings(const alloc_impl<value_alloc_type>& a, uint16_t n = 0) |
| 1357 | : alloc_impl<value_alloc_type>(a), num_buckets(n) {} |
| 1358 | Settings(const Settings& s) |
| 1359 | : alloc_impl<value_alloc_type>(s), num_buckets(s.num_buckets) {} |
| 1360 | |
| 1361 | uint16_t num_buckets; // limits GROUP_SIZE to 64K |
| 1362 | }; |
| 1363 | |
| 1364 | // The actual data |
| 1365 | pointer group; // (small) array of T's |
| 1366 | Settings settings; // allocator and num_buckets |
| 1367 | unsigned char |
| 1368 | bitmap[(GROUP_SIZE - 1) / 8 + 1]; // fancy math is so we round up |
| 1369 | }; |
| 1370 | |
| 1371 | // We need a global swap as well |
| 1372 | template <class T, uint16_t GROUP_SIZE, class Alloc> |
| 1373 | inline void swap(sparsegroup<T, GROUP_SIZE, Alloc>& x, |
| 1374 | sparsegroup<T, GROUP_SIZE, Alloc>& y) { |
| 1375 | x.swap(y); |
| 1376 | } |
| 1377 | |
| 1378 | // --------------------------------------------------------------------------- |
| 1379 | |
| 1380 | template <class T, uint16_t GROUP_SIZE = DEFAULT_SPARSEGROUP_SIZE, |
| 1381 | class Alloc = libc_allocator_with_realloc<T>> |
| 1382 | class sparsetable { |
| 1383 | private: |
| 1384 | using value_alloc_type = |
| 1385 | typename std::allocator_traits<Alloc>::template rebind_alloc<T>; |
| 1386 | using vector_alloc = |
| 1387 | typename std::allocator_traits<Alloc>::template rebind_alloc< |
| 1388 | sparsegroup<T, GROUP_SIZE, value_alloc_type>>; |
| 1389 | |
| 1390 | public: |
| 1391 | // Basic types |
| 1392 | typedef T value_type; // stolen from stl_vector.h |
| 1393 | typedef Alloc allocator_type; |
| 1394 | typedef typename value_alloc_type::size_type size_type; |
| 1395 | typedef typename value_alloc_type::difference_type difference_type; |
| 1396 | typedef typename value_alloc_type::reference reference; |
| 1397 | typedef typename value_alloc_type::const_reference const_reference; |
| 1398 | typedef typename value_alloc_type::pointer pointer; |
| 1399 | typedef typename value_alloc_type::const_pointer const_pointer; |
| 1400 | typedef table_iterator<sparsetable<T, GROUP_SIZE, Alloc>> iterator; |
| 1401 | typedef const_table_iterator<sparsetable<T, GROUP_SIZE, Alloc>> |
| 1402 | const_iterator; |
| 1403 | typedef table_element_adaptor<sparsetable<T, GROUP_SIZE, Alloc>> |
| 1404 | element_adaptor; |
| 1405 | typedef std::reverse_iterator<const_iterator> const_reverse_iterator; |
| 1406 | typedef std::reverse_iterator<iterator> reverse_iterator; // from iterator.h |
| 1407 | |
| 1408 | // These are our special iterators, that go over non-empty buckets in a |
| 1409 | // table. These aren't const only because you can change non-empty bcks. |
| 1410 | typedef two_d_iterator<std::vector< |
| 1411 | sparsegroup<value_type, GROUP_SIZE, value_alloc_type>, vector_alloc>> |
| 1412 | nonempty_iterator; |
| 1413 | typedef const_two_d_iterator<std::vector< |
| 1414 | sparsegroup<value_type, GROUP_SIZE, value_alloc_type>, vector_alloc>> |
| 1415 | const_nonempty_iterator; |
| 1416 | typedef std::reverse_iterator<nonempty_iterator> reverse_nonempty_iterator; |
| 1417 | typedef std::reverse_iterator<const_nonempty_iterator> |
| 1418 | const_reverse_nonempty_iterator; |
| 1419 | // Another special iterator: it frees memory as it iterates (used to resize) |
| 1420 | typedef destructive_two_d_iterator<std::vector< |
| 1421 | sparsegroup<value_type, GROUP_SIZE, value_alloc_type>, vector_alloc>> |
| 1422 | destructive_iterator; |
| 1423 | |
| 1424 | // Iterator functions |
| 1425 | iterator begin() { return iterator(this, 0); } |
| 1426 | const_iterator begin() const { return const_iterator(this, 0); } |
| 1427 | iterator end() { return iterator(this, size()); } |
| 1428 | const_iterator end() const { return const_iterator(this, size()); } |
| 1429 | reverse_iterator rbegin() { return reverse_iterator(end()); } |
| 1430 | const_reverse_iterator rbegin() const { |
| 1431 | return const_reverse_iterator(end()); |
| 1432 | } |
| 1433 | reverse_iterator rend() { return reverse_iterator(begin()); } |
| 1434 | const_reverse_iterator rend() const { |
| 1435 | return const_reverse_iterator(begin()); |
| 1436 | } |
| 1437 | |
| 1438 | // Versions for our special non-empty iterator |
| 1439 | nonempty_iterator nonempty_begin() { |
| 1440 | return nonempty_iterator(groups.begin(), groups.end(), groups.begin()); |
| 1441 | } |
| 1442 | const_nonempty_iterator nonempty_begin() const { |
| 1443 | return const_nonempty_iterator(groups.begin(), groups.end(), |
| 1444 | groups.begin()); |
| 1445 | } |
| 1446 | nonempty_iterator nonempty_end() { |
| 1447 | return nonempty_iterator(groups.begin(), groups.end(), groups.end()); |
| 1448 | } |
| 1449 | const_nonempty_iterator nonempty_end() const { |
| 1450 | return const_nonempty_iterator(groups.begin(), groups.end(), groups.end()); |
| 1451 | } |
| 1452 | reverse_nonempty_iterator nonempty_rbegin() { |
| 1453 | return reverse_nonempty_iterator(nonempty_end()); |
| 1454 | } |
| 1455 | const_reverse_nonempty_iterator nonempty_rbegin() const { |
| 1456 | return const_reverse_nonempty_iterator(nonempty_end()); |
| 1457 | } |
| 1458 | reverse_nonempty_iterator nonempty_rend() { |
| 1459 | return reverse_nonempty_iterator(nonempty_begin()); |
| 1460 | } |
| 1461 | const_reverse_nonempty_iterator nonempty_rend() const { |
| 1462 | return const_reverse_nonempty_iterator(nonempty_begin()); |
| 1463 | } |
| 1464 | destructive_iterator destructive_begin() { |
| 1465 | return destructive_iterator(groups.begin(), groups.end(), groups.begin()); |
| 1466 | } |
| 1467 | destructive_iterator destructive_end() { |
| 1468 | return destructive_iterator(groups.begin(), groups.end(), groups.end()); |
| 1469 | } |
| 1470 | |
| 1471 | typedef sparsegroup<value_type, GROUP_SIZE, allocator_type> group_type; |
| 1472 | using group_vector_type_allocator_type = |
| 1473 | typename std::allocator_traits<Alloc>::template rebind_alloc<group_type>; |
| 1474 | typedef std::vector<group_type, group_vector_type_allocator_type> |
| 1475 | group_vector_type; |
| 1476 | |
| 1477 | typedef typename group_vector_type::reference GroupsReference; |
| 1478 | typedef typename group_vector_type::const_reference GroupsConstReference; |
| 1479 | typedef typename group_vector_type::iterator GroupsIterator; |
| 1480 | typedef typename group_vector_type::const_iterator GroupsConstIterator; |
| 1481 | |
| 1482 | // How to deal with the proper group |
| 1483 | static size_type num_groups(size_type num) { // how many to hold num buckets |
| 1484 | return num == 0 ? 0 : ((num - 1) / GROUP_SIZE) + 1; |
| 1485 | } |
| 1486 | |
| 1487 | uint16_t pos_in_group(size_type i) const { |
| 1488 | return static_cast<uint16_t>(i % GROUP_SIZE); |
| 1489 | } |
| 1490 | size_type group_num(size_type i) const { return i / GROUP_SIZE; } |
| 1491 | GroupsReference which_group(size_type i) { return groups[group_num(i)]; } |
| 1492 | GroupsConstReference which_group(size_type i) const { |
| 1493 | return groups[group_num(i)]; |
| 1494 | } |
| 1495 | |
| 1496 | public: |
| 1497 | // Constructors -- default, normal (when you specify size), and copy |
| 1498 | explicit sparsetable(size_type sz = 0, Alloc alloc = Alloc()) |
| 1499 | : groups(vector_alloc(alloc)), settings(alloc, sz) { |
| 1500 | groups.resize(num_groups(sz), group_type(settings)); |
| 1501 | } |
| 1502 | // We can get away with using the default copy constructor, |
| 1503 | // and default destructor, and hence the default operator=. Huzzah! |
| 1504 | |
| 1505 | // Many STL algorithms use swap instead of copy constructors |
| 1506 | void swap(sparsetable& x) { |
| 1507 | std::swap(groups, x.groups); // defined in stl_algobase.h |
| 1508 | std::swap(settings.table_size, x.settings.table_size); |
| 1509 | std::swap(settings.num_buckets, x.settings.num_buckets); |
| 1510 | } |
| 1511 | |
| 1512 | // It's always nice to be able to clear a table without deallocating it |
| 1513 | void clear() { |
| 1514 | GroupsIterator group; |
| 1515 | for (group = groups.begin(); group != groups.end(); ++group) { |
| 1516 | group->clear(); |
| 1517 | } |
| 1518 | settings.num_buckets = 0; |
| 1519 | } |
| 1520 | |
| 1521 | // ACCESSOR FUNCTIONS for the things we templatize on, basically |
| 1522 | allocator_type get_allocator() const { return allocator_type(settings); } |
| 1523 | |
| 1524 | // Functions that tell you about size. |
| 1525 | // NOTE: empty() is non-intuitive! It does not tell you the number |
| 1526 | // of not-empty buckets (use num_nonempty() for that). Instead |
| 1527 | // it says whether you've allocated any buckets or not. |
| 1528 | size_type size() const { return settings.table_size; } |
| 1529 | size_type max_size() const { return settings.max_size(); } |
| 1530 | bool empty() const { return settings.table_size == 0; } |
| 1531 | // We also may want to know how many *used* buckets there are |
| 1532 | size_type num_nonempty() const { return settings.num_buckets; } |
| 1533 | |
| 1534 | // OK, we'll let you resize one of these puppies |
| 1535 | void resize(size_type new_size) { |
| 1536 | groups.resize(num_groups(new_size), group_type(settings)); |
| 1537 | if (new_size < settings.table_size) { |
| 1538 | // lower num_buckets, clear last group |
| 1539 | if (pos_in_group(new_size) > 0) // need to clear inside last group |
| 1540 | groups.back().erase(groups.back().begin() + pos_in_group(new_size), |
| 1541 | groups.back().end()); |
| 1542 | settings.num_buckets = 0; // refigure # of used buckets |
| 1543 | GroupsConstIterator group; |
| 1544 | for (group = groups.begin(); group != groups.end(); ++group) |
| 1545 | settings.num_buckets += group->num_nonempty(); |
| 1546 | } |
| 1547 | settings.table_size = new_size; |
| 1548 | } |
| 1549 | |
| 1550 | // We let you see if a bucket is non-empty without retrieving it |
| 1551 | bool test(size_type i) const { |
| 1552 | assert(i < settings.table_size); |
| 1553 | return which_group(i).test(pos_in_group(i)); |
| 1554 | } |
| 1555 | bool test(iterator pos) const { |
| 1556 | return which_group(pos.pos).test(pos_in_group(pos.pos)); |
| 1557 | } |
| 1558 | bool test(const_iterator pos) const { |
| 1559 | return which_group(pos.pos).test(pos_in_group(pos.pos)); |
| 1560 | } |
| 1561 | |
| 1562 | // We only return const_references because it's really hard to |
| 1563 | // return something settable for empty buckets. Use set() instead. |
| 1564 | const_reference get(size_type i) const { |
| 1565 | assert(i < settings.table_size); |
| 1566 | return which_group(i).get(pos_in_group(i)); |
| 1567 | } |
| 1568 | |
| 1569 | // TODO(csilvers): make protected + friend |
| 1570 | // This is used by sparse_hashtable to get an element from the table |
| 1571 | // when we know it exists (because the caller has called test(i)). |
| 1572 | const_reference unsafe_get(size_type i) const { |
| 1573 | assert(i < settings.table_size); |
| 1574 | assert(test(i)); |
| 1575 | return which_group(i).unsafe_get(pos_in_group(i)); |
| 1576 | } |
| 1577 | |
| 1578 | // TODO(csilvers): make protected + friend element_adaptor |
| 1579 | reference mutating_get(size_type i) { // fills bucket i before getting |
| 1580 | assert(i < settings.table_size); |
| 1581 | typename group_type::size_type old_numbuckets = |
| 1582 | which_group(i).num_nonempty(); |
| 1583 | reference retval = which_group(i).mutating_get(pos_in_group(i)); |
| 1584 | settings.num_buckets += which_group(i).num_nonempty() - old_numbuckets; |
| 1585 | return retval; |
| 1586 | } |
| 1587 | |
| 1588 | // Syntactic sugar. As in sparsegroup, the non-const version is harder |
| 1589 | const_reference operator[](size_type i) const { return get(i); } |
| 1590 | |
| 1591 | element_adaptor operator[](size_type i) { return element_adaptor(this, i); } |
| 1592 | |
| 1593 | // Needed for hashtables, gets as a nonempty_iterator. Crashes for empty |
| 1594 | // bcks |
| 1595 | const_nonempty_iterator get_iter(size_type i) const { |
| 1596 | assert(test(i)); // how can a nonempty_iterator point to an empty bucket? |
| 1597 | return const_nonempty_iterator( |
| 1598 | groups.begin(), groups.end(), groups.begin() + group_num(i), |
| 1599 | (groups[group_num(i)].nonempty_begin() + |
| 1600 | groups[group_num(i)].pos_to_offset(pos_in_group(i)))); |
| 1601 | } |
| 1602 | // For nonempty we can return a non-const version |
| 1603 | nonempty_iterator get_iter(size_type i) { |
| 1604 | assert(test(i)); // how can a nonempty_iterator point to an empty bucket? |
| 1605 | return nonempty_iterator( |
| 1606 | groups.begin(), groups.end(), groups.begin() + group_num(i), |
| 1607 | (groups[group_num(i)].nonempty_begin() + |
| 1608 | groups[group_num(i)].pos_to_offset(pos_in_group(i)))); |
| 1609 | } |
| 1610 | |
| 1611 | // And the reverse transformation. |
| 1612 | size_type get_pos(const const_nonempty_iterator& it) const { |
| 1613 | difference_type current_row = it.row_current - it.row_begin; |
| 1614 | difference_type current_col = |
| 1615 | (it.col_current - groups[current_row].nonempty_begin()); |
| 1616 | return ((current_row * GROUP_SIZE) + |
| 1617 | groups[current_row].offset_to_pos(current_col)); |
| 1618 | } |
| 1619 | |
| 1620 | // This returns a reference to the inserted item (which is a copy of val) |
| 1621 | // The trick is to figure out whether we're replacing or inserting anew |
| 1622 | reference set(size_type i, const_reference val) { |
| 1623 | assert(i < settings.table_size); |
| 1624 | typename group_type::size_type old_numbuckets = |
| 1625 | which_group(i).num_nonempty(); |
| 1626 | reference retval = which_group(i).set(pos_in_group(i), val); |
| 1627 | settings.num_buckets += which_group(i).num_nonempty() - old_numbuckets; |
| 1628 | return retval; |
| 1629 | } |
| 1630 | |
| 1631 | // This takes the specified elements out of the table. This is |
| 1632 | // "undefining", rather than "clearing". |
| 1633 | void erase(size_type i) { |
| 1634 | assert(i < settings.table_size); |
| 1635 | typename group_type::size_type old_numbuckets = |
| 1636 | which_group(i).num_nonempty(); |
| 1637 | which_group(i).erase(pos_in_group(i)); |
| 1638 | settings.num_buckets += which_group(i).num_nonempty() - old_numbuckets; |
| 1639 | } |
| 1640 | |
| 1641 | void erase(iterator pos) { erase(pos.pos); } |
| 1642 | |
| 1643 | void erase(iterator start_it, iterator end_it) { |
| 1644 | // This could be more efficient, but then we'd need to figure |
| 1645 | // out if we spanned groups or not. Doesn't seem worth it. |
| 1646 | for (; start_it != end_it; ++start_it) erase(start_it); |
| 1647 | } |
| 1648 | |
| 1649 | // We support reading and writing tables to disk. We don't store |
| 1650 | // the actual array contents (which we don't know how to store), |
| 1651 | // just the groups and sizes. Returns true if all went ok. |
| 1652 | |
| 1653 | private: |
| 1654 | // Every time the disk format changes, this should probably change too |
| 1655 | typedef unsigned long MagicNumberType; |
| 1656 | static const MagicNumberType MAGIC_NUMBER = 0x24687531; |
| 1657 | |
| 1658 | // Old versions of this code write all data in 32 bits. We need to |
| 1659 | // support these files as well as having support for 64-bit systems. |
| 1660 | // So we use the following encoding scheme: for values < 2^32-1, we |
| 1661 | // store in 4 bytes in big-endian order. For values > 2^32, we |
| 1662 | // store 0xFFFFFFF followed by 8 bytes in big-endian order. This |
| 1663 | // causes us to mis-read old-version code that stores exactly |
| 1664 | // 0xFFFFFFF, but I don't think that is likely to have happened for |
| 1665 | // these particular values. |
| 1666 | template <typename OUTPUT, typename IntType> |
| 1667 | static bool write_32_or_64(OUTPUT* fp, IntType value) { |
| 1668 | if (value < 0xFFFFFFFFULL) { // fits in 4 bytes |
| 1669 | if (!sparsehash_internal::write_bigendian_number(fp, value, 4)) |
| 1670 | return false; |
| 1671 | } else { |
| 1672 | if (!sparsehash_internal::write_bigendian_number(fp, 0xFFFFFFFFUL, 4)) |
| 1673 | return false; |
| 1674 | if (!sparsehash_internal::write_bigendian_number(fp, value, 8)) |
| 1675 | return false; |
| 1676 | } |
| 1677 | return true; |
| 1678 | } |
| 1679 | |
| 1680 | template <typename INPUT, typename IntType> |
| 1681 | static bool read_32_or_64(INPUT* fp, IntType* value) { // reads into value |
| 1682 | MagicNumberType first4 = 0; // a convenient 32-bit unsigned type |
| 1683 | if (!sparsehash_internal::read_bigendian_number(fp, &first4, 4)) |
| 1684 | return false; |
| 1685 | if (first4 < 0xFFFFFFFFULL) { |
| 1686 | *value = first4; |
| 1687 | } else { |
| 1688 | if (!sparsehash_internal::read_bigendian_number(fp, value, 8)) |
| 1689 | return false; |
| 1690 | } |
| 1691 | return true; |
| 1692 | } |
| 1693 | |
| 1694 | public: |
| 1695 | // read/write_metadata() and read_write/nopointer_data() are DEPRECATED. |
| 1696 | // Use serialize() and unserialize(), below, for new code. |
| 1697 | |
| 1698 | template <typename OUTPUT> |
| 1699 | bool write_metadata(OUTPUT* fp) const { |
| 1700 | if (!write_32_or_64(fp, MAGIC_NUMBER)) return false; |
| 1701 | if (!write_32_or_64(fp, settings.table_size)) return false; |
| 1702 | if (!write_32_or_64(fp, settings.num_buckets)) return false; |
| 1703 | |
| 1704 | GroupsConstIterator group; |
| 1705 | for (group = groups.begin(); group != groups.end(); ++group) |
| 1706 | if (group->write_metadata(fp) == false) return false; |
| 1707 | return true; |
| 1708 | } |
| 1709 | |
| 1710 | // Reading destroys the old table contents! Returns true if read ok. |
| 1711 | template <typename INPUT> |
| 1712 | bool read_metadata(INPUT* fp) { |
| 1713 | size_type magic_read = 0; |
| 1714 | if (!read_32_or_64(fp, &magic_read)) return false; |
| 1715 | if (magic_read != MAGIC_NUMBER) { |
| 1716 | clear(); // just to be consistent |
| 1717 | return false; |
| 1718 | } |
| 1719 | |
| 1720 | if (!read_32_or_64(fp, &settings.table_size)) return false; |
| 1721 | if (!read_32_or_64(fp, &settings.num_buckets)) return false; |
| 1722 | |
| 1723 | resize(settings.table_size); // so the vector's sized ok |
| 1724 | GroupsIterator group; |
| 1725 | for (group = groups.begin(); group != groups.end(); ++group) |
| 1726 | if (group->read_metadata(fp) == false) return false; |
| 1727 | return true; |
| 1728 | } |
| 1729 | |
| 1730 | // This code is identical to that for SparseGroup |
| 1731 | // If your keys and values are simple enough, we can write them |
| 1732 | // to disk for you. "simple enough" means no pointers. |
| 1733 | // However, we don't try to normalize endianness |
| 1734 | bool write_nopointer_data(FILE* fp) const { |
| 1735 | for (const_nonempty_iterator it = nonempty_begin(); it != nonempty_end(); |
| 1736 | ++it) { |
| 1737 | if (!fwrite(&*it, sizeof(*it), 1, fp)) return false; |
| 1738 | } |
| 1739 | return true; |
| 1740 | } |
| 1741 | |
| 1742 | // When reading, we have to override the potential const-ness of *it |
| 1743 | bool read_nopointer_data(FILE* fp) { |
| 1744 | for (nonempty_iterator it = nonempty_begin(); it != nonempty_end(); ++it) { |
| 1745 | if (!fread(reinterpret_cast<void*>(&(*it)), sizeof(*it), 1, fp)) |
| 1746 | return false; |
| 1747 | } |
| 1748 | return true; |
| 1749 | } |
| 1750 | |
| 1751 | // INPUT and OUTPUT must be either a FILE, *or* a C++ stream |
| 1752 | // (istream, ostream, etc) *or* a class providing |
| 1753 | // Read(void*, size_t) and Write(const void*, size_t) |
| 1754 | // (respectively), which writes a buffer into a stream |
| 1755 | // (which the INPUT/OUTPUT instance presumably owns). |
| 1756 | |
| 1757 | typedef sparsehash_internal::pod_serializer<value_type> NopointerSerializer; |
| 1758 | |
| 1759 | // ValueSerializer: a functor. operator()(OUTPUT*, const value_type&) |
| 1760 | template <typename ValueSerializer, typename OUTPUT> |
| 1761 | bool serialize(ValueSerializer serializer, OUTPUT* fp) { |
| 1762 | if (!write_metadata(fp)) return false; |
| 1763 | for (const_nonempty_iterator it = nonempty_begin(); it != nonempty_end(); |
| 1764 | ++it) { |
| 1765 | if (!serializer(fp, *it)) return false; |
| 1766 | } |
| 1767 | return true; |
| 1768 | } |
| 1769 | |
| 1770 | // ValueSerializer: a functor. operator()(INPUT*, value_type*) |
| 1771 | template <typename ValueSerializer, typename INPUT> |
| 1772 | bool unserialize(ValueSerializer serializer, INPUT* fp) { |
| 1773 | clear(); |
| 1774 | if (!read_metadata(fp)) return false; |
| 1775 | for (nonempty_iterator it = nonempty_begin(); it != nonempty_end(); ++it) { |
| 1776 | if (!serializer(fp, &*it)) return false; |
| 1777 | } |
| 1778 | return true; |
| 1779 | } |
| 1780 | |
| 1781 | // Comparisons. Note the comparisons are pretty arbitrary: we |
| 1782 | // compare values of the first index that isn't equal (using default |
| 1783 | // value for empty buckets). |
| 1784 | bool operator==(const sparsetable& x) const { |
| 1785 | return (settings.table_size == x.settings.table_size && |
| 1786 | settings.num_buckets == x.settings.num_buckets && |
| 1787 | groups == x.groups); |
| 1788 | } |
| 1789 | |
| 1790 | bool operator<(const sparsetable& x) const { |
| 1791 | return std::lexicographical_compare(begin(), end(), x.begin(), x.end()); |
| 1792 | } |
| 1793 | bool operator!=(const sparsetable& x) const { return !(*this == x); } |
| 1794 | bool operator<=(const sparsetable& x) const { return !(x < *this); } |
| 1795 | bool operator>(const sparsetable& x) const { return x < *this; } |
| 1796 | bool operator>=(const sparsetable& x) const { return !(*this < x); } |
| 1797 | |
| 1798 | private: |
| 1799 | // Package allocator with table_size and num_buckets to eliminate memory |
| 1800 | // needed for the zero-size allocator. |
| 1801 | // If new fields are added to this class, we should add them to |
| 1802 | // operator= and swap. |
| 1803 | class Settings : public allocator_type { |
| 1804 | public: |
| 1805 | typedef typename allocator_type::size_type size_type; |
| 1806 | |
| 1807 | Settings(const allocator_type& a, size_type sz = 0, size_type n = 0) |
| 1808 | : allocator_type(a), table_size(sz), num_buckets(n) {} |
| 1809 | |
| 1810 | Settings(const Settings& s) |
| 1811 | : allocator_type(s), |
| 1812 | table_size(s.table_size), |
| 1813 | num_buckets(s.num_buckets) {} |
| 1814 | |
| 1815 | size_type table_size; // how many buckets they want |
| 1816 | size_type num_buckets; // number of non-empty buckets |
| 1817 | }; |
| 1818 | |
| 1819 | // The actual data |
| 1820 | group_vector_type groups; // our list of groups |
| 1821 | Settings settings; // allocator, table size, buckets |
| 1822 | }; |
| 1823 | |
| 1824 | // We need a global swap as well |
| 1825 | template <class T, uint16_t GROUP_SIZE, class Alloc> |
| 1826 | inline void swap(sparsetable<T, GROUP_SIZE, Alloc>& x, |
| 1827 | sparsetable<T, GROUP_SIZE, Alloc>& y) { |
| 1828 | x.swap(y); |
| 1829 | } |
| 1830 | } // namespace google |
| 1831 |
Generated while processing ClickHouse/dbms/src/Common/tests/integer_hash_tables_and_hashes.cpp
Generated on 2020-Jan-04 from project ClickHouse revision 19.17.6
Powered by 
Code Browser 2.1
Generator usage only permitted with license.