| 1 | /** |
| 2 | * Copyright 2013, GitHub, Inc |
| 3 | * Copyright 2009-2013, Daniel Lemire, Cliff Moon, |
| 4 | * David McIntosh, Robert Becho, Google Inc. and Veronika Zenz |
| 5 | * |
| 6 | * This program is free software; you can redistribute it and/or |
| 7 | * modify it under the terms of the GNU General Public License |
| 8 | * as published by the Free Software Foundation; either version 2 |
| 9 | * of the License, or (at your option) any later version. |
| 10 | * |
| 11 | * This program is distributed in the hope that it will be useful, |
| 12 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 13 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 14 | * GNU General Public License for more details. |
| 15 | * |
| 16 | * You should have received a copy of the GNU General Public License |
| 17 | * along with this program; if not, see <http://www.gnu.org/licenses/>. |
| 18 | */ |
| 19 | |
| 20 | #include "git-compat-util.h" |
| 21 | #include "ewok.h" |
| 22 | #include "ewok_rlw.h" |
| 23 | |
| 24 | static inline size_t min_size(size_t a, size_t b) |
| 25 | { |
| 26 | return a < b ? a : b; |
| 27 | } |
| 28 | |
| 29 | static inline size_t max_size(size_t a, size_t b) |
| 30 | { |
| 31 | return a > b ? a : b; |
| 32 | } |
| 33 | |
| 34 | static inline void buffer_grow(struct ewah_bitmap *self, size_t new_size) |
| 35 | { |
| 36 | size_t rlw_offset = (uint8_t *)self->rlw - (uint8_t *)self->buffer; |
| 37 | ALLOC_GROW(self->buffer, new_size, self->alloc_size); |
| 38 | self->rlw = self->buffer + (rlw_offset / sizeof(eword_t)); |
| 39 | } |
| 40 | |
| 41 | static inline void buffer_push(struct ewah_bitmap *self, eword_t value) |
| 42 | { |
| 43 | buffer_grow(self, self->buffer_size + 1); |
| 44 | self->buffer[self->buffer_size++] = value; |
| 45 | } |
| 46 | |
| 47 | static void buffer_push_rlw(struct ewah_bitmap *self, eword_t value) |
| 48 | { |
| 49 | buffer_push(self, value); |
| 50 | self->rlw = self->buffer + self->buffer_size - 1; |
| 51 | } |
| 52 | |
| 53 | static size_t add_empty_words(struct ewah_bitmap *self, int v, size_t number) |
| 54 | { |
| 55 | size_t added = 0; |
| 56 | eword_t runlen, can_add; |
| 57 | |
| 58 | if (rlw_get_run_bit(self->rlw) != v && rlw_size(self->rlw) == 0) { |
| 59 | rlw_set_run_bit(self->rlw, v); |
| 60 | } else if (rlw_get_literal_words(self->rlw) != 0 || |
| 61 | rlw_get_run_bit(self->rlw) != v) { |
| 62 | buffer_push_rlw(self, 0); |
| 63 | if (v) rlw_set_run_bit(self->rlw, v); |
| 64 | added++; |
| 65 | } |
| 66 | |
| 67 | runlen = rlw_get_running_len(self->rlw); |
| 68 | can_add = min_size(number, RLW_LARGEST_RUNNING_COUNT - runlen); |
| 69 | |
| 70 | rlw_set_running_len(self->rlw, runlen + can_add); |
| 71 | number -= can_add; |
| 72 | |
| 73 | while (number >= RLW_LARGEST_RUNNING_COUNT) { |
| 74 | buffer_push_rlw(self, 0); |
| 75 | added++; |
| 76 | if (v) rlw_set_run_bit(self->rlw, v); |
| 77 | rlw_set_running_len(self->rlw, RLW_LARGEST_RUNNING_COUNT); |
| 78 | number -= RLW_LARGEST_RUNNING_COUNT; |
| 79 | } |
| 80 | |
| 81 | if (number > 0) { |
| 82 | buffer_push_rlw(self, 0); |
| 83 | added++; |
| 84 | |
| 85 | if (v) rlw_set_run_bit(self->rlw, v); |
| 86 | rlw_set_running_len(self->rlw, number); |
| 87 | } |
| 88 | |
| 89 | return added; |
| 90 | } |
| 91 | |
| 92 | size_t ewah_add_empty_words(struct ewah_bitmap *self, int v, size_t number) |
| 93 | { |
| 94 | if (number == 0) |
| 95 | return 0; |
| 96 | |
| 97 | self->bit_size += number * BITS_IN_EWORD; |
| 98 | return add_empty_words(self, v, number); |
| 99 | } |
| 100 | |
| 101 | static size_t add_literal(struct ewah_bitmap *self, eword_t new_data) |
| 102 | { |
| 103 | eword_t current_num = rlw_get_literal_words(self->rlw); |
| 104 | |
| 105 | if (current_num >= RLW_LARGEST_LITERAL_COUNT) { |
| 106 | buffer_push_rlw(self, 0); |
| 107 | |
| 108 | rlw_set_literal_words(self->rlw, 1); |
| 109 | buffer_push(self, new_data); |
| 110 | return 2; |
| 111 | } |
| 112 | |
| 113 | rlw_set_literal_words(self->rlw, current_num + 1); |
| 114 | |
| 115 | /* sanity check */ |
| 116 | assert(rlw_get_literal_words(self->rlw) == current_num + 1); |
| 117 | |
| 118 | buffer_push(self, new_data); |
| 119 | return 1; |
| 120 | } |
| 121 | |
| 122 | void ewah_add_dirty_words( |
| 123 | struct ewah_bitmap *self, const eword_t *buffer, |
| 124 | size_t number, int negate) |
| 125 | { |
| 126 | size_t literals, can_add; |
| 127 | |
| 128 | while (1) { |
| 129 | literals = rlw_get_literal_words(self->rlw); |
| 130 | can_add = min_size(number, RLW_LARGEST_LITERAL_COUNT - literals); |
| 131 | |
| 132 | rlw_set_literal_words(self->rlw, literals + can_add); |
| 133 | |
| 134 | buffer_grow(self, self->buffer_size + can_add); |
| 135 | |
| 136 | if (negate) { |
| 137 | size_t i; |
| 138 | for (i = 0; i < can_add; ++i) |
| 139 | self->buffer[self->buffer_size++] = ~buffer[i]; |
| 140 | } else { |
| 141 | memcpy(self->buffer + self->buffer_size, |
| 142 | buffer, can_add * sizeof(eword_t)); |
| 143 | self->buffer_size += can_add; |
| 144 | } |
| 145 | |
| 146 | self->bit_size += can_add * BITS_IN_EWORD; |
| 147 | |
| 148 | if (number - can_add == 0) |
| 149 | break; |
| 150 | |
| 151 | buffer_push_rlw(self, 0); |
| 152 | buffer += can_add; |
| 153 | number -= can_add; |
| 154 | } |
| 155 | } |
| 156 | |
| 157 | static size_t add_empty_word(struct ewah_bitmap *self, int v) |
| 158 | { |
| 159 | int no_literal = (rlw_get_literal_words(self->rlw) == 0); |
| 160 | eword_t run_len = rlw_get_running_len(self->rlw); |
| 161 | |
| 162 | if (no_literal && run_len == 0) { |
| 163 | rlw_set_run_bit(self->rlw, v); |
| 164 | assert(rlw_get_run_bit(self->rlw) == v); |
| 165 | } |
| 166 | |
| 167 | if (no_literal && rlw_get_run_bit(self->rlw) == v && |
| 168 | run_len < RLW_LARGEST_RUNNING_COUNT) { |
| 169 | rlw_set_running_len(self->rlw, run_len + 1); |
| 170 | assert(rlw_get_running_len(self->rlw) == run_len + 1); |
| 171 | return 0; |
| 172 | } else { |
| 173 | buffer_push_rlw(self, 0); |
| 174 | |
| 175 | assert(rlw_get_running_len(self->rlw) == 0); |
| 176 | assert(rlw_get_run_bit(self->rlw) == 0); |
| 177 | assert(rlw_get_literal_words(self->rlw) == 0); |
| 178 | |
| 179 | rlw_set_run_bit(self->rlw, v); |
| 180 | assert(rlw_get_run_bit(self->rlw) == v); |
| 181 | |
| 182 | rlw_set_running_len(self->rlw, 1); |
| 183 | assert(rlw_get_running_len(self->rlw) == 1); |
| 184 | assert(rlw_get_literal_words(self->rlw) == 0); |
| 185 | return 1; |
| 186 | } |
| 187 | } |
| 188 | |
| 189 | size_t ewah_add(struct ewah_bitmap *self, eword_t word) |
| 190 | { |
| 191 | self->bit_size += BITS_IN_EWORD; |
| 192 | |
| 193 | if (word == 0) |
| 194 | return add_empty_word(self, 0); |
| 195 | |
| 196 | if (word == (eword_t)(~0)) |
| 197 | return add_empty_word(self, 1); |
| 198 | |
| 199 | return add_literal(self, word); |
| 200 | } |
| 201 | |
| 202 | void ewah_set(struct ewah_bitmap *self, size_t i) |
| 203 | { |
| 204 | const size_t dist = |
| 205 | DIV_ROUND_UP(i + 1, BITS_IN_EWORD) - |
| 206 | DIV_ROUND_UP(self->bit_size, BITS_IN_EWORD); |
| 207 | |
| 208 | assert(i >= self->bit_size); |
| 209 | |
| 210 | self->bit_size = i + 1; |
| 211 | |
| 212 | if (dist > 0) { |
| 213 | if (dist > 1) |
| 214 | add_empty_words(self, 0, dist - 1); |
| 215 | |
| 216 | add_literal(self, (eword_t)1 << (i % BITS_IN_EWORD)); |
| 217 | return; |
| 218 | } |
| 219 | |
| 220 | if (rlw_get_literal_words(self->rlw) == 0) { |
| 221 | rlw_set_running_len(self->rlw, |
| 222 | rlw_get_running_len(self->rlw) - 1); |
| 223 | add_literal(self, (eword_t)1 << (i % BITS_IN_EWORD)); |
| 224 | return; |
| 225 | } |
| 226 | |
| 227 | self->buffer[self->buffer_size - 1] |= |
| 228 | ((eword_t)1 << (i % BITS_IN_EWORD)); |
| 229 | |
| 230 | /* check if we just completed a stream of 1s */ |
| 231 | if (self->buffer[self->buffer_size - 1] == (eword_t)(~0)) { |
| 232 | self->buffer[--self->buffer_size] = 0; |
| 233 | rlw_set_literal_words(self->rlw, |
| 234 | rlw_get_literal_words(self->rlw) - 1); |
| 235 | add_empty_word(self, 1); |
| 236 | } |
| 237 | } |
| 238 | |
| 239 | void ewah_each_bit(struct ewah_bitmap *self, void (*callback)(size_t, void*), void *payload) |
| 240 | { |
| 241 | size_t pos = 0; |
| 242 | size_t pointer = 0; |
| 243 | size_t k; |
| 244 | |
| 245 | while (pointer < self->buffer_size) { |
| 246 | eword_t *word = &self->buffer[pointer]; |
| 247 | |
| 248 | if (rlw_get_run_bit(word)) { |
| 249 | size_t len = rlw_get_running_len(word) * BITS_IN_EWORD; |
| 250 | for (k = 0; k < len; ++k, ++pos) |
| 251 | callback(pos, payload); |
| 252 | } else { |
| 253 | pos += rlw_get_running_len(word) * BITS_IN_EWORD; |
| 254 | } |
| 255 | |
| 256 | ++pointer; |
| 257 | |
| 258 | for (k = 0; k < rlw_get_literal_words(word); ++k) { |
| 259 | /* todo: zero count optimization */ |
| 260 | for (size_t c = 0; c < BITS_IN_EWORD; ++c, ++pos) { |
| 261 | if ((self->buffer[pointer] & ((eword_t)1 << c)) != 0) |
| 262 | callback(pos, payload); |
| 263 | } |
| 264 | |
| 265 | ++pointer; |
| 266 | } |
| 267 | } |
| 268 | } |
| 269 | |
| 270 | /** |
| 271 | * Clear all the bits in the bitmap. Does not free or resize |
| 272 | * memory. |
| 273 | */ |
| 274 | static void ewah_clear(struct ewah_bitmap *self) |
| 275 | { |
| 276 | self->buffer_size = 1; |
| 277 | self->buffer[0] = 0; |
| 278 | self->bit_size = 0; |
| 279 | self->rlw = self->buffer; |
| 280 | } |
| 281 | |
| 282 | struct ewah_bitmap *ewah_new(void) |
| 283 | { |
| 284 | struct ewah_bitmap *self; |
| 285 | |
| 286 | self = xmalloc(sizeof(struct ewah_bitmap)); |
| 287 | self->alloc_size = 32; |
| 288 | ALLOC_ARRAY(self->buffer, self->alloc_size); |
| 289 | |
| 290 | ewah_clear(self); |
| 291 | return self; |
| 292 | } |
| 293 | |
| 294 | void ewah_free(struct ewah_bitmap *self) |
| 295 | { |
| 296 | if (!self) |
| 297 | return; |
| 298 | |
| 299 | if (self->alloc_size) |
| 300 | free(self->buffer); |
| 301 | |
| 302 | free(self); |
| 303 | } |
| 304 | |
| 305 | static void read_new_rlw(struct ewah_iterator *it) |
| 306 | { |
| 307 | const eword_t *word = NULL; |
| 308 | |
| 309 | it->literals = 0; |
| 310 | it->compressed = 0; |
| 311 | |
| 312 | while (1) { |
| 313 | word = &it->buffer[it->pointer]; |
| 314 | |
| 315 | it->rl = rlw_get_running_len(word); |
| 316 | it->lw = rlw_get_literal_words(word); |
| 317 | it->b = rlw_get_run_bit(word); |
| 318 | |
| 319 | if (it->rl || it->lw) |
| 320 | return; |
| 321 | |
| 322 | if (it->pointer < it->buffer_size - 1) { |
| 323 | it->pointer++; |
| 324 | } else { |
| 325 | it->pointer = it->buffer_size; |
| 326 | return; |
| 327 | } |
| 328 | } |
| 329 | } |
| 330 | |
| 331 | int ewah_iterator_next(eword_t *next, struct ewah_iterator *it) |
| 332 | { |
| 333 | if (it->pointer >= it->buffer_size) |
| 334 | return 0; |
| 335 | |
| 336 | if (it->compressed < it->rl) { |
| 337 | it->compressed++; |
| 338 | *next = it->b ? (eword_t)(~0) : 0; |
| 339 | } else { |
| 340 | assert(it->literals < it->lw); |
| 341 | |
| 342 | it->literals++; |
| 343 | it->pointer++; |
| 344 | |
| 345 | assert(it->pointer < it->buffer_size); |
| 346 | |
| 347 | *next = it->buffer[it->pointer]; |
| 348 | } |
| 349 | |
| 350 | if (it->compressed == it->rl && it->literals == it->lw) { |
| 351 | if (++it->pointer < it->buffer_size) |
| 352 | read_new_rlw(it); |
| 353 | } |
| 354 | |
| 355 | return 1; |
| 356 | } |
| 357 | |
| 358 | void ewah_iterator_init(struct ewah_iterator *it, struct ewah_bitmap *parent) |
| 359 | { |
| 360 | it->buffer = parent->buffer; |
| 361 | it->buffer_size = parent->buffer_size; |
| 362 | it->pointer = 0; |
| 363 | |
| 364 | it->lw = 0; |
| 365 | it->rl = 0; |
| 366 | it->compressed = 0; |
| 367 | it->literals = 0; |
| 368 | it->b = 0; |
| 369 | |
| 370 | if (it->pointer < it->buffer_size) |
| 371 | read_new_rlw(it); |
| 372 | } |
| 373 | |
| 374 | void ewah_or_iterator_init(struct ewah_or_iterator *it, |
| 375 | struct ewah_bitmap **parents, size_t nr) |
| 376 | { |
| 377 | size_t i; |
| 378 | |
| 379 | memset(it, 0, sizeof(*it)); |
| 380 | |
| 381 | ALLOC_ARRAY(it->its, nr); |
| 382 | for (i = 0; i < nr; i++) |
| 383 | ewah_iterator_init(&it->its[it->nr++], parents[i]); |
| 384 | } |
| 385 | |
| 386 | int ewah_or_iterator_next(eword_t *next, struct ewah_or_iterator *it) |
| 387 | { |
| 388 | eword_t buf, out = 0; |
| 389 | size_t i; |
| 390 | int ret = 0; |
| 391 | |
| 392 | for (i = 0; i < it->nr; i++) |
| 393 | if (ewah_iterator_next(&buf, &it->its[i])) { |
| 394 | out |= buf; |
| 395 | ret = 1; |
| 396 | } |
| 397 | |
| 398 | *next = out; |
| 399 | return ret; |
| 400 | } |
| 401 | |
| 402 | void ewah_or_iterator_release(struct ewah_or_iterator *it) |
| 403 | { |
| 404 | free(it->its); |
| 405 | } |
| 406 | |
| 407 | void ewah_xor( |
| 408 | struct ewah_bitmap *ewah_i, |
| 409 | struct ewah_bitmap *ewah_j, |
| 410 | struct ewah_bitmap *out) |
| 411 | { |
| 412 | struct rlw_iterator rlw_i; |
| 413 | struct rlw_iterator rlw_j; |
| 414 | size_t literals; |
| 415 | |
| 416 | rlwit_init(&rlw_i, ewah_i); |
| 417 | rlwit_init(&rlw_j, ewah_j); |
| 418 | |
| 419 | while (rlwit_word_size(&rlw_i) > 0 && rlwit_word_size(&rlw_j) > 0) { |
| 420 | while (rlw_i.rlw.running_len > 0 || rlw_j.rlw.running_len > 0) { |
| 421 | struct rlw_iterator *prey, *predator; |
| 422 | size_t index; |
| 423 | int negate_words; |
| 424 | |
| 425 | if (rlw_i.rlw.running_len < rlw_j.rlw.running_len) { |
| 426 | prey = &rlw_i; |
| 427 | predator = &rlw_j; |
| 428 | } else { |
| 429 | prey = &rlw_j; |
| 430 | predator = &rlw_i; |
| 431 | } |
| 432 | |
| 433 | negate_words = !!predator->rlw.running_bit; |
| 434 | index = rlwit_discharge(prey, out, |
| 435 | predator->rlw.running_len, negate_words); |
| 436 | |
| 437 | ewah_add_empty_words(out, negate_words, |
| 438 | predator->rlw.running_len - index); |
| 439 | |
| 440 | rlwit_discard_first_words(predator, |
| 441 | predator->rlw.running_len); |
| 442 | } |
| 443 | |
| 444 | literals = min_size( |
| 445 | rlw_i.rlw.literal_words, |
| 446 | rlw_j.rlw.literal_words); |
| 447 | |
| 448 | if (literals) { |
| 449 | size_t k; |
| 450 | |
| 451 | for (k = 0; k < literals; ++k) { |
| 452 | ewah_add(out, |
| 453 | rlw_i.buffer[rlw_i.literal_word_start + k] ^ |
| 454 | rlw_j.buffer[rlw_j.literal_word_start + k] |
| 455 | ); |
| 456 | } |
| 457 | |
| 458 | rlwit_discard_first_words(&rlw_i, literals); |
| 459 | rlwit_discard_first_words(&rlw_j, literals); |
| 460 | } |
| 461 | } |
| 462 | |
| 463 | if (rlwit_word_size(&rlw_i) > 0) |
| 464 | rlwit_discharge(&rlw_i, out, ~0, 0); |
| 465 | else |
| 466 | rlwit_discharge(&rlw_j, out, ~0, 0); |
| 467 | |
| 468 | out->bit_size = max_size(ewah_i->bit_size, ewah_j->bit_size); |
| 469 | } |
| 470 | |
| 471 | #define BITMAP_POOL_MAX 16 |
| 472 | static struct ewah_bitmap *bitmap_pool[BITMAP_POOL_MAX]; |
| 473 | static size_t bitmap_pool_size; |
| 474 | |
| 475 | struct ewah_bitmap *ewah_pool_new(void) |
| 476 | { |
| 477 | if (bitmap_pool_size) |
| 478 | return bitmap_pool[--bitmap_pool_size]; |
| 479 | |
| 480 | return ewah_new(); |
| 481 | } |
| 482 | |
| 483 | void ewah_pool_free(struct ewah_bitmap *self) |
| 484 | { |
| 485 | if (!self) |
| 486 | return; |
| 487 | |
| 488 | if (bitmap_pool_size == BITMAP_POOL_MAX || |
| 489 | self->alloc_size == 0) { |
| 490 | ewah_free(self); |
| 491 | return; |
| 492 | } |
| 493 | |
| 494 | ewah_clear(self); |
| 495 | bitmap_pool[bitmap_pool_size++] = self; |
| 496 | } |
| 497 | |
| 498 | uint32_t ewah_checksum(struct ewah_bitmap *self) |
| 499 | { |
| 500 | const uint8_t *p = (uint8_t *)self->buffer; |
| 501 | uint32_t crc = (uint32_t)self->bit_size; |
| 502 | size_t size = self->buffer_size * sizeof(eword_t); |
| 503 | |
| 504 | while (size--) |
| 505 | crc = (crc << 5) - crc + (uint32_t)*p++; |
| 506 | |
| 507 | return crc; |
| 508 | } |