| 1 | // Details for a templated general-case hybrid-radix string_sort. |
| 2 | |
| 3 | // Copyright Steven J. Ross 2001 - 2014. |
| 4 | // Distributed under the Boost Software License, Version 1.0. |
| 5 | // (See accompanying file LICENSE_1_0.txt or copy at |
| 6 | // http://www.boost.org/LICENSE_1_0.txt) |
| 7 | |
| 8 | // See http://www.boost.org/libs/sort for library home page. |
| 9 | |
| 10 | /* |
| 11 | Some improvements suggested by: |
| 12 | Phil Endecott and Frank Gennari |
| 13 | */ |
| 14 | |
| 15 | #ifndef BOOST_SORT_SPREADSORT_DETAIL_SPREAD_SORT_HPP |
| 16 | #define BOOST_SORT_SPREADSORT_DETAIL_SPREAD_SORT_HPP |
| 17 | #include <algorithm> |
| 18 | #include <vector> |
| 19 | #include <cstring> |
| 20 | #include <limits> |
| 21 | #include <functional> |
| 22 | #include <boost/static_assert.hpp> |
| 23 | #include <boost/utility/enable_if.hpp> |
| 24 | #include <boost/sort/spreadsort/detail/constants.hpp> |
| 25 | #include <boost/sort/spreadsort/detail/spreadsort_common.hpp> |
| 26 | #include <boost/cstdint.hpp> |
| 27 | |
| 28 | namespace boost { |
| 29 | namespace sort { |
| 30 | namespace spreadsort { |
| 31 | namespace detail { |
| 32 | static const int max_step_size = 64; |
| 33 | |
| 34 | //Offsetting on identical characters. This function works a chunk of |
| 35 | //characters at a time for cache efficiency and optimal worst-case |
| 36 | //performance. |
| 37 | template<class RandomAccessIter, class Unsigned_char_type> |
| 38 | inline void |
| 39 | update_offset(RandomAccessIter first, RandomAccessIter finish, |
| 40 | size_t &char_offset) |
| 41 | { |
| 42 | const int char_size = sizeof(Unsigned_char_type); |
| 43 | size_t nextOffset = char_offset; |
| 44 | int step_size = max_step_size / char_size; |
| 45 | while (true) { |
| 46 | RandomAccessIter curr = first; |
| 47 | do { |
| 48 | //Ignore empties, but if the nextOffset would exceed the length or |
| 49 | //not match, exit; we've found the last matching character |
| 50 | //This will reduce the step_size if the current step doesn't match. |
| 51 | if ((*curr).size() > char_offset) { |
| 52 | if((*curr).size() <= (nextOffset + step_size)) { |
| 53 | step_size = (*curr).size() - nextOffset - 1; |
| 54 | if (step_size < 1) { |
| 55 | char_offset = nextOffset; |
| 56 | return; |
| 57 | } |
| 58 | } |
| 59 | const int step_byte_size = step_size * char_size; |
| 60 | if (memcmp(curr->data() + nextOffset, first->data() + nextOffset, |
| 61 | step_byte_size) != 0) { |
| 62 | if (step_size == 1) { |
| 63 | char_offset = nextOffset; |
| 64 | return; |
| 65 | } |
| 66 | step_size = (step_size > 4) ? 4 : 1; |
| 67 | continue; |
| 68 | } |
| 69 | } |
| 70 | ++curr; |
| 71 | } while (curr != finish); |
| 72 | nextOffset += step_size; |
| 73 | } |
| 74 | } |
| 75 | |
| 76 | //Offsetting on identical characters. This function works a character |
| 77 | //at a time for optimal worst-case performance. |
| 78 | template<class RandomAccessIter, class Get_char, class Get_length> |
| 79 | inline void |
| 80 | update_offset(RandomAccessIter first, RandomAccessIter finish, |
| 81 | size_t &char_offset, Get_char get_character, Get_length length) |
| 82 | { |
| 83 | size_t nextOffset = char_offset; |
| 84 | while (true) { |
| 85 | RandomAccessIter curr = first; |
| 86 | do { |
| 87 | //ignore empties, but if the nextOffset would exceed the length or |
| 88 | //not match, exit; we've found the last matching character |
| 89 | if (length(*curr) > char_offset && (length(*curr) <= (nextOffset + 1) |
| 90 | || get_character((*curr), nextOffset) != get_character((*first), nextOffset))) { |
| 91 | char_offset = nextOffset; |
| 92 | return; |
| 93 | } |
| 94 | } while (++curr != finish); |
| 95 | ++nextOffset; |
| 96 | } |
| 97 | } |
| 98 | |
| 99 | //This comparison functor assumes strings are identical up to char_offset |
| 100 | template<class Data_type, class Unsigned_char_type> |
| 101 | struct offset_less_than { |
| 102 | offset_less_than(size_t char_offset) : fchar_offset(char_offset){} |
| 103 | inline bool operator()(const Data_type &x, const Data_type &y) const |
| 104 | { |
| 105 | size_t minSize = (std::min)(x.size(), y.size()); |
| 106 | for (size_t u = fchar_offset; u < minSize; ++u) { |
| 107 | BOOST_STATIC_ASSERT(sizeof(x[u]) == sizeof(Unsigned_char_type)); |
| 108 | if (static_cast<Unsigned_char_type>(x[u]) != |
| 109 | static_cast<Unsigned_char_type>(y[u])) { |
| 110 | return static_cast<Unsigned_char_type>(x[u]) < |
| 111 | static_cast<Unsigned_char_type>(y[u]); |
| 112 | } |
| 113 | } |
| 114 | return x.size() < y.size(); |
| 115 | } |
| 116 | size_t fchar_offset; |
| 117 | }; |
| 118 | |
| 119 | //Compares strings assuming they are identical up to char_offset |
| 120 | template<class Data_type, class Unsigned_char_type> |
| 121 | struct offset_greater_than { |
| 122 | offset_greater_than(size_t char_offset) : fchar_offset(char_offset){} |
| 123 | inline bool operator()(const Data_type &x, const Data_type &y) const |
| 124 | { |
| 125 | size_t minSize = (std::min)(x.size(), y.size()); |
| 126 | for (size_t u = fchar_offset; u < minSize; ++u) { |
| 127 | BOOST_STATIC_ASSERT(sizeof(x[u]) == sizeof(Unsigned_char_type)); |
| 128 | if (static_cast<Unsigned_char_type>(x[u]) != |
| 129 | static_cast<Unsigned_char_type>(y[u])) { |
| 130 | return static_cast<Unsigned_char_type>(x[u]) > |
| 131 | static_cast<Unsigned_char_type>(y[u]); |
| 132 | } |
| 133 | } |
| 134 | return x.size() > y.size(); |
| 135 | } |
| 136 | size_t fchar_offset; |
| 137 | }; |
| 138 | |
| 139 | //This comparison functor assumes strings are identical up to char_offset |
| 140 | template<class Data_type, class Get_char, class Get_length> |
| 141 | struct offset_char_less_than { |
| 142 | offset_char_less_than(size_t char_offset) : fchar_offset(char_offset){} |
| 143 | inline bool operator()(const Data_type &x, const Data_type &y) const |
| 144 | { |
| 145 | size_t minSize = (std::min)(length(x), length(y)); |
| 146 | for (size_t u = fchar_offset; u < minSize; ++u) { |
| 147 | if (get_character(x, u) != get_character(y, u)) { |
| 148 | return get_character(x, u) < get_character(y, u); |
| 149 | } |
| 150 | } |
| 151 | return length(x) < length(y); |
| 152 | } |
| 153 | size_t fchar_offset; |
| 154 | Get_char get_character; |
| 155 | Get_length length; |
| 156 | }; |
| 157 | |
| 158 | //String sorting recursive implementation |
| 159 | template <class RandomAccessIter, class Unsigned_char_type> |
| 160 | inline void |
| 161 | string_sort_rec(RandomAccessIter first, RandomAccessIter last, |
| 162 | size_t char_offset, |
| 163 | std::vector<RandomAccessIter> &bin_cache, |
| 164 | unsigned cache_offset, size_t *bin_sizes) |
| 165 | { |
| 166 | typedef typename std::iterator_traits<RandomAccessIter>::value_type |
| 167 | Data_type; |
| 168 | //This section makes handling of long identical substrings much faster |
| 169 | //with a mild average performance impact. |
| 170 | //Iterate to the end of the empties. If all empty, return |
| 171 | while ((*first).size() <= char_offset) { |
| 172 | if (++first == last) |
| 173 | return; |
| 174 | } |
| 175 | RandomAccessIter finish = last - 1; |
| 176 | //Getting the last non-empty |
| 177 | for (;(*finish).size() <= char_offset; --finish); |
| 178 | ++finish; |
| 179 | //Offsetting on identical characters. This section works |
| 180 | //a few characters at a time for optimal worst-case performance. |
| 181 | update_offset<RandomAccessIter, Unsigned_char_type>(first, finish, |
| 182 | char_offset); |
| 183 | |
| 184 | const unsigned bin_count = (1 << (sizeof(Unsigned_char_type)*8)); |
| 185 | //Equal worst-case of radix and comparison is when bin_count = n*log(n). |
| 186 | const unsigned max_size = bin_count; |
| 187 | const unsigned membin_count = bin_count + 1; |
| 188 | unsigned cache_end; |
| 189 | RandomAccessIter * bins = size_bins(bin_sizes, bin_cache, cache_offset, |
| 190 | cache_end, membin_count) + 1; |
| 191 | |
| 192 | //Calculating the size of each bin; this takes roughly 10% of runtime |
| 193 | for (RandomAccessIter current = first; current != last; ++current) { |
| 194 | if ((*current).size() <= char_offset) { |
| 195 | bin_sizes[0]++; |
| 196 | } |
| 197 | else |
| 198 | bin_sizes[static_cast<Unsigned_char_type>((*current)[char_offset]) |
| 199 | + 1]++; |
| 200 | } |
| 201 | //Assign the bin positions |
| 202 | bin_cache[cache_offset] = first; |
| 203 | for (unsigned u = 0; u < membin_count - 1; u++) |
| 204 | bin_cache[cache_offset + u + 1] = |
| 205 | bin_cache[cache_offset + u] + bin_sizes[u]; |
| 206 | |
| 207 | //Swap into place |
| 208 | RandomAccessIter next_bin_start = first; |
| 209 | //handling empty bins |
| 210 | RandomAccessIter * local_bin = &(bin_cache[cache_offset]); |
| 211 | next_bin_start += bin_sizes[0]; |
| 212 | RandomAccessIter * target_bin; |
| 213 | //Iterating over each element in the bin of empties |
| 214 | for (RandomAccessIter current = *local_bin; current < next_bin_start; |
| 215 | ++current) { |
| 216 | //empties belong in this bin |
| 217 | while ((*current).size() > char_offset) { |
| 218 | target_bin = |
| 219 | bins + static_cast<Unsigned_char_type>((*current)[char_offset]); |
| 220 | iter_swap(current, (*target_bin)++); |
| 221 | } |
| 222 | } |
| 223 | *local_bin = next_bin_start; |
| 224 | //iterate backwards to find the last bin with elements in it |
| 225 | //this saves iterations in multiple loops |
| 226 | unsigned last_bin = bin_count - 1; |
| 227 | for (; last_bin && !bin_sizes[last_bin + 1]; --last_bin); |
| 228 | //This dominates runtime, mostly in the swap and bin lookups |
| 229 | for (unsigned u = 0; u < last_bin; ++u) { |
| 230 | local_bin = bins + u; |
| 231 | next_bin_start += bin_sizes[u + 1]; |
| 232 | //Iterating over each element in this bin |
| 233 | for (RandomAccessIter current = *local_bin; current < next_bin_start; |
| 234 | ++current) { |
| 235 | //Swapping into place until the correct element has been swapped in |
| 236 | for (target_bin = bins + static_cast<Unsigned_char_type> |
| 237 | ((*current)[char_offset]); target_bin != local_bin; |
| 238 | target_bin = bins + static_cast<Unsigned_char_type> |
| 239 | ((*current)[char_offset])) iter_swap(current, (*target_bin)++); |
| 240 | } |
| 241 | *local_bin = next_bin_start; |
| 242 | } |
| 243 | bins[last_bin] = last; |
| 244 | //Recursing |
| 245 | RandomAccessIter lastPos = bin_cache[cache_offset]; |
| 246 | //Skip this loop for empties |
| 247 | for (unsigned u = cache_offset + 1; u < cache_offset + last_bin + 2; |
| 248 | lastPos = bin_cache[u], ++u) { |
| 249 | size_t count = bin_cache[u] - lastPos; |
| 250 | //don't sort unless there are at least two items to Compare |
| 251 | if (count < 2) |
| 252 | continue; |
| 253 | //using boost::sort::pdqsort if its worst-case is better |
| 254 | if (count < max_size) |
| 255 | boost::sort::pdqsort(lastPos, bin_cache[u], |
| 256 | offset_less_than<Data_type, Unsigned_char_type>(char_offset + 1)); |
| 257 | else |
| 258 | string_sort_rec<RandomAccessIter, Unsigned_char_type>(lastPos, |
| 259 | bin_cache[u], char_offset + 1, bin_cache, cache_end, bin_sizes); |
| 260 | } |
| 261 | } |
| 262 | |
| 263 | //Sorts strings in reverse order, with empties at the end |
| 264 | template <class RandomAccessIter, class Unsigned_char_type> |
| 265 | inline void |
| 266 | reverse_string_sort_rec(RandomAccessIter first, RandomAccessIter last, |
| 267 | size_t char_offset, |
| 268 | std::vector<RandomAccessIter> &bin_cache, |
| 269 | unsigned cache_offset, |
| 270 | size_t *bin_sizes) |
| 271 | { |
| 272 | typedef typename std::iterator_traits<RandomAccessIter>::value_type |
| 273 | Data_type; |
| 274 | //This section makes handling of long identical substrings much faster |
| 275 | //with a mild average performance impact. |
| 276 | RandomAccessIter curr = first; |
| 277 | //Iterate to the end of the empties. If all empty, return |
| 278 | while ((*curr).size() <= char_offset) { |
| 279 | if (++curr == last) |
| 280 | return; |
| 281 | } |
| 282 | //Getting the last non-empty |
| 283 | while ((*(--last)).size() <= char_offset); |
| 284 | ++last; |
| 285 | //Offsetting on identical characters. This section works |
| 286 | //a few characters at a time for optimal worst-case performance. |
| 287 | update_offset<RandomAccessIter, Unsigned_char_type>(curr, last, |
| 288 | char_offset); |
| 289 | RandomAccessIter * target_bin; |
| 290 | |
| 291 | const unsigned bin_count = (1 << (sizeof(Unsigned_char_type)*8)); |
| 292 | //Equal worst-case of radix and comparison when bin_count = n*log(n). |
| 293 | const unsigned max_size = bin_count; |
| 294 | const unsigned membin_count = bin_count + 1; |
| 295 | const unsigned max_bin = bin_count - 1; |
| 296 | unsigned cache_end; |
| 297 | RandomAccessIter * bins = size_bins(bin_sizes, bin_cache, cache_offset, |
| 298 | cache_end, membin_count); |
| 299 | RandomAccessIter * end_bin = &(bin_cache[cache_offset + max_bin]); |
| 300 | |
| 301 | //Calculating the size of each bin; this takes roughly 10% of runtime |
| 302 | for (RandomAccessIter current = first; current != last; ++current) { |
| 303 | if ((*current).size() <= char_offset) { |
| 304 | bin_sizes[bin_count]++; |
| 305 | } |
| 306 | else |
| 307 | bin_sizes[max_bin - static_cast<Unsigned_char_type> |
| 308 | ((*current)[char_offset])]++; |
| 309 | } |
| 310 | //Assign the bin positions |
| 311 | bin_cache[cache_offset] = first; |
| 312 | for (unsigned u = 0; u < membin_count - 1; u++) |
| 313 | bin_cache[cache_offset + u + 1] = |
| 314 | bin_cache[cache_offset + u] + bin_sizes[u]; |
| 315 | |
| 316 | //Swap into place |
| 317 | RandomAccessIter next_bin_start = last; |
| 318 | //handling empty bins |
| 319 | RandomAccessIter * local_bin = &(bin_cache[cache_offset + bin_count]); |
| 320 | RandomAccessIter lastFull = *local_bin; |
| 321 | //Iterating over each element in the bin of empties |
| 322 | for (RandomAccessIter current = *local_bin; current < next_bin_start; |
| 323 | ++current) { |
| 324 | //empties belong in this bin |
| 325 | while ((*current).size() > char_offset) { |
| 326 | target_bin = |
| 327 | end_bin - static_cast<Unsigned_char_type>((*current)[char_offset]); |
| 328 | iter_swap(current, (*target_bin)++); |
| 329 | } |
| 330 | } |
| 331 | *local_bin = next_bin_start; |
| 332 | next_bin_start = first; |
| 333 | //iterate backwards to find the last non-empty bin |
| 334 | //this saves iterations in multiple loops |
| 335 | unsigned last_bin = max_bin; |
| 336 | for (; last_bin && !bin_sizes[last_bin]; --last_bin); |
| 337 | //This dominates runtime, mostly in the swap and bin lookups |
| 338 | for (unsigned u = 0; u < last_bin; ++u) { |
| 339 | local_bin = bins + u; |
| 340 | next_bin_start += bin_sizes[u]; |
| 341 | //Iterating over each element in this bin |
| 342 | for (RandomAccessIter current = *local_bin; current < next_bin_start; |
| 343 | ++current) { |
| 344 | //Swapping into place until the correct element has been swapped in |
| 345 | for (target_bin = |
| 346 | end_bin - static_cast<Unsigned_char_type>((*current)[char_offset]); |
| 347 | target_bin != local_bin; |
| 348 | target_bin = |
| 349 | end_bin - static_cast<Unsigned_char_type>((*current)[char_offset])) |
| 350 | iter_swap(current, (*target_bin)++); |
| 351 | } |
| 352 | *local_bin = next_bin_start; |
| 353 | } |
| 354 | bins[last_bin] = lastFull; |
| 355 | //Recursing |
| 356 | RandomAccessIter lastPos = first; |
| 357 | //Skip this loop for empties |
| 358 | for (unsigned u = cache_offset; u <= cache_offset + last_bin; |
| 359 | lastPos = bin_cache[u], ++u) { |
| 360 | size_t count = bin_cache[u] - lastPos; |
| 361 | //don't sort unless there are at least two items to Compare |
| 362 | if (count < 2) |
| 363 | continue; |
| 364 | //using boost::sort::pdqsort if its worst-case is better |
| 365 | if (count < max_size) |
| 366 | boost::sort::pdqsort(lastPos, bin_cache[u], offset_greater_than<Data_type, |
| 367 | Unsigned_char_type>(char_offset + 1)); |
| 368 | else |
| 369 | reverse_string_sort_rec<RandomAccessIter, Unsigned_char_type> |
| 370 | (lastPos, bin_cache[u], char_offset + 1, bin_cache, cache_end, bin_sizes); |
| 371 | } |
| 372 | } |
| 373 | |
| 374 | //String sorting recursive implementation |
| 375 | template <class RandomAccessIter, class Unsigned_char_type, class Get_char, |
| 376 | class Get_length> |
| 377 | inline void |
| 378 | string_sort_rec(RandomAccessIter first, RandomAccessIter last, |
| 379 | size_t char_offset, std::vector<RandomAccessIter> &bin_cache, |
| 380 | unsigned cache_offset, size_t *bin_sizes, |
| 381 | Get_char get_character, Get_length length) |
| 382 | { |
| 383 | typedef typename std::iterator_traits<RandomAccessIter>::value_type |
| 384 | Data_type; |
| 385 | //This section makes handling of long identical substrings much faster |
| 386 | //with a mild average performance impact. |
| 387 | //Iterate to the end of the empties. If all empty, return |
| 388 | while (length(*first) <= char_offset) { |
| 389 | if (++first == last) |
| 390 | return; |
| 391 | } |
| 392 | RandomAccessIter finish = last - 1; |
| 393 | //Getting the last non-empty |
| 394 | for (;length(*finish) <= char_offset; --finish); |
| 395 | ++finish; |
| 396 | update_offset(first, finish, char_offset, get_character, length); |
| 397 | |
| 398 | const unsigned bin_count = (1 << (sizeof(Unsigned_char_type)*8)); |
| 399 | //Equal worst-case of radix and comparison is when bin_count = n*log(n). |
| 400 | const unsigned max_size = bin_count; |
| 401 | const unsigned membin_count = bin_count + 1; |
| 402 | unsigned cache_end; |
| 403 | RandomAccessIter * bins = size_bins(bin_sizes, bin_cache, cache_offset, |
| 404 | cache_end, membin_count) + 1; |
| 405 | |
| 406 | //Calculating the size of each bin; this takes roughly 10% of runtime |
| 407 | for (RandomAccessIter current = first; current != last; ++current) { |
| 408 | if (length(*current) <= char_offset) { |
| 409 | bin_sizes[0]++; |
| 410 | } |
| 411 | else |
| 412 | bin_sizes[get_character((*current), char_offset) + 1]++; |
| 413 | } |
| 414 | //Assign the bin positions |
| 415 | bin_cache[cache_offset] = first; |
| 416 | for (unsigned u = 0; u < membin_count - 1; u++) |
| 417 | bin_cache[cache_offset + u + 1] = |
| 418 | bin_cache[cache_offset + u] + bin_sizes[u]; |
| 419 | |
| 420 | //Swap into place |
| 421 | RandomAccessIter next_bin_start = first; |
| 422 | //handling empty bins |
| 423 | RandomAccessIter * local_bin = &(bin_cache[cache_offset]); |
| 424 | next_bin_start += bin_sizes[0]; |
| 425 | RandomAccessIter * target_bin; |
| 426 | //Iterating over each element in the bin of empties |
| 427 | for (RandomAccessIter current = *local_bin; current < next_bin_start; |
| 428 | ++current) { |
| 429 | //empties belong in this bin |
| 430 | while (length(*current) > char_offset) { |
| 431 | target_bin = bins + get_character((*current), char_offset); |
| 432 | iter_swap(current, (*target_bin)++); |
| 433 | } |
| 434 | } |
| 435 | *local_bin = next_bin_start; |
| 436 | //iterate backwards to find the last bin with elements in it |
| 437 | //this saves iterations in multiple loops |
| 438 | unsigned last_bin = bin_count - 1; |
| 439 | for (; last_bin && !bin_sizes[last_bin + 1]; --last_bin); |
| 440 | //This dominates runtime, mostly in the swap and bin lookups |
| 441 | for (unsigned ii = 0; ii < last_bin; ++ii) { |
| 442 | local_bin = bins + ii; |
| 443 | next_bin_start += bin_sizes[ii + 1]; |
| 444 | //Iterating over each element in this bin |
| 445 | for (RandomAccessIter current = *local_bin; current < next_bin_start; |
| 446 | ++current) { |
| 447 | //Swapping into place until the correct element has been swapped in |
| 448 | for (target_bin = bins + get_character((*current), char_offset); |
| 449 | target_bin != local_bin; |
| 450 | target_bin = bins + get_character((*current), char_offset)) |
| 451 | iter_swap(current, (*target_bin)++); |
| 452 | } |
| 453 | *local_bin = next_bin_start; |
| 454 | } |
| 455 | bins[last_bin] = last; |
| 456 | |
| 457 | //Recursing |
| 458 | RandomAccessIter lastPos = bin_cache[cache_offset]; |
| 459 | //Skip this loop for empties |
| 460 | for (unsigned u = cache_offset + 1; u < cache_offset + last_bin + 2; |
| 461 | lastPos = bin_cache[u], ++u) { |
| 462 | size_t count = bin_cache[u] - lastPos; |
| 463 | //don't sort unless there are at least two items to Compare |
| 464 | if (count < 2) |
| 465 | continue; |
| 466 | //using boost::sort::pdqsort if its worst-case is better |
| 467 | if (count < max_size) |
| 468 | boost::sort::pdqsort(lastPos, bin_cache[u], offset_char_less_than<Data_type, |
| 469 | Get_char, Get_length>(char_offset + 1)); |
| 470 | else |
| 471 | string_sort_rec<RandomAccessIter, Unsigned_char_type, Get_char, |
| 472 | Get_length>(lastPos, bin_cache[u], char_offset + 1, bin_cache, |
| 473 | cache_end, bin_sizes, get_character, length); |
| 474 | } |
| 475 | } |
| 476 | |
| 477 | //String sorting recursive implementation |
| 478 | template <class RandomAccessIter, class Unsigned_char_type, class Get_char, |
| 479 | class Get_length, class Compare> |
| 480 | inline void |
| 481 | string_sort_rec(RandomAccessIter first, RandomAccessIter last, |
| 482 | size_t char_offset, std::vector<RandomAccessIter> &bin_cache, |
| 483 | unsigned cache_offset, size_t *bin_sizes, |
| 484 | Get_char get_character, Get_length length, Compare comp) |
| 485 | { |
| 486 | //This section makes handling of long identical substrings much faster |
| 487 | //with a mild average performance impact. |
| 488 | //Iterate to the end of the empties. If all empty, return |
| 489 | while (length(*first) <= char_offset) { |
| 490 | if (++first == last) |
| 491 | return; |
| 492 | } |
| 493 | RandomAccessIter finish = last - 1; |
| 494 | //Getting the last non-empty |
| 495 | for (;length(*finish) <= char_offset; --finish); |
| 496 | ++finish; |
| 497 | update_offset(first, finish, char_offset, get_character, length); |
| 498 | |
| 499 | const unsigned bin_count = (1 << (sizeof(Unsigned_char_type)*8)); |
| 500 | //Equal worst-case of radix and comparison is when bin_count = n*log(n). |
| 501 | const unsigned max_size = bin_count; |
| 502 | const unsigned membin_count = bin_count + 1; |
| 503 | unsigned cache_end; |
| 504 | RandomAccessIter * bins = size_bins(bin_sizes, bin_cache, cache_offset, |
| 505 | cache_end, membin_count) + 1; |
| 506 | |
| 507 | //Calculating the size of each bin; this takes roughly 10% of runtime |
| 508 | for (RandomAccessIter current = first; current != last; ++current) { |
| 509 | if (length(*current) <= char_offset) { |
| 510 | bin_sizes[0]++; |
| 511 | } |
| 512 | else |
| 513 | bin_sizes[get_character((*current), char_offset) + 1]++; |
| 514 | } |
| 515 | //Assign the bin positions |
| 516 | bin_cache[cache_offset] = first; |
| 517 | for (unsigned u = 0; u < membin_count - 1; u++) |
| 518 | bin_cache[cache_offset + u + 1] = |
| 519 | bin_cache[cache_offset + u] + bin_sizes[u]; |
| 520 | |
| 521 | //Swap into place |
| 522 | RandomAccessIter next_bin_start = first; |
| 523 | //handling empty bins |
| 524 | RandomAccessIter * local_bin = &(bin_cache[cache_offset]); |
| 525 | next_bin_start += bin_sizes[0]; |
| 526 | RandomAccessIter * target_bin; |
| 527 | //Iterating over each element in the bin of empties |
| 528 | for (RandomAccessIter current = *local_bin; current < next_bin_start; |
| 529 | ++current) { |
| 530 | //empties belong in this bin |
| 531 | while (length(*current) > char_offset) { |
| 532 | target_bin = bins + get_character((*current), char_offset); |
| 533 | iter_swap(current, (*target_bin)++); |
| 534 | } |
| 535 | } |
| 536 | *local_bin = next_bin_start; |
| 537 | //iterate backwards to find the last bin with elements in it |
| 538 | //this saves iterations in multiple loops |
| 539 | unsigned last_bin = bin_count - 1; |
| 540 | for (; last_bin && !bin_sizes[last_bin + 1]; --last_bin); |
| 541 | //This dominates runtime, mostly in the swap and bin lookups |
| 542 | for (unsigned u = 0; u < last_bin; ++u) { |
| 543 | local_bin = bins + u; |
| 544 | next_bin_start += bin_sizes[u + 1]; |
| 545 | //Iterating over each element in this bin |
| 546 | for (RandomAccessIter current = *local_bin; current < next_bin_start; |
| 547 | ++current) { |
| 548 | //Swapping into place until the correct element has been swapped in |
| 549 | for (target_bin = bins + get_character((*current), char_offset); |
| 550 | target_bin != local_bin; |
| 551 | target_bin = bins + get_character((*current), char_offset)) |
| 552 | iter_swap(current, (*target_bin)++); |
| 553 | } |
| 554 | *local_bin = next_bin_start; |
| 555 | } |
| 556 | bins[last_bin] = last; |
| 557 | |
| 558 | //Recursing |
| 559 | RandomAccessIter lastPos = bin_cache[cache_offset]; |
| 560 | //Skip this loop for empties |
| 561 | for (unsigned u = cache_offset + 1; u < cache_offset + last_bin + 2; |
| 562 | lastPos = bin_cache[u], ++u) { |
| 563 | size_t count = bin_cache[u] - lastPos; |
| 564 | //don't sort unless there are at least two items to Compare |
| 565 | if (count < 2) |
| 566 | continue; |
| 567 | //using boost::sort::pdqsort if its worst-case is better |
| 568 | if (count < max_size) |
| 569 | boost::sort::pdqsort(lastPos, bin_cache[u], comp); |
| 570 | else |
| 571 | string_sort_rec<RandomAccessIter, Unsigned_char_type, Get_char, |
| 572 | Get_length, Compare> |
| 573 | (lastPos, bin_cache[u], char_offset + 1, bin_cache, cache_end, |
| 574 | bin_sizes, get_character, length, comp); |
| 575 | } |
| 576 | } |
| 577 | |
| 578 | //Sorts strings in reverse order, with empties at the end |
| 579 | template <class RandomAccessIter, class Unsigned_char_type, class Get_char, |
| 580 | class Get_length, class Compare> |
| 581 | inline void |
| 582 | reverse_string_sort_rec(RandomAccessIter first, RandomAccessIter last, |
| 583 | size_t char_offset, std::vector<RandomAccessIter> &bin_cache, |
| 584 | unsigned cache_offset, size_t *bin_sizes, |
| 585 | Get_char get_character, Get_length length, Compare comp) |
| 586 | { |
| 587 | //This section makes handling of long identical substrings much faster |
| 588 | //with a mild average performance impact. |
| 589 | RandomAccessIter curr = first; |
| 590 | //Iterate to the end of the empties. If all empty, return |
| 591 | while (length(*curr) <= char_offset) { |
| 592 | if (++curr == last) |
| 593 | return; |
| 594 | } |
| 595 | //Getting the last non-empty |
| 596 | while (length(*(--last)) <= char_offset); |
| 597 | ++last; |
| 598 | //Offsetting on identical characters. This section works |
| 599 | //a character at a time for optimal worst-case performance. |
| 600 | update_offset(curr, last, char_offset, get_character, length); |
| 601 | |
| 602 | const unsigned bin_count = (1 << (sizeof(Unsigned_char_type)*8)); |
| 603 | //Equal worst-case of radix and comparison is when bin_count = n*log(n). |
| 604 | const unsigned max_size = bin_count; |
| 605 | const unsigned membin_count = bin_count + 1; |
| 606 | const unsigned max_bin = bin_count - 1; |
| 607 | unsigned cache_end; |
| 608 | RandomAccessIter * bins = size_bins(bin_sizes, bin_cache, cache_offset, |
| 609 | cache_end, membin_count); |
| 610 | RandomAccessIter *end_bin = &(bin_cache[cache_offset + max_bin]); |
| 611 | |
| 612 | //Calculating the size of each bin; this takes roughly 10% of runtime |
| 613 | for (RandomAccessIter current = first; current != last; ++current) { |
| 614 | if (length(*current) <= char_offset) { |
| 615 | bin_sizes[bin_count]++; |
| 616 | } |
| 617 | else |
| 618 | bin_sizes[max_bin - get_character((*current), char_offset)]++; |
| 619 | } |
| 620 | //Assign the bin positions |
| 621 | bin_cache[cache_offset] = first; |
| 622 | for (unsigned u = 0; u < membin_count - 1; u++) |
| 623 | bin_cache[cache_offset + u + 1] = |
| 624 | bin_cache[cache_offset + u] + bin_sizes[u]; |
| 625 | |
| 626 | //Swap into place |
| 627 | RandomAccessIter next_bin_start = last; |
| 628 | //handling empty bins |
| 629 | RandomAccessIter * local_bin = &(bin_cache[cache_offset + bin_count]); |
| 630 | RandomAccessIter lastFull = *local_bin; |
| 631 | RandomAccessIter * target_bin; |
| 632 | //Iterating over each element in the bin of empties |
| 633 | for (RandomAccessIter current = *local_bin; current < next_bin_start; |
| 634 | ++current) { |
| 635 | //empties belong in this bin |
| 636 | while (length(*current) > char_offset) { |
| 637 | target_bin = end_bin - get_character((*current), char_offset); |
| 638 | iter_swap(current, (*target_bin)++); |
| 639 | } |
| 640 | } |
| 641 | *local_bin = next_bin_start; |
| 642 | next_bin_start = first; |
| 643 | //iterate backwards to find the last bin with elements in it |
| 644 | //this saves iterations in multiple loops |
| 645 | unsigned last_bin = max_bin; |
| 646 | for (; last_bin && !bin_sizes[last_bin]; --last_bin); |
| 647 | //This dominates runtime, mostly in the swap and bin lookups |
| 648 | for (unsigned u = 0; u < last_bin; ++u) { |
| 649 | local_bin = bins + u; |
| 650 | next_bin_start += bin_sizes[u]; |
| 651 | //Iterating over each element in this bin |
| 652 | for (RandomAccessIter current = *local_bin; current < next_bin_start; |
| 653 | ++current) { |
| 654 | //Swapping into place until the correct element has been swapped in |
| 655 | for (target_bin = end_bin - get_character((*current), char_offset); |
| 656 | target_bin != local_bin; |
| 657 | target_bin = end_bin - get_character((*current), char_offset)) |
| 658 | iter_swap(current, (*target_bin)++); |
| 659 | } |
| 660 | *local_bin = next_bin_start; |
| 661 | } |
| 662 | bins[last_bin] = lastFull; |
| 663 | //Recursing |
| 664 | RandomAccessIter lastPos = first; |
| 665 | //Skip this loop for empties |
| 666 | for (unsigned u = cache_offset; u <= cache_offset + last_bin; |
| 667 | lastPos = bin_cache[u], ++u) { |
| 668 | size_t count = bin_cache[u] - lastPos; |
| 669 | //don't sort unless there are at least two items to Compare |
| 670 | if (count < 2) |
| 671 | continue; |
| 672 | //using boost::sort::pdqsort if its worst-case is better |
| 673 | if (count < max_size) |
| 674 | boost::sort::pdqsort(lastPos, bin_cache[u], comp); |
| 675 | else |
| 676 | reverse_string_sort_rec<RandomAccessIter, Unsigned_char_type, |
| 677 | Get_char, Get_length, Compare> |
| 678 | (lastPos, bin_cache[u], char_offset + 1, bin_cache, cache_end, |
| 679 | bin_sizes, get_character, length, comp); |
| 680 | } |
| 681 | } |
| 682 | |
| 683 | //Holds the bin vector and makes the initial recursive call |
| 684 | template <class RandomAccessIter, class Unsigned_char_type> |
| 685 | inline typename boost::enable_if_c< sizeof(Unsigned_char_type) <= 2, void |
| 686 | >::type |
| 687 | string_sort(RandomAccessIter first, RandomAccessIter last, |
| 688 | Unsigned_char_type) |
| 689 | { |
| 690 | size_t bin_sizes[(1 << (8 * sizeof(Unsigned_char_type))) + 1]; |
| 691 | std::vector<RandomAccessIter> bin_cache; |
| 692 | string_sort_rec<RandomAccessIter, Unsigned_char_type> |
| 693 | (first, last, 0, bin_cache, 0, bin_sizes); |
| 694 | } |
| 695 | |
| 696 | template <class RandomAccessIter, class Unsigned_char_type> |
| 697 | inline typename boost::disable_if_c< sizeof(Unsigned_char_type) <= 2, void |
| 698 | >::type |
| 699 | string_sort(RandomAccessIter first, RandomAccessIter last, |
| 700 | Unsigned_char_type) |
| 701 | { |
| 702 | // Use boost::sort::pdqsort if the char_type is too large for string_sort. |
| 703 | boost::sort::pdqsort(first, last); |
| 704 | } |
| 705 | |
| 706 | //Holds the bin vector and makes the initial recursive call |
| 707 | template <class RandomAccessIter, class Unsigned_char_type> |
| 708 | inline typename boost::enable_if_c< sizeof(Unsigned_char_type) <= 2, void |
| 709 | >::type |
| 710 | reverse_string_sort(RandomAccessIter first, RandomAccessIter last, |
| 711 | Unsigned_char_type) |
| 712 | { |
| 713 | size_t bin_sizes[(1 << (8 * sizeof(Unsigned_char_type))) + 1]; |
| 714 | std::vector<RandomAccessIter> bin_cache; |
| 715 | reverse_string_sort_rec<RandomAccessIter, Unsigned_char_type> |
| 716 | (first, last, 0, bin_cache, 0, bin_sizes); |
| 717 | } |
| 718 | |
| 719 | template <class RandomAccessIter, class Unsigned_char_type> |
| 720 | inline typename boost::disable_if_c< sizeof(Unsigned_char_type) <= 2, void |
| 721 | >::type |
| 722 | reverse_string_sort(RandomAccessIter first, RandomAccessIter last, |
| 723 | Unsigned_char_type) |
| 724 | { |
| 725 | typedef typename std::iterator_traits<RandomAccessIter>::value_type |
| 726 | Data_type; |
| 727 | // Use boost::sort::pdqsort if the char_type is too large for string_sort. |
| 728 | boost::sort::pdqsort(first, last, std::greater<Data_type>()); |
| 729 | } |
| 730 | |
| 731 | //Holds the bin vector and makes the initial recursive call |
| 732 | template <class RandomAccessIter, class Get_char, class Get_length, |
| 733 | class Unsigned_char_type> |
| 734 | inline typename boost::enable_if_c< sizeof(Unsigned_char_type) <= 2, void |
| 735 | >::type |
| 736 | string_sort(RandomAccessIter first, RandomAccessIter last, |
| 737 | Get_char get_character, Get_length length, Unsigned_char_type) |
| 738 | { |
| 739 | size_t bin_sizes[(1 << (8 * sizeof(Unsigned_char_type))) + 1]; |
| 740 | std::vector<RandomAccessIter> bin_cache; |
| 741 | string_sort_rec<RandomAccessIter, Unsigned_char_type, Get_char, |
| 742 | Get_length>(first, last, 0, bin_cache, 0, bin_sizes, get_character, length); |
| 743 | } |
| 744 | |
| 745 | template <class RandomAccessIter, class Get_char, class Get_length, |
| 746 | class Unsigned_char_type> |
| 747 | inline typename boost::disable_if_c< sizeof(Unsigned_char_type) <= 2, void |
| 748 | >::type |
| 749 | string_sort(RandomAccessIter first, RandomAccessIter last, |
| 750 | Get_char get_character, Get_length length, Unsigned_char_type) |
| 751 | { |
| 752 | // Use boost::sort::pdqsort if the char_type is too large for string_sort. |
| 753 | boost::sort::pdqsort(first, last); |
| 754 | } |
| 755 | |
| 756 | //Holds the bin vector and makes the initial recursive call |
| 757 | template <class RandomAccessIter, class Get_char, class Get_length, |
| 758 | class Compare, class Unsigned_char_type> |
| 759 | inline typename boost::enable_if_c< sizeof(Unsigned_char_type) <= 2, void |
| 760 | >::type |
| 761 | string_sort(RandomAccessIter first, RandomAccessIter last, |
| 762 | Get_char get_character, Get_length length, Compare comp, Unsigned_char_type) |
| 763 | { |
| 764 | size_t bin_sizes[(1 << (8 * sizeof(Unsigned_char_type))) + 1]; |
| 765 | std::vector<RandomAccessIter> bin_cache; |
| 766 | string_sort_rec<RandomAccessIter, Unsigned_char_type, Get_char |
| 767 | , Get_length, Compare> |
| 768 | (first, last, 0, bin_cache, 0, bin_sizes, get_character, length, comp); |
| 769 | } |
| 770 | |
| 771 | //disable_if_c was refusing to compile, so rewrote to use enable_if_c |
| 772 | template <class RandomAccessIter, class Get_char, class Get_length, |
| 773 | class Compare, class Unsigned_char_type> |
| 774 | inline typename boost::enable_if_c< (sizeof(Unsigned_char_type) > 2), void |
| 775 | >::type |
| 776 | string_sort(RandomAccessIter first, RandomAccessIter last, |
| 777 | Get_char get_character, Get_length length, Compare comp, Unsigned_char_type) |
| 778 | { |
| 779 | // Use boost::sort::pdqsort if the char_type is too large for string_sort. |
| 780 | boost::sort::pdqsort(first, last, comp); |
| 781 | } |
| 782 | |
| 783 | //Holds the bin vector and makes the initial recursive call |
| 784 | template <class RandomAccessIter, class Get_char, class Get_length, |
| 785 | class Compare, class Unsigned_char_type> |
| 786 | inline typename boost::enable_if_c< sizeof(Unsigned_char_type) <= 2, void |
| 787 | >::type |
| 788 | reverse_string_sort(RandomAccessIter first, RandomAccessIter last, |
| 789 | Get_char get_character, Get_length length, Compare comp, Unsigned_char_type) |
| 790 | { |
| 791 | size_t bin_sizes[(1 << (8 * sizeof(Unsigned_char_type))) + 1]; |
| 792 | std::vector<RandomAccessIter> bin_cache; |
| 793 | reverse_string_sort_rec<RandomAccessIter, Unsigned_char_type, Get_char, |
| 794 | Get_length, Compare> |
| 795 | (first, last, 0, bin_cache, 0, bin_sizes, get_character, length, comp); |
| 796 | } |
| 797 | |
| 798 | template <class RandomAccessIter, class Get_char, class Get_length, |
| 799 | class Compare, class Unsigned_char_type> |
| 800 | inline typename boost::disable_if_c< sizeof(Unsigned_char_type) <= 2, void |
| 801 | >::type |
| 802 | reverse_string_sort(RandomAccessIter first, RandomAccessIter last, |
| 803 | Get_char get_character, Get_length length, Compare comp, Unsigned_char_type) |
| 804 | { |
| 805 | // Use boost::sort::pdqsort if the char_type is too large for string_sort. |
| 806 | boost::sort::pdqsort(first, last, comp); |
| 807 | } |
| 808 | } |
| 809 | } |
| 810 | } |
| 811 | } |
| 812 | |
| 813 | #endif |
| 814 | |