| 1 |
/*------------------------------------------------------------------------- |
| 2 |
* |
| 3 |
* bitmapset.c |
| 4 |
* PostgreSQL generic bitmap set package |
| 5 |
* |
| 6 |
* A bitmap set can represent any set of nonnegative integers, although |
| 7 |
* it is mainly intended for sets where the maximum value is not large, |
| 8 |
* say at most a few hundred. By convention, we always represent a set with |
| 9 |
* the minimum possible number of words, i.e, there are never any trailing |
| 10 |
* zero words. Enforcing this requires that an empty set is represented as |
| 11 |
* NULL. Because an empty Bitmapset is represented as NULL, a non-NULL |
| 12 |
* Bitmapset always has at least 1 Bitmapword. We can exploit this fact to |
| 13 |
* speed up various loops over the Bitmapset's words array by using "do while" |
| 14 |
* loops instead of "for" loops. This means the code does not waste time |
| 15 |
* checking the loop condition before the first iteration. For Bitmapsets |
| 16 |
* containing only a single word (likely the majority of them) this halves the |
| 17 |
* number of loop condition checks. |
| 18 |
* |
| 19 |
* Callers must ensure that the set returned by functions in this file which |
| 20 |
* adjust the members of an existing set is assigned to all pointers pointing |
| 21 |
* to that existing set. No guarantees are made that we'll ever modify the |
| 22 |
* existing set in-place and return it. |
| 23 |
* |
| 24 |
* To help find bugs caused by callers failing to record the return value of |
| 25 |
* the function which manipulates an existing set, we support building with |
| 26 |
* REALLOCATE_BITMAPSETS. This results in the set being reallocated each time |
| 27 |
* the set is altered and the existing being pfreed. This is useful as if any |
| 28 |
* references still exist to the old set, we're more likely to notice as |
| 29 |
* any users of the old set will be accessing pfree'd memory. This option is |
| 30 |
* only intended to be used for debugging. |
| 31 |
* |
| 32 |
* Copyright (c) 2003-2025, PostgreSQL Global Development Group |
| 33 |
* |
| 34 |
* IDENTIFICATION |
| 35 |
* src/backend/nodes/bitmapset.c |
| 36 |
* |
| 37 |
*------------------------------------------------------------------------- |
| 38 |
*/ |
| 39 |
#include "postgres.h" |
| 40 |
|
| 41 |
#include "common/hashfn.h" |
| 42 |
#include "nodes/bitmapset.h" |
| 43 |
#include "nodes/pg_list.h" |
| 44 |
#include "port/pg_bitutils.h" |
| 45 |
|
| 46 |
|
| 47 |
#define WORDNUM(x) ((x) / BITS_PER_BITMAPWORD) |
| 48 |
#define BITNUM(x) ((x) % BITS_PER_BITMAPWORD) |
| 49 |
|
| 50 |
#define BITMAPSET_SIZE(nwords) \ |
| 51 |
(offsetof(Bitmapset, words) + (nwords) * sizeof(bitmapword)) |
| 52 |
|
| 53 |
/*---------- |
| 54 |
* This is a well-known cute trick for isolating the rightmost one-bit |
| 55 |
* in a word. It assumes two's complement arithmetic. Consider any |
| 56 |
* nonzero value, and focus attention on the rightmost one. The value is |
| 57 |
* then something like |
| 58 |
* xxxxxx10000 |
| 59 |
* where x's are unspecified bits. The two's complement negative is formed |
| 60 |
* by inverting all the bits and adding one. Inversion gives |
| 61 |
* yyyyyy01111 |
| 62 |
* where each y is the inverse of the corresponding x. Incrementing gives |
| 63 |
* yyyyyy10000 |
| 64 |
* and then ANDing with the original value gives |
| 65 |
* 00000010000 |
| 66 |
* This works for all cases except original value = zero, where of course |
| 67 |
* we get zero. |
| 68 |
*---------- |
| 69 |
*/ |
| 70 |
#define RIGHTMOST_ONE(x) ((signedbitmapword) (x) & -((signedbitmapword) (x))) |
| 71 |
|
| 72 |
#define HAS_MULTIPLE_ONES(x) ((bitmapword) RIGHTMOST_ONE(x) != (x)) |
| 73 |
|
| 74 |
#ifdef USE_ASSERT_CHECKING |
| 75 |
/* |
| 76 |
* bms_is_valid_set - for cassert builds to check for valid sets |
| 77 |
*/ |
| 78 |
static bool |
| 79 |
bms_is_valid_set(const Bitmapset *a) |
| 80 |
{ |
| 81 |
/* NULL is the correct representation of an empty set */ |
| 82 |
if (a == NULL) |
| 83 |
return true; |
| 84 |
|
| 85 |
/* check the node tag is set correctly. pfree'd pointer, maybe? */ |
| 86 |
if (!IsA(a, Bitmapset)) |
| 87 |
return false; |
| 88 |
|
| 89 |
/* trailing zero words are not allowed */ |
| 90 |
if (a->words[a->nwords - 1] == 0) |
| 91 |
return false; |
| 92 |
|
| 93 |
return true; |
| 94 |
} |
| 95 |
#endif |
| 96 |
|
| 97 |
#ifdef REALLOCATE_BITMAPSETS |
| 98 |
/* |
| 99 |
* bms_copy_and_free |
| 100 |
* Only required in REALLOCATE_BITMAPSETS builds. Provide a simple way |
| 101 |
* to return a freshly allocated set and pfree the original. |
| 102 |
* |
| 103 |
* Note: callers which accept multiple sets must be careful when calling this |
| 104 |
* function to clone one parameter as other parameters may point to the same |
| 105 |
* set. A good option is to call this just before returning the resulting |
| 106 |
* set. |
| 107 |
*/ |
| 108 |
static Bitmapset * |
| 109 |
bms_copy_and_free(Bitmapset *a) |
| 110 |
{ |
| 111 |
Bitmapset *c = bms_copy(a); |
| 112 |
|
| 113 |
bms_free(a); |
| 114 |
return c; |
| 115 |
} |
| 116 |
#endif |
| 117 |
|
| 118 |
/* |
| 119 |
* bms_copy - make a palloc'd copy of a bitmapset |
| 120 |
*/ |
| 121 |
Bitmapset * |
| 122 |
bms_copy(const Bitmapset *a) |
| 123 |
{ |
| 124 |
Bitmapset *result; |
| 125 |
size_t size; |
| 126 |
|
| 127 |
Assert(bms_is_valid_set(a)); |
| 128 |
|
| 129 |
if (a == NULL) |
| 130 |
return NULL; |
| 131 |
|
| 132 |
size = BITMAPSET_SIZE(a->nwords); |
| 133 |
result = (Bitmapset *) palloc(size); |
| 134 |
memcpy(result, a, size); |
| 135 |
return result; |
| 136 |
} |
| 137 |
|
| 138 |
/* |
| 139 |
* bms_equal - are two bitmapsets equal? or both NULL? |
| 140 |
*/ |
| 141 |
bool |
| 142 |
bms_equal(const Bitmapset *a, const Bitmapset *b) |
| 143 |
{ |
| 144 |
int i; |
| 145 |
|
| 146 |
Assert(bms_is_valid_set(a)); |
| 147 |
Assert(bms_is_valid_set(b)); |
| 148 |
|
| 149 |
/* Handle cases where either input is NULL */ |
| 150 |
if (a == NULL) |
| 151 |
{ |
| 152 |
if (b == NULL) |
| 153 |
return true; |
| 154 |
return false; |
| 155 |
} |
| 156 |
else if (b == NULL) |
| 157 |
return false; |
| 158 |
|
| 159 |
/* can't be equal if the word counts don't match */ |
| 160 |
if (a->nwords != b->nwords) |
| 161 |
return false; |
| 162 |
|
| 163 |
/* check each word matches */ |
| 164 |
i = 0; |
| 165 |
do |
| 166 |
{ |
| 167 |
if (a->words[i] != b->words[i]) |
| 168 |
return false; |
| 169 |
} while (++i < a->nwords); |
| 170 |
|
| 171 |
return true; |
| 172 |
} |
| 173 |
|
| 174 |
/* |
| 175 |
* bms_compare - qsort-style comparator for bitmapsets |
| 176 |
* |
| 177 |
* This guarantees to report values as equal iff bms_equal would say they are |
| 178 |
* equal. Otherwise, the highest-numbered bit that is set in one value but |
| 179 |
* not the other determines the result. (This rule means that, for example, |
| 180 |
* {6} is greater than {5}, which seems plausible.) |
| 181 |
*/ |
| 182 |
int |
| 183 |
bms_compare(const Bitmapset *a, const Bitmapset *b) |
| 184 |
{ |
| 185 |
int i; |
| 186 |
|
| 187 |
Assert(bms_is_valid_set(a)); |
| 188 |
Assert(bms_is_valid_set(b)); |
| 189 |
|
| 190 |
/* Handle cases where either input is NULL */ |
| 191 |
if (a == NULL) |
| 192 |
return (b == NULL) ? 0 : -1; |
| 193 |
else if (b == NULL) |
| 194 |
return +1; |
| 195 |
|
| 196 |
/* the set with the most words must be greater */ |
| 197 |
if (a->nwords != b->nwords) |
| 198 |
return (a->nwords > b->nwords) ? +1 : -1; |
| 199 |
|
| 200 |
i = a->nwords - 1; |
| 201 |
do |
| 202 |
{ |
| 203 |
bitmapword aw = a->words[i]; |
| 204 |
bitmapword bw = b->words[i]; |
| 205 |
|
| 206 |
if (aw != bw) |
| 207 |
return (aw > bw) ? +1 : -1; |
| 208 |
} while (--i >= 0); |
| 209 |
return 0; |
| 210 |
} |
| 211 |
|
| 212 |
/* |
| 213 |
* bms_make_singleton - build a bitmapset containing a single member |
| 214 |
*/ |
| 215 |
Bitmapset * |
| 216 |
bms_make_singleton(int x) |
| 217 |
{ |
| 218 |
Bitmapset *result; |
| 219 |
int wordnum, |
| 220 |
bitnum; |
| 221 |
|
| 222 |
if (x < 0) |
| 223 |
elog(ERROR, "negative bitmapset member not allowed"); |
| 224 |
wordnum = WORDNUM(x); |
| 225 |
bitnum = BITNUM(x); |
| 226 |
result = (Bitmapset *) palloc0(BITMAPSET_SIZE(wordnum + 1)); |
| 227 |
result->type = T_Bitmapset; |
| 228 |
result->nwords = wordnum + 1; |
| 229 |
result->words[wordnum] = ((bitmapword) 1 << bitnum); |
| 230 |
return result; |
| 231 |
} |
| 232 |
|
| 233 |
/* |
| 234 |
* bms_free - free a bitmapset |
| 235 |
* |
| 236 |
* Same as pfree except for allowing NULL input |
| 237 |
*/ |
| 238 |
void |
| 239 |
bms_free(Bitmapset *a) |
| 240 |
{ |
| 241 |
if (a) |
| 242 |
pfree(a); |
| 243 |
} |
| 244 |
|
| 245 |
|
| 246 |
/* |
| 247 |
* bms_union - create and return a new set containing all members from both |
| 248 |
* input sets. Both inputs are left unmodified. |
| 249 |
*/ |
| 250 |
Bitmapset * |
| 251 |
bms_union(const Bitmapset *a, const Bitmapset *b) |
| 252 |
{ |
| 253 |
Bitmapset *result; |
| 254 |
const Bitmapset *other; |
| 255 |
int otherlen; |
| 256 |
int i; |
| 257 |
|
| 258 |
Assert(bms_is_valid_set(a)); |
| 259 |
Assert(bms_is_valid_set(b)); |
| 260 |
|
| 261 |
/* Handle cases where either input is NULL */ |
| 262 |
if (a == NULL) |
| 263 |
return bms_copy(b); |
| 264 |
if (b == NULL) |
| 265 |
return bms_copy(a); |
| 266 |
/* Identify shorter and longer input; copy the longer one */ |
| 267 |
if (a->nwords <= b->nwords) |
| 268 |
{ |
| 269 |
result = bms_copy(b); |
| 270 |
other = a; |
| 271 |
} |
| 272 |
else |
| 273 |
{ |
| 274 |
result = bms_copy(a); |
| 275 |
other = b; |
| 276 |
} |
| 277 |
/* And union the shorter input into the result */ |
| 278 |
otherlen = other->nwords; |
| 279 |
i = 0; |
| 280 |
do |
| 281 |
{ |
| 282 |
result->words[i] |= other->words[i]; |
| 283 |
} while (++i < otherlen); |
| 284 |
return result; |
| 285 |
} |
| 286 |
|
| 287 |
/* |
| 288 |
* bms_intersect - create and return a new set containing members which both |
| 289 |
* input sets have in common. Both inputs are left unmodified. |
| 290 |
*/ |
| 291 |
Bitmapset * |
| 292 |
bms_intersect(const Bitmapset *a, const Bitmapset *b) |
| 293 |
{ |
| 294 |
Bitmapset *result; |
| 295 |
const Bitmapset *other; |
| 296 |
int lastnonzero; |
| 297 |
int resultlen; |
| 298 |
int i; |
| 299 |
|
| 300 |
Assert(bms_is_valid_set(a)); |
| 301 |
Assert(bms_is_valid_set(b)); |
| 302 |
|
| 303 |
/* Handle cases where either input is NULL */ |
| 304 |
if (a == NULL || b == NULL) |
| 305 |
return NULL; |
| 306 |
|
| 307 |
/* Identify shorter and longer input; copy the shorter one */ |
| 308 |
if (a->nwords <= b->nwords) |
| 309 |
{ |
| 310 |
result = bms_copy(a); |
| 311 |
other = b; |
| 312 |
} |
| 313 |
else |
| 314 |
{ |
| 315 |
result = bms_copy(b); |
| 316 |
other = a; |
| 317 |
} |
| 318 |
/* And intersect the longer input with the result */ |
| 319 |
resultlen = result->nwords; |
| 320 |
lastnonzero = -1; |
| 321 |
i = 0; |
| 322 |
do |
| 323 |
{ |
| 324 |
result->words[i] &= other->words[i]; |
| 325 |
|
| 326 |
if (result->words[i] != 0) |
| 327 |
lastnonzero = i; |
| 328 |
} while (++i < resultlen); |
| 329 |
/* If we computed an empty result, we must return NULL */ |
| 330 |
if (lastnonzero == -1) |
| 331 |
{ |
| 332 |
pfree(result); |
| 333 |
return NULL; |
| 334 |
} |
| 335 |
|
| 336 |
/* get rid of trailing zero words */ |
| 337 |
result->nwords = lastnonzero + 1; |
| 338 |
return result; |
| 339 |
} |
| 340 |
|
| 341 |
/* |
| 342 |
* bms_difference - create and return a new set containing all the members of |
| 343 |
* 'a' without the members of 'b'. |
| 344 |
*/ |
| 345 |
Bitmapset * |
| 346 |
bms_difference(const Bitmapset *a, const Bitmapset *b) |
| 347 |
{ |
| 348 |
Bitmapset *result; |
| 349 |
int i; |
| 350 |
|
| 351 |
Assert(bms_is_valid_set(a)); |
| 352 |
Assert(bms_is_valid_set(b)); |
| 353 |
|
| 354 |
/* Handle cases where either input is NULL */ |
| 355 |
if (a == NULL) |
| 356 |
return NULL; |
| 357 |
if (b == NULL) |
| 358 |
return bms_copy(a); |
| 359 |
|
| 360 |
/* |
| 361 |
* In Postgres' usage, an empty result is a very common case, so it's |
| 362 |
* worth optimizing for that by testing bms_nonempty_difference(). This |
| 363 |
* saves us a palloc/pfree cycle compared to checking after-the-fact. |
| 364 |
*/ |
| 365 |
if (!bms_nonempty_difference(a, b)) |
| 366 |
return NULL; |
| 367 |
|
| 368 |
/* Copy the left input */ |
| 369 |
result = bms_copy(a); |
| 370 |
|
| 371 |
/* And remove b's bits from result */ |
| 372 |
if (result->nwords > b->nwords) |
| 373 |
{ |
| 374 |
/* |
| 375 |
* We'll never need to remove trailing zero words when 'a' has more |
| 376 |
* words than 'b' as the additional words must be non-zero. |
| 377 |
*/ |
| 378 |
i = 0; |
| 379 |
do |
| 380 |
{ |
| 381 |
result->words[i] &= ~b->words[i]; |
| 382 |
} while (++i < b->nwords); |
| 383 |
} |
| 384 |
else |
| 385 |
{ |
| 386 |
int lastnonzero = -1; |
| 387 |
|
| 388 |
/* we may need to remove trailing zero words from the result. */ |
| 389 |
i = 0; |
| 390 |
do |
| 391 |
{ |
| 392 |
result->words[i] &= ~b->words[i]; |
| 393 |
|
| 394 |
/* remember the last non-zero word */ |
| 395 |
if (result->words[i] != 0) |
| 396 |
lastnonzero = i; |
| 397 |
} while (++i < result->nwords); |
| 398 |
|
| 399 |
/* trim off trailing zero words */ |
| 400 |
result->nwords = lastnonzero + 1; |
| 401 |
} |
| 402 |
Assert(result->nwords != 0); |
| 403 |
|
| 404 |
/* Need not check for empty result, since we handled that case above */ |
| 405 |
return result; |
| 406 |
} |
| 407 |
|
| 408 |
/* |
| 409 |
* bms_is_subset - is A a subset of B? |
| 410 |
*/ |
| 411 |
bool |
| 412 |
bms_is_subset(const Bitmapset *a, const Bitmapset *b) |
| 413 |
{ |
| 414 |
int i; |
| 415 |
|
| 416 |
Assert(bms_is_valid_set(a)); |
| 417 |
Assert(bms_is_valid_set(b)); |
| 418 |
|
| 419 |
/* Handle cases where either input is NULL */ |
| 420 |
if (a == NULL) |
| 421 |
return true; /* empty set is a subset of anything */ |
| 422 |
if (b == NULL) |
| 423 |
return false; |
| 424 |
|
| 425 |
/* 'a' can't be a subset of 'b' if it contains more words */ |
| 426 |
if (a->nwords > b->nwords) |
| 427 |
return false; |
| 428 |
|
| 429 |
/* Check all 'a' members are set in 'b' */ |
| 430 |
i = 0; |
| 431 |
do |
| 432 |
{ |
| 433 |
if ((a->words[i] & ~b->words[i]) != 0) |
| 434 |
return false; |
| 435 |
} while (++i < a->nwords); |
| 436 |
return true; |
| 437 |
} |
| 438 |
|
| 439 |
/* |
| 440 |
* bms_subset_compare - compare A and B for equality/subset relationships |
| 441 |
* |
| 442 |
* This is more efficient than testing bms_is_subset in both directions. |
| 443 |
*/ |
| 444 |
BMS_Comparison |
| 445 |
bms_subset_compare(const Bitmapset *a, const Bitmapset *b) |
| 446 |
{ |
| 447 |
BMS_Comparison result; |
| 448 |
int shortlen; |
| 449 |
int i; |
| 450 |
|
| 451 |
Assert(bms_is_valid_set(a)); |
| 452 |
Assert(bms_is_valid_set(b)); |
| 453 |
|
| 454 |
/* Handle cases where either input is NULL */ |
| 455 |
if (a == NULL) |
| 456 |
{ |
| 457 |
if (b == NULL) |
| 458 |
return BMS_EQUAL; |
| 459 |
return BMS_SUBSET1; |
| 460 |
} |
| 461 |
if (b == NULL) |
| 462 |
return BMS_SUBSET2; |
| 463 |
|
| 464 |
/* Check common words */ |
| 465 |
result = BMS_EQUAL; /* status so far */ |
| 466 |
shortlen = Min(a->nwords, b->nwords); |
| 467 |
i = 0; |
| 468 |
do |
| 469 |
{ |
| 470 |
bitmapword aword = a->words[i]; |
| 471 |
bitmapword bword = b->words[i]; |
| 472 |
|
| 473 |
if ((aword & ~bword) != 0) |
| 474 |
{ |
| 475 |
/* a is not a subset of b */ |
| 476 |
if (result == BMS_SUBSET1) |
| 477 |
return BMS_DIFFERENT; |
| 478 |
result = BMS_SUBSET2; |
| 479 |
} |
| 480 |
if ((bword & ~aword) != 0) |
| 481 |
{ |
| 482 |
/* b is not a subset of a */ |
| 483 |
if (result == BMS_SUBSET2) |
| 484 |
return BMS_DIFFERENT; |
| 485 |
result = BMS_SUBSET1; |
| 486 |
} |
| 487 |
} while (++i < shortlen); |
| 488 |
/* Check extra words */ |
| 489 |
if (a->nwords > b->nwords) |
| 490 |
{ |
| 491 |
/* if a has more words then a is not a subset of b */ |
| 492 |
if (result == BMS_SUBSET1) |
| 493 |
return BMS_DIFFERENT; |
| 494 |
return BMS_SUBSET2; |
| 495 |
} |
| 496 |
else if (a->nwords < b->nwords) |
| 497 |
{ |
| 498 |
/* if b has more words then b is not a subset of a */ |
| 499 |
if (result == BMS_SUBSET2) |
| 500 |
return BMS_DIFFERENT; |
| 501 |
return BMS_SUBSET1; |
| 502 |
} |
| 503 |
return result; |
| 504 |
} |
| 505 |
|
| 506 |
/* |
| 507 |
* bms_is_member - is X a member of A? |
| 508 |
*/ |
| 509 |
bool |
| 510 |
bms_is_member(int x, const Bitmapset *a) |
| 511 |
{ |
| 512 |
int wordnum, |
| 513 |
bitnum; |
| 514 |
|
| 515 |
Assert(bms_is_valid_set(a)); |
| 516 |
|
| 517 |
/* XXX better to just return false for x<0 ? */ |
| 518 |
if (x < 0) |
| 519 |
elog(ERROR, "negative bitmapset member not allowed"); |
| 520 |
if (a == NULL) |
| 521 |
return false; |
| 522 |
|
| 523 |
wordnum = WORDNUM(x); |
| 524 |
bitnum = BITNUM(x); |
| 525 |
if (wordnum >= a->nwords) |
| 526 |
return false; |
| 527 |
if ((a->words[wordnum] & ((bitmapword) 1 << bitnum)) != 0) |
| 528 |
return true; |
| 529 |
return false; |
| 530 |
} |
| 531 |
|
| 532 |
/* |
| 533 |
* bms_member_index |
| 534 |
* determine 0-based index of member x in the bitmap |
| 535 |
* |
| 536 |
* Returns (-1) when x is not a member. |
| 537 |
*/ |
| 538 |
int |
| 539 |
bms_member_index(Bitmapset *a, int x) |
| 540 |
{ |
| 541 |
int i; |
| 542 |
int bitnum; |
| 543 |
int wordnum; |
| 544 |
int result = 0; |
| 545 |
bitmapword mask; |
| 546 |
|
| 547 |
Assert(bms_is_valid_set(a)); |
| 548 |
|
| 549 |
/* return -1 if not a member of the bitmap */ |
| 550 |
if (!bms_is_member(x, a)) |
| 551 |
return -1; |
| 552 |
|
| 553 |
wordnum = WORDNUM(x); |
| 554 |
bitnum = BITNUM(x); |
| 555 |
|
| 556 |
/* count bits in preceding words */ |
| 557 |
for (i = 0; i < wordnum; i++) |
| 558 |
{ |
| 559 |
bitmapword w = a->words[i]; |
| 560 |
|
| 561 |
/* No need to count the bits in a zero word */ |
| 562 |
if (w != 0) |
| 563 |
result += bmw_popcount(w); |
| 564 |
} |
| 565 |
|
| 566 |
/* |
| 567 |
* Now add bits of the last word, but only those before the item. We can |
| 568 |
* do that by applying a mask and then using popcount again. To get |
| 569 |
* 0-based index, we want to count only preceding bits, not the item |
| 570 |
* itself, so we subtract 1. |
| 571 |
*/ |
| 572 |
mask = ((bitmapword) 1 << bitnum) - 1; |
| 573 |
result += bmw_popcount(a->words[wordnum] & mask); |
| 574 |
|
| 575 |
return result; |
| 576 |
} |
| 577 |
|
| 578 |
/* |
| 579 |
* bms_overlap - do sets overlap (ie, have a nonempty intersection)? |
| 580 |
*/ |
| 581 |
bool |
| 582 |
bms_overlap(const Bitmapset *a, const Bitmapset *b) |
| 583 |
{ |
| 584 |
int shortlen; |
| 585 |
int i; |
| 586 |
|
| 587 |
Assert(bms_is_valid_set(a)); |
| 588 |
Assert(bms_is_valid_set(b)); |
| 589 |
|
| 590 |
/* Handle cases where either input is NULL */ |
| 591 |
if (a == NULL || b == NULL) |
| 592 |
return false; |
| 593 |
/* Check words in common */ |
| 594 |
shortlen = Min(a->nwords, b->nwords); |
| 595 |
i = 0; |
| 596 |
do |
| 597 |
{ |
| 598 |
if ((a->words[i] & b->words[i]) != 0) |
| 599 |
return true; |
| 600 |
} while (++i < shortlen); |
| 601 |
return false; |
| 602 |
} |
| 603 |
|
| 604 |
/* |
| 605 |
* bms_overlap_list - does a set overlap an integer list? |
| 606 |
*/ |
| 607 |
bool |
| 608 |
bms_overlap_list(const Bitmapset *a, const List *b) |
| 609 |
{ |
| 610 |
ListCell *lc; |
| 611 |
int wordnum, |
| 612 |
bitnum; |
| 613 |
|
| 614 |
Assert(bms_is_valid_set(a)); |
| 615 |
|
| 616 |
if (a == NULL || b == NIL) |
| 617 |
return false; |
| 618 |
|
| 619 |
foreach(lc, b) |
| 620 |
{ |
| 621 |
int x = lfirst_int(lc); |
| 622 |
|
| 623 |
if (x < 0) |
| 624 |
elog(ERROR, "negative bitmapset member not allowed"); |
| 625 |
wordnum = WORDNUM(x); |
| 626 |
bitnum = BITNUM(x); |
| 627 |
if (wordnum < a->nwords) |
| 628 |
if ((a->words[wordnum] & ((bitmapword) 1 << bitnum)) != 0) |
| 629 |
return true; |
| 630 |
} |
| 631 |
|
| 632 |
return false; |
| 633 |
} |
| 634 |
|
| 635 |
/* |
| 636 |
* bms_nonempty_difference - do sets have a nonempty difference? |
| 637 |
* |
| 638 |
* i.e., are any members set in 'a' that are not also set in 'b'. |
| 639 |
*/ |
| 640 |
bool |
| 641 |
bms_nonempty_difference(const Bitmapset *a, const Bitmapset *b) |
| 642 |
{ |
| 643 |
int i; |
| 644 |
|
| 645 |
Assert(bms_is_valid_set(a)); |
| 646 |
Assert(bms_is_valid_set(b)); |
| 647 |
|
| 648 |
/* Handle cases where either input is NULL */ |
| 649 |
if (a == NULL) |
| 650 |
return false; |
| 651 |
if (b == NULL) |
| 652 |
return true; |
| 653 |
/* if 'a' has more words then it must contain additional members */ |
| 654 |
if (a->nwords > b->nwords) |
| 655 |
return true; |
| 656 |
/* Check all 'a' members are set in 'b' */ |
| 657 |
i = 0; |
| 658 |
do |
| 659 |
{ |
| 660 |
if ((a->words[i] & ~b->words[i]) != 0) |
| 661 |
return true; |
| 662 |
} while (++i < a->nwords); |
| 663 |
return false; |
| 664 |
} |
| 665 |
|
| 666 |
/* |
| 667 |
* bms_singleton_member - return the sole integer member of set |
| 668 |
* |
| 669 |
* Raises error if |a| is not 1. |
| 670 |
*/ |
| 671 |
int |
| 672 |
bms_singleton_member(const Bitmapset *a) |
| 673 |
{ |
| 674 |
int result = -1; |
| 675 |
int nwords; |
| 676 |
int wordnum; |
| 677 |
|
| 678 |
Assert(bms_is_valid_set(a)); |
| 679 |
|
| 680 |
if (a == NULL) |
| 681 |
elog(ERROR, "bitmapset is empty"); |
| 682 |
|
| 683 |
nwords = a->nwords; |
| 684 |
wordnum = 0; |
| 685 |
do |
| 686 |
{ |
| 687 |
bitmapword w = a->words[wordnum]; |
| 688 |
|
| 689 |
if (w != 0) |
| 690 |
{ |
| 691 |
if (result >= 0 || HAS_MULTIPLE_ONES(w)) |
| 692 |
elog(ERROR, "bitmapset has multiple members"); |
| 693 |
result = wordnum * BITS_PER_BITMAPWORD; |
| 694 |
result += bmw_rightmost_one_pos(w); |
| 695 |
} |
| 696 |
} while (++wordnum < nwords); |
| 697 |
|
| 698 |
/* we don't expect non-NULL sets to be empty */ |
| 699 |
Assert(result >= 0); |
| 700 |
return result; |
| 701 |
} |
| 702 |
|
| 703 |
/* |
| 704 |
* bms_get_singleton_member |
| 705 |
* |
| 706 |
* Test whether the given set is a singleton. |
| 707 |
* If so, set *member to the value of its sole member, and return true. |
| 708 |
* If not, return false, without changing *member. |
| 709 |
* |
| 710 |
* This is more convenient and faster than calling bms_membership() and then |
| 711 |
* bms_singleton_member(), if we don't care about distinguishing empty sets |
| 712 |
* from multiple-member sets. |
| 713 |
*/ |
| 714 |
bool |
| 715 |
bms_get_singleton_member(const Bitmapset *a, int *member) |
| 716 |
{ |
| 717 |
int result = -1; |
| 718 |
int nwords; |
| 719 |
int wordnum; |
| 720 |
|
| 721 |
Assert(bms_is_valid_set(a)); |
| 722 |
|
| 723 |
if (a == NULL) |
| 724 |
return false; |
| 725 |
|
| 726 |
nwords = a->nwords; |
| 727 |
wordnum = 0; |
| 728 |
do |
| 729 |
{ |
| 730 |
bitmapword w = a->words[wordnum]; |
| 731 |
|
| 732 |
if (w != 0) |
| 733 |
{ |
| 734 |
if (result >= 0 || HAS_MULTIPLE_ONES(w)) |
| 735 |
return false; |
| 736 |
result = wordnum * BITS_PER_BITMAPWORD; |
| 737 |
result += bmw_rightmost_one_pos(w); |
| 738 |
} |
| 739 |
} while (++wordnum < nwords); |
| 740 |
|
| 741 |
/* we don't expect non-NULL sets to be empty */ |
| 742 |
Assert(result >= 0); |
| 743 |
*member = result; |
| 744 |
return true; |
| 745 |
} |
| 746 |
|
| 747 |
/* |
| 748 |
* bms_num_members - count members of set |
| 749 |
*/ |
| 750 |
int |
| 751 |
bms_num_members(const Bitmapset *a) |
| 752 |
{ |
| 753 |
int result = 0; |
| 754 |
int nwords; |
| 755 |
int wordnum; |
| 756 |
|
| 757 |
Assert(bms_is_valid_set(a)); |
| 758 |
|
| 759 |
if (a == NULL) |
| 760 |
return 0; |
| 761 |
|
| 762 |
nwords = a->nwords; |
| 763 |
wordnum = 0; |
| 764 |
do |
| 765 |
{ |
| 766 |
bitmapword w = a->words[wordnum]; |
| 767 |
|
| 768 |
/* No need to count the bits in a zero word */ |
| 769 |
if (w != 0) |
| 770 |
result += bmw_popcount(w); |
| 771 |
} while (++wordnum < nwords); |
| 772 |
return result; |
| 773 |
} |
| 774 |
|
| 775 |
/* |
| 776 |
* bms_membership - does a set have zero, one, or multiple members? |
| 777 |
* |
| 778 |
* This is faster than making an exact count with bms_num_members(). |
| 779 |
*/ |
| 780 |
BMS_Membership |
| 781 |
bms_membership(const Bitmapset *a) |
| 782 |
{ |
| 783 |
BMS_Membership result = BMS_EMPTY_SET; |
| 784 |
int nwords; |
| 785 |
int wordnum; |
| 786 |
|
| 787 |
Assert(bms_is_valid_set(a)); |
| 788 |
|
| 789 |
if (a == NULL) |
| 790 |
return BMS_EMPTY_SET; |
| 791 |
|
| 792 |
nwords = a->nwords; |
| 793 |
wordnum = 0; |
| 794 |
do |
| 795 |
{ |
| 796 |
bitmapword w = a->words[wordnum]; |
| 797 |
|
| 798 |
if (w != 0) |
| 799 |
{ |
| 800 |
if (result != BMS_EMPTY_SET || HAS_MULTIPLE_ONES(w)) |
| 801 |
return BMS_MULTIPLE; |
| 802 |
result = BMS_SINGLETON; |
| 803 |
} |
| 804 |
} while (++wordnum < nwords); |
| 805 |
return result; |
| 806 |
} |
| 807 |
|
| 808 |
|
| 809 |
/* |
| 810 |
* bms_add_member - add a specified member to set |
| 811 |
* |
| 812 |
* 'a' is recycled when possible. |
| 813 |
*/ |
| 814 |
Bitmapset * |
| 815 |
bms_add_member(Bitmapset *a, int x) |
| 816 |
{ |
| 817 |
int wordnum, |
| 818 |
bitnum; |
| 819 |
|
| 820 |
Assert(bms_is_valid_set(a)); |
| 821 |
|
| 822 |
if (x < 0) |
| 823 |
elog(ERROR, "negative bitmapset member not allowed"); |
| 824 |
if (a == NULL) |
| 825 |
return bms_make_singleton(x); |
| 826 |
|
| 827 |
wordnum = WORDNUM(x); |
| 828 |
bitnum = BITNUM(x); |
| 829 |
|
| 830 |
/* enlarge the set if necessary */ |
| 831 |
if (wordnum >= a->nwords) |
| 832 |
{ |
| 833 |
int oldnwords = a->nwords; |
| 834 |
int i; |
| 835 |
|
| 836 |
a = (Bitmapset *) repalloc(a, BITMAPSET_SIZE(wordnum + 1)); |
| 837 |
a->nwords = wordnum + 1; |
| 838 |
/* zero out the enlarged portion */ |
| 839 |
i = oldnwords; |
| 840 |
do |
| 841 |
{ |
| 842 |
a->words[i] = 0; |
| 843 |
} while (++i < a->nwords); |
| 844 |
} |
| 845 |
|
| 846 |
a->words[wordnum] |= ((bitmapword) 1 << bitnum); |
| 847 |
|
| 848 |
#ifdef REALLOCATE_BITMAPSETS |
| 849 |
|
| 850 |
/* |
| 851 |
* There's no guarantee that the repalloc returned a new pointer, so copy |
| 852 |
* and free unconditionally here. |
| 853 |
*/ |
| 854 |
a = bms_copy_and_free(a); |
| 855 |
#endif |
| 856 |
|
| 857 |
return a; |
| 858 |
} |
| 859 |
|
| 860 |
/* |
| 861 |
* bms_del_member - remove a specified member from set |
| 862 |
* |
| 863 |
* No error if x is not currently a member of set |
| 864 |
* |
| 865 |
* 'a' is recycled when possible. |
| 866 |
*/ |
| 867 |
Bitmapset * |
| 868 |
bms_del_member(Bitmapset *a, int x) |
| 869 |
{ |
| 870 |
int wordnum, |
| 871 |
bitnum; |
| 872 |
|
| 873 |
Assert(bms_is_valid_set(a)); |
| 874 |
|
| 875 |
if (x < 0) |
| 876 |
elog(ERROR, "negative bitmapset member not allowed"); |
| 877 |
if (a == NULL) |
| 878 |
return NULL; |
| 879 |
|
| 880 |
wordnum = WORDNUM(x); |
| 881 |
bitnum = BITNUM(x); |
| 882 |
|
| 883 |
#ifdef REALLOCATE_BITMAPSETS |
| 884 |
a = bms_copy_and_free(a); |
| 885 |
#endif |
| 886 |
|
| 887 |
/* member can't exist. Return 'a' unmodified */ |
| 888 |
if (unlikely(wordnum >= a->nwords)) |
| 889 |
return a; |
| 890 |
|
| 891 |
a->words[wordnum] &= ~((bitmapword) 1 << bitnum); |
| 892 |
|
| 893 |
/* when last word becomes empty, trim off all trailing empty words */ |
| 894 |
if (a->words[wordnum] == 0 && wordnum == a->nwords - 1) |
| 895 |
{ |
| 896 |
/* find the last non-empty word and make that the new final word */ |
| 897 |
for (int i = wordnum - 1; i >= 0; i--) |
| 898 |
{ |
| 899 |
if (a->words[i] != 0) |
| 900 |
{ |
| 901 |
a->nwords = i + 1; |
| 902 |
return a; |
| 903 |
} |
| 904 |
} |
| 905 |
|
| 906 |
/* the set is now empty */ |
| 907 |
pfree(a); |
| 908 |
return NULL; |
| 909 |
} |
| 910 |
return a; |
| 911 |
} |
| 912 |
|
| 913 |
/* |
| 914 |
* bms_add_members - like bms_union, but left input is recycled when possible |
| 915 |
*/ |
| 916 |
Bitmapset * |
| 917 |
bms_add_members(Bitmapset *a, const Bitmapset *b) |
| 918 |
{ |
| 919 |
Bitmapset *result; |
| 920 |
const Bitmapset *other; |
| 921 |
int otherlen; |
| 922 |
int i; |
| 923 |
|
| 924 |
Assert(bms_is_valid_set(a)); |
| 925 |
Assert(bms_is_valid_set(b)); |
| 926 |
|
| 927 |
/* Handle cases where either input is NULL */ |
| 928 |
if (a == NULL) |
| 929 |
return bms_copy(b); |
| 930 |
if (b == NULL) |
| 931 |
{ |
| 932 |
#ifdef REALLOCATE_BITMAPSETS |
| 933 |
a = bms_copy_and_free(a); |
| 934 |
#endif |
| 935 |
|
| 936 |
return a; |
| 937 |
} |
| 938 |
/* Identify shorter and longer input; copy the longer one if needed */ |
| 939 |
if (a->nwords < b->nwords) |
| 940 |
{ |
| 941 |
result = bms_copy(b); |
| 942 |
other = a; |
| 943 |
} |
| 944 |
else |
| 945 |
{ |
| 946 |
result = a; |
| 947 |
other = b; |
| 948 |
} |
| 949 |
/* And union the shorter input into the result */ |
| 950 |
otherlen = other->nwords; |
| 951 |
i = 0; |
| 952 |
do |
| 953 |
{ |
| 954 |
result->words[i] |= other->words[i]; |
| 955 |
} while (++i < otherlen); |
| 956 |
if (result != a) |
| 957 |
pfree(a); |
| 958 |
#ifdef REALLOCATE_BITMAPSETS |
| 959 |
else |
| 960 |
result = bms_copy_and_free(result); |
| 961 |
#endif |
| 962 |
|
| 963 |
return result; |
| 964 |
} |
| 965 |
|
| 966 |
/* |
| 967 |
* bms_replace_members |
| 968 |
* Remove all existing members from 'a' and repopulate the set with members |
| 969 |
* from 'b', recycling 'a', when possible. |
| 970 |
*/ |
| 971 |
Bitmapset * |
| 972 |
bms_replace_members(Bitmapset *a, const Bitmapset *b) |
| 973 |
{ |
| 974 |
int i; |
| 975 |
|
| 976 |
Assert(bms_is_valid_set(a)); |
| 977 |
Assert(bms_is_valid_set(b)); |
| 978 |
|
| 979 |
if (a == NULL) |
| 980 |
return bms_copy(b); |
| 981 |
if (b == NULL) |
| 982 |
{ |
| 983 |
pfree(a); |
| 984 |
return NULL; |
| 985 |
} |
| 986 |
|
| 987 |
if (a->nwords < b->nwords) |
| 988 |
a = (Bitmapset *) repalloc(a, BITMAPSET_SIZE(b->nwords)); |
| 989 |
|
| 990 |
i = 0; |
| 991 |
do |
| 992 |
{ |
| 993 |
a->words[i] = b->words[i]; |
| 994 |
} while (++i < b->nwords); |
| 995 |
|
| 996 |
a->nwords = b->nwords; |
| 997 |
|
| 998 |
#ifdef REALLOCATE_BITMAPSETS |
| 999 |
|
| 1000 |
/* |
| 1001 |
* There's no guarantee that the repalloc returned a new pointer, so copy |
| 1002 |
* and free unconditionally here. |
| 1003 |
*/ |
| 1004 |
a = bms_copy_and_free(a); |
| 1005 |
#endif |
| 1006 |
|
| 1007 |
return a; |
| 1008 |
} |
| 1009 |
|
| 1010 |
/* |
| 1011 |
* bms_add_range |
| 1012 |
* Add members in the range of 'lower' to 'upper' to the set. |
| 1013 |
* |
| 1014 |
* Note this could also be done by calling bms_add_member in a loop, however, |
| 1015 |
* using this function will be faster when the range is large as we work at |
| 1016 |
* the bitmapword level rather than at bit level. |
| 1017 |
*/ |
| 1018 |
Bitmapset * |
| 1019 |
bms_add_range(Bitmapset *a, int lower, int upper) |
| 1020 |
{ |
| 1021 |
int lwordnum, |
| 1022 |
lbitnum, |
| 1023 |
uwordnum, |
| 1024 |
ushiftbits, |
| 1025 |
wordnum; |
| 1026 |
|
| 1027 |
Assert(bms_is_valid_set(a)); |
| 1028 |
|
| 1029 |
/* do nothing if nothing is called for, without further checking */ |
| 1030 |
if (upper < lower) |
| 1031 |
{ |
| 1032 |
#ifdef REALLOCATE_BITMAPSETS |
| 1033 |
a = bms_copy_and_free(a); |
| 1034 |
#endif |
| 1035 |
|
| 1036 |
return a; |
| 1037 |
} |
| 1038 |
|
| 1039 |
if (lower < 0) |
| 1040 |
elog(ERROR, "negative bitmapset member not allowed"); |
| 1041 |
uwordnum = WORDNUM(upper); |
| 1042 |
|
| 1043 |
if (a == NULL) |
| 1044 |
{ |
| 1045 |
a = (Bitmapset *) palloc0(BITMAPSET_SIZE(uwordnum + 1)); |
| 1046 |
a->type = T_Bitmapset; |
| 1047 |
a->nwords = uwordnum + 1; |
| 1048 |
} |
| 1049 |
else if (uwordnum >= a->nwords) |
| 1050 |
{ |
| 1051 |
int oldnwords = a->nwords; |
| 1052 |
int i; |
| 1053 |
|
| 1054 |
/* ensure we have enough words to store the upper bit */ |
| 1055 |
a = (Bitmapset *) repalloc(a, BITMAPSET_SIZE(uwordnum + 1)); |
| 1056 |
a->nwords = uwordnum + 1; |
| 1057 |
/* zero out the enlarged portion */ |
| 1058 |
i = oldnwords; |
| 1059 |
do |
| 1060 |
{ |
| 1061 |
a->words[i] = 0; |
| 1062 |
} while (++i < a->nwords); |
| 1063 |
} |
| 1064 |
|
| 1065 |
wordnum = lwordnum = WORDNUM(lower); |
| 1066 |
|
| 1067 |
lbitnum = BITNUM(lower); |
| 1068 |
ushiftbits = BITS_PER_BITMAPWORD - (BITNUM(upper) + 1); |
| 1069 |
|
| 1070 |
/* |
| 1071 |
* Special case when lwordnum is the same as uwordnum we must perform the |
| 1072 |
* upper and lower masking on the word. |
| 1073 |
*/ |
| 1074 |
if (lwordnum == uwordnum) |
| 1075 |
{ |
| 1076 |
a->words[lwordnum] |= ~(bitmapword) (((bitmapword) 1 << lbitnum) - 1) |
| 1077 |
& (~(bitmapword) 0) >> ushiftbits; |
| 1078 |
} |
| 1079 |
else |
| 1080 |
{ |
| 1081 |
/* turn on lbitnum and all bits left of it */ |
| 1082 |
a->words[wordnum++] |= ~(bitmapword) (((bitmapword) 1 << lbitnum) - 1); |
| 1083 |
|
| 1084 |
/* turn on all bits for any intermediate words */ |
| 1085 |
while (wordnum < uwordnum) |
| 1086 |
a->words[wordnum++] = ~(bitmapword) 0; |
| 1087 |
|
| 1088 |
/* turn on upper's bit and all bits right of it. */ |
| 1089 |
a->words[uwordnum] |= (~(bitmapword) 0) >> ushiftbits; |
| 1090 |
} |
| 1091 |
|
| 1092 |
#ifdef REALLOCATE_BITMAPSETS |
| 1093 |
|
| 1094 |
/* |
| 1095 |
* There's no guarantee that the repalloc returned a new pointer, so copy |
| 1096 |
* and free unconditionally here. |
| 1097 |
*/ |
| 1098 |
a = bms_copy_and_free(a); |
| 1099 |
#endif |
| 1100 |
|
| 1101 |
return a; |
| 1102 |
} |
| 1103 |
|
| 1104 |
/* |
| 1105 |
* bms_int_members - like bms_intersect, but left input is recycled when |
| 1106 |
* possible |
| 1107 |
*/ |
| 1108 |
Bitmapset * |
| 1109 |
bms_int_members(Bitmapset *a, const Bitmapset *b) |
| 1110 |
{ |
| 1111 |
int lastnonzero; |
| 1112 |
int shortlen; |
| 1113 |
int i; |
| 1114 |
|
| 1115 |
Assert(bms_is_valid_set(a)); |
| 1116 |
Assert(bms_is_valid_set(b)); |
| 1117 |
|
| 1118 |
/* Handle cases where either input is NULL */ |
| 1119 |
if (a == NULL) |
| 1120 |
return NULL; |
| 1121 |
if (b == NULL) |
| 1122 |
{ |
| 1123 |
pfree(a); |
| 1124 |
return NULL; |
| 1125 |
} |
| 1126 |
|
| 1127 |
/* Intersect b into a; we need never copy */ |
| 1128 |
shortlen = Min(a->nwords, b->nwords); |
| 1129 |
lastnonzero = -1; |
| 1130 |
i = 0; |
| 1131 |
do |
| 1132 |
{ |
| 1133 |
a->words[i] &= b->words[i]; |
| 1134 |
|
| 1135 |
if (a->words[i] != 0) |
| 1136 |
lastnonzero = i; |
| 1137 |
} while (++i < shortlen); |
| 1138 |
|
| 1139 |
/* If we computed an empty result, we must return NULL */ |
| 1140 |
if (lastnonzero == -1) |
| 1141 |
{ |
| 1142 |
pfree(a); |
| 1143 |
return NULL; |
| 1144 |
} |
| 1145 |
|
| 1146 |
/* get rid of trailing zero words */ |
| 1147 |
a->nwords = lastnonzero + 1; |
| 1148 |
|
| 1149 |
#ifdef REALLOCATE_BITMAPSETS |
| 1150 |
a = bms_copy_and_free(a); |
| 1151 |
#endif |
| 1152 |
|
| 1153 |
return a; |
| 1154 |
} |
| 1155 |
|
| 1156 |
/* |
| 1157 |
* bms_del_members - delete members in 'a' that are set in 'b'. 'a' is |
| 1158 |
* recycled when possible. |
| 1159 |
*/ |
| 1160 |
Bitmapset * |
| 1161 |
bms_del_members(Bitmapset *a, const Bitmapset *b) |
| 1162 |
{ |
| 1163 |
int i; |
| 1164 |
|
| 1165 |
Assert(bms_is_valid_set(a)); |
| 1166 |
Assert(bms_is_valid_set(b)); |
| 1167 |
|
| 1168 |
/* Handle cases where either input is NULL */ |
| 1169 |
if (a == NULL) |
| 1170 |
return NULL; |
| 1171 |
if (b == NULL) |
| 1172 |
{ |
| 1173 |
#ifdef REALLOCATE_BITMAPSETS |
| 1174 |
a = bms_copy_and_free(a); |
| 1175 |
#endif |
| 1176 |
|
| 1177 |
return a; |
| 1178 |
} |
| 1179 |
|
| 1180 |
/* Remove b's bits from a; we need never copy */ |
| 1181 |
if (a->nwords > b->nwords) |
| 1182 |
{ |
| 1183 |
/* |
| 1184 |
* We'll never need to remove trailing zero words when 'a' has more |
| 1185 |
* words than 'b'. |
| 1186 |
*/ |
| 1187 |
i = 0; |
| 1188 |
do |
| 1189 |
{ |
| 1190 |
a->words[i] &= ~b->words[i]; |
| 1191 |
} while (++i < b->nwords); |
| 1192 |
} |
| 1193 |
else |
| 1194 |
{ |
| 1195 |
int lastnonzero = -1; |
| 1196 |
|
| 1197 |
/* we may need to remove trailing zero words from the result. */ |
| 1198 |
i = 0; |
| 1199 |
do |
| 1200 |
{ |
| 1201 |
a->words[i] &= ~b->words[i]; |
| 1202 |
|
| 1203 |
/* remember the last non-zero word */ |
| 1204 |
if (a->words[i] != 0) |
| 1205 |
lastnonzero = i; |
| 1206 |
} while (++i < a->nwords); |
| 1207 |
|
| 1208 |
/* check if 'a' has become empty */ |
| 1209 |
if (lastnonzero == -1) |
| 1210 |
{ |
| 1211 |
pfree(a); |
| 1212 |
return NULL; |
| 1213 |
} |
| 1214 |
|
| 1215 |
/* trim off any trailing zero words */ |
| 1216 |
a->nwords = lastnonzero + 1; |
| 1217 |
} |
| 1218 |
|
| 1219 |
#ifdef REALLOCATE_BITMAPSETS |
| 1220 |
a = bms_copy_and_free(a); |
| 1221 |
#endif |
| 1222 |
|
| 1223 |
return a; |
| 1224 |
} |
| 1225 |
|
| 1226 |
/* |
| 1227 |
* bms_join - like bms_union, but *either* input *may* be recycled |
| 1228 |
*/ |
| 1229 |
Bitmapset * |
| 1230 |
bms_join(Bitmapset *a, Bitmapset *b) |
| 1231 |
{ |
| 1232 |
Bitmapset *result; |
| 1233 |
Bitmapset *other; |
| 1234 |
int otherlen; |
| 1235 |
int i; |
| 1236 |
|
| 1237 |
Assert(bms_is_valid_set(a)); |
| 1238 |
Assert(bms_is_valid_set(b)); |
| 1239 |
|
| 1240 |
/* Handle cases where either input is NULL */ |
| 1241 |
if (a == NULL) |
| 1242 |
{ |
| 1243 |
#ifdef REALLOCATE_BITMAPSETS |
| 1244 |
b = bms_copy_and_free(b); |
| 1245 |
#endif |
| 1246 |
|
| 1247 |
return b; |
| 1248 |
} |
| 1249 |
if (b == NULL) |
| 1250 |
{ |
| 1251 |
#ifdef REALLOCATE_BITMAPSETS |
| 1252 |
a = bms_copy_and_free(a); |
| 1253 |
#endif |
| 1254 |
|
| 1255 |
return a; |
| 1256 |
} |
| 1257 |
|
| 1258 |
/* Identify shorter and longer input; use longer one as result */ |
| 1259 |
if (a->nwords < b->nwords) |
| 1260 |
{ |
| 1261 |
result = b; |
| 1262 |
other = a; |
| 1263 |
} |
| 1264 |
else |
| 1265 |
{ |
| 1266 |
result = a; |
| 1267 |
other = b; |
| 1268 |
} |
| 1269 |
/* And union the shorter input into the result */ |
| 1270 |
otherlen = other->nwords; |
| 1271 |
i = 0; |
| 1272 |
do |
| 1273 |
{ |
| 1274 |
result->words[i] |= other->words[i]; |
| 1275 |
} while (++i < otherlen); |
| 1276 |
if (other != result) /* pure paranoia */ |
| 1277 |
pfree(other); |
| 1278 |
|
| 1279 |
#ifdef REALLOCATE_BITMAPSETS |
| 1280 |
result = bms_copy_and_free(result); |
| 1281 |
#endif |
| 1282 |
|
| 1283 |
return result; |
| 1284 |
} |
| 1285 |
|
| 1286 |
/* |
| 1287 |
* bms_next_member - find next member of a set |
| 1288 |
* |
| 1289 |
* Returns smallest member greater than "prevbit", or -2 if there is none. |
| 1290 |
* "prevbit" must NOT be less than -1, or the behavior is unpredictable. |
| 1291 |
* |
| 1292 |
* This is intended as support for iterating through the members of a set. |
| 1293 |
* The typical pattern is |
| 1294 |
* |
| 1295 |
* x = -1; |
| 1296 |
* while ((x = bms_next_member(inputset, x)) >= 0) |
| 1297 |
* process member x; |
| 1298 |
* |
| 1299 |
* Notice that when there are no more members, we return -2, not -1 as you |
| 1300 |
* might expect. The rationale for that is to allow distinguishing the |
| 1301 |
* loop-not-started state (x == -1) from the loop-completed state (x == -2). |
| 1302 |
* It makes no difference in simple loop usage, but complex iteration logic |
| 1303 |
* might need such an ability. |
| 1304 |
*/ |
| 1305 |
int |
| 1306 |
bms_next_member(const Bitmapset *a, int prevbit) |
| 1307 |
{ |
| 1308 |
int nwords; |
| 1309 |
int wordnum; |
| 1310 |
bitmapword mask; |
| 1311 |
|
| 1312 |
Assert(bms_is_valid_set(a)); |
| 1313 |
|
| 1314 |
if (a == NULL) |
| 1315 |
return -2; |
| 1316 |
nwords = a->nwords; |
| 1317 |
prevbit++; |
| 1318 |
mask = (~(bitmapword) 0) << BITNUM(prevbit); |
| 1319 |
for (wordnum = WORDNUM(prevbit); wordnum < nwords; wordnum++) |
| 1320 |
{ |
| 1321 |
bitmapword w = a->words[wordnum]; |
| 1322 |
|
| 1323 |
/* ignore bits before prevbit */ |
| 1324 |
w &= mask; |
| 1325 |
|
| 1326 |
if (w != 0) |
| 1327 |
{ |
| 1328 |
int result; |
| 1329 |
|
| 1330 |
result = wordnum * BITS_PER_BITMAPWORD; |
| 1331 |
result += bmw_rightmost_one_pos(w); |
| 1332 |
return result; |
| 1333 |
} |
| 1334 |
|
| 1335 |
/* in subsequent words, consider all bits */ |
| 1336 |
mask = (~(bitmapword) 0); |
| 1337 |
} |
| 1338 |
return -2; |
| 1339 |
} |
| 1340 |
|
| 1341 |
/* |
| 1342 |
* bms_prev_member - find prev member of a set |
| 1343 |
* |
| 1344 |
* Returns largest member less than "prevbit", or -2 if there is none. |
| 1345 |
* "prevbit" must NOT be more than one above the highest possible bit that can |
| 1346 |
* be set at the Bitmapset at its current size. |
| 1347 |
* |
| 1348 |
* To ease finding the highest set bit for the initial loop, the special |
| 1349 |
* prevbit value of -1 can be passed to have the function find the highest |
| 1350 |
* valued member in the set. |
| 1351 |
* |
| 1352 |
* This is intended as support for iterating through the members of a set in |
| 1353 |
* reverse. The typical pattern is |
| 1354 |
* |
| 1355 |
* x = -1; |
| 1356 |
* while ((x = bms_prev_member(inputset, x)) >= 0) |
| 1357 |
* process member x; |
| 1358 |
* |
| 1359 |
* Notice that when there are no more members, we return -2, not -1 as you |
| 1360 |
* might expect. The rationale for that is to allow distinguishing the |
| 1361 |
* loop-not-started state (x == -1) from the loop-completed state (x == -2). |
| 1362 |
* It makes no difference in simple loop usage, but complex iteration logic |
| 1363 |
* might need such an ability. |
| 1364 |
*/ |
| 1365 |
|
| 1366 |
int |
| 1367 |
bms_prev_member(const Bitmapset *a, int prevbit) |
| 1368 |
{ |
| 1369 |
int wordnum; |
| 1370 |
int ushiftbits; |
| 1371 |
bitmapword mask; |
| 1372 |
|
| 1373 |
Assert(bms_is_valid_set(a)); |
| 1374 |
|
| 1375 |
/* |
| 1376 |
* If set is NULL or if there are no more bits to the right then we've |
| 1377 |
* nothing to do. |
| 1378 |
*/ |
| 1379 |
if (a == NULL || prevbit == 0) |
| 1380 |
return -2; |
| 1381 |
|
| 1382 |
/* transform -1 to the highest possible bit we could have set */ |
| 1383 |
if (prevbit == -1) |
| 1384 |
prevbit = a->nwords * BITS_PER_BITMAPWORD - 1; |
| 1385 |
else |
| 1386 |
prevbit--; |
| 1387 |
|
| 1388 |
ushiftbits = BITS_PER_BITMAPWORD - (BITNUM(prevbit) + 1); |
| 1389 |
mask = (~(bitmapword) 0) >> ushiftbits; |
| 1390 |
for (wordnum = WORDNUM(prevbit); wordnum >= 0; wordnum--) |
| 1391 |
{ |
| 1392 |
bitmapword w = a->words[wordnum]; |
| 1393 |
|
| 1394 |
/* mask out bits left of prevbit */ |
| 1395 |
w &= mask; |
| 1396 |
|
| 1397 |
if (w != 0) |
| 1398 |
{ |
| 1399 |
int result; |
| 1400 |
|
| 1401 |
result = wordnum * BITS_PER_BITMAPWORD; |
| 1402 |
result += bmw_leftmost_one_pos(w); |
| 1403 |
return result; |
| 1404 |
} |
| 1405 |
|
| 1406 |
/* in subsequent words, consider all bits */ |
| 1407 |
mask = (~(bitmapword) 0); |
| 1408 |
} |
| 1409 |
return -2; |
| 1410 |
} |
| 1411 |
|
| 1412 |
/* |
| 1413 |
* bms_hash_value - compute a hash key for a Bitmapset |
| 1414 |
*/ |
| 1415 |
uint32 |
| 1416 |
bms_hash_value(const Bitmapset *a) |
| 1417 |
{ |
| 1418 |
Assert(bms_is_valid_set(a)); |
| 1419 |
|
| 1420 |
if (a == NULL) |
| 1421 |
return 0; /* All empty sets hash to 0 */ |
| 1422 |
return DatumGetUInt32(hash_any((const unsigned char *) a->words, |
| 1423 |
a->nwords * sizeof(bitmapword))); |
| 1424 |
} |
| 1425 |
|
| 1426 |
/* |
| 1427 |
* bitmap_hash - hash function for keys that are (pointers to) Bitmapsets |
| 1428 |
* |
| 1429 |
* Note: don't forget to specify bitmap_match as the match function! |
| 1430 |
*/ |
| 1431 |
uint32 |
| 1432 |
bitmap_hash(const void *key, Size keysize) |
| 1433 |
{ |
| 1434 |
Assert(keysize == sizeof(Bitmapset *)); |
| 1435 |
return bms_hash_value(*((const Bitmapset *const *) key)); |
| 1436 |
} |
| 1437 |
|
| 1438 |
/* |
| 1439 |
* bitmap_match - match function to use with bitmap_hash |
| 1440 |
*/ |
| 1441 |
int |
| 1442 |
bitmap_match(const void *key1, const void *key2, Size keysize) |
| 1443 |
{ |
| 1444 |
Assert(keysize == sizeof(Bitmapset *)); |
| 1445 |
return !bms_equal(*((const Bitmapset *const *) key1), |
| 1446 |
*((const Bitmapset *const *) key2)); |
| 1447 |
} |