PostgreSQL Source Code  git master
hashfn.c
Go to the documentation of this file.
1 /*-------------------------------------------------------------------------
2  *
3  * hashfn.c
4  * Generic hashing functions, and hash functions for use in dynahash.c
5  * hashtables
6  *
7  *
8  * Portions Copyright (c) 1996-2019, PostgreSQL Global Development Group
9  * Portions Copyright (c) 1994, Regents of the University of California
10  *
11  *
12  * IDENTIFICATION
13  * src/backend/utils/hash/hashfn.c
14  *
15  * NOTES
16  * It is expected that every bit of a hash function's 32-bit result is
17  * as random as every other; failure to ensure this is likely to lead
18  * to poor performance of hash tables. In most cases a hash
19  * function should use hash_any() or its variant hash_uint32().
20  *
21  *-------------------------------------------------------------------------
22  */
23 #include "postgres.h"
24 
25 #include "fmgr.h"
26 #include "nodes/bitmapset.h"
27 #include "utils/hashutils.h"
28 #include "utils/hsearch.h"
29 
30 
31 /*
32  * This hash function was written by Bob Jenkins
33  * (bob_jenkins@burtleburtle.net), and superficially adapted
34  * for PostgreSQL by Neil Conway. For more information on this
35  * hash function, see http://burtleburtle.net/bob/hash/doobs.html,
36  * or Bob's article in Dr. Dobb's Journal, Sept. 1997.
37  *
38  * In the current code, we have adopted Bob's 2006 update of his hash
39  * function to fetch the data a word at a time when it is suitably aligned.
40  * This makes for a useful speedup, at the cost of having to maintain
41  * four code paths (aligned vs unaligned, and little-endian vs big-endian).
42  * It also uses two separate mixing functions mix() and final(), instead
43  * of a slower multi-purpose function.
44  */
45 
46 /* Get a bit mask of the bits set in non-uint32 aligned addresses */
47 #define UINT32_ALIGN_MASK (sizeof(uint32) - 1)
48 
49 /* Rotate a uint32 value left by k bits - note multiple evaluation! */
50 #define rot(x,k) (((x)<<(k)) | ((x)>>(32-(k))))
51 
52 /*----------
53  * mix -- mix 3 32-bit values reversibly.
54  *
55  * This is reversible, so any information in (a,b,c) before mix() is
56  * still in (a,b,c) after mix().
57  *
58  * If four pairs of (a,b,c) inputs are run through mix(), or through
59  * mix() in reverse, there are at least 32 bits of the output that
60  * are sometimes the same for one pair and different for another pair.
61  * This was tested for:
62  * * pairs that differed by one bit, by two bits, in any combination
63  * of top bits of (a,b,c), or in any combination of bottom bits of
64  * (a,b,c).
65  * * "differ" is defined as +, -, ^, or ~^. For + and -, I transformed
66  * the output delta to a Gray code (a^(a>>1)) so a string of 1's (as
67  * is commonly produced by subtraction) look like a single 1-bit
68  * difference.
69  * * the base values were pseudorandom, all zero but one bit set, or
70  * all zero plus a counter that starts at zero.
71  *
72  * This does not achieve avalanche. There are input bits of (a,b,c)
73  * that fail to affect some output bits of (a,b,c), especially of a. The
74  * most thoroughly mixed value is c, but it doesn't really even achieve
75  * avalanche in c.
76  *
77  * This allows some parallelism. Read-after-writes are good at doubling
78  * the number of bits affected, so the goal of mixing pulls in the opposite
79  * direction from the goal of parallelism. I did what I could. Rotates
80  * seem to cost as much as shifts on every machine I could lay my hands on,
81  * and rotates are much kinder to the top and bottom bits, so I used rotates.
82  *----------
83  */
84 #define mix(a,b,c) \
85 { \
86  a -= c; a ^= rot(c, 4); c += b; \
87  b -= a; b ^= rot(a, 6); a += c; \
88  c -= b; c ^= rot(b, 8); b += a; \
89  a -= c; a ^= rot(c,16); c += b; \
90  b -= a; b ^= rot(a,19); a += c; \
91  c -= b; c ^= rot(b, 4); b += a; \
92 }
93 
94 /*----------
95  * final -- final mixing of 3 32-bit values (a,b,c) into c
96  *
97  * Pairs of (a,b,c) values differing in only a few bits will usually
98  * produce values of c that look totally different. This was tested for
99  * * pairs that differed by one bit, by two bits, in any combination
100  * of top bits of (a,b,c), or in any combination of bottom bits of
101  * (a,b,c).
102  * * "differ" is defined as +, -, ^, or ~^. For + and -, I transformed
103  * the output delta to a Gray code (a^(a>>1)) so a string of 1's (as
104  * is commonly produced by subtraction) look like a single 1-bit
105  * difference.
106  * * the base values were pseudorandom, all zero but one bit set, or
107  * all zero plus a counter that starts at zero.
108  *
109  * The use of separate functions for mix() and final() allow for a
110  * substantial performance increase since final() does not need to
111  * do well in reverse, but is does need to affect all output bits.
112  * mix(), on the other hand, does not need to affect all output
113  * bits (affecting 32 bits is enough). The original hash function had
114  * a single mixing operation that had to satisfy both sets of requirements
115  * and was slower as a result.
116  *----------
117  */
118 #define final(a,b,c) \
119 { \
120  c ^= b; c -= rot(b,14); \
121  a ^= c; a -= rot(c,11); \
122  b ^= a; b -= rot(a,25); \
123  c ^= b; c -= rot(b,16); \
124  a ^= c; a -= rot(c, 4); \
125  b ^= a; b -= rot(a,14); \
126  c ^= b; c -= rot(b,24); \
127 }
128 
129 /*
130  * hash_any() -- hash a variable-length key into a 32-bit value
131  * k : the key (the unaligned variable-length array of bytes)
132  * len : the length of the key, counting by bytes
133  *
134  * Returns a uint32 value. Every bit of the key affects every bit of
135  * the return value. Every 1-bit and 2-bit delta achieves avalanche.
136  * About 6*len+35 instructions. The best hash table sizes are powers
137  * of 2. There is no need to do mod a prime (mod is sooo slow!).
138  * If you need less than 32 bits, use a bitmask.
139  *
140  * This procedure must never throw elog(ERROR); the ResourceOwner code
141  * relies on this not to fail.
142  *
143  * Note: we could easily change this function to return a 64-bit hash value
144  * by using the final values of both b and c. b is perhaps a little less
145  * well mixed than c, however.
146  */
147 Datum
148 hash_any(const unsigned char *k, int keylen)
149 {
150  uint32 a,
151  b,
152  c,
153  len;
154 
155  /* Set up the internal state */
156  len = keylen;
157  a = b = c = 0x9e3779b9 + len + 3923095;
158 
159  /* If the source pointer is word-aligned, we use word-wide fetches */
160  if (((uintptr_t) k & UINT32_ALIGN_MASK) == 0)
161  {
162  /* Code path for aligned source data */
163  const uint32 *ka = (const uint32 *) k;
164 
165  /* handle most of the key */
166  while (len >= 12)
167  {
168  a += ka[0];
169  b += ka[1];
170  c += ka[2];
171  mix(a, b, c);
172  ka += 3;
173  len -= 12;
174  }
175 
176  /* handle the last 11 bytes */
177  k = (const unsigned char *) ka;
178 #ifdef WORDS_BIGENDIAN
179  switch (len)
180  {
181  case 11:
182  c += ((uint32) k[10] << 8);
183  /* fall through */
184  case 10:
185  c += ((uint32) k[9] << 16);
186  /* fall through */
187  case 9:
188  c += ((uint32) k[8] << 24);
189  /* fall through */
190  case 8:
191  /* the lowest byte of c is reserved for the length */
192  b += ka[1];
193  a += ka[0];
194  break;
195  case 7:
196  b += ((uint32) k[6] << 8);
197  /* fall through */
198  case 6:
199  b += ((uint32) k[5] << 16);
200  /* fall through */
201  case 5:
202  b += ((uint32) k[4] << 24);
203  /* fall through */
204  case 4:
205  a += ka[0];
206  break;
207  case 3:
208  a += ((uint32) k[2] << 8);
209  /* fall through */
210  case 2:
211  a += ((uint32) k[1] << 16);
212  /* fall through */
213  case 1:
214  a += ((uint32) k[0] << 24);
215  /* case 0: nothing left to add */
216  }
217 #else /* !WORDS_BIGENDIAN */
218  switch (len)
219  {
220  case 11:
221  c += ((uint32) k[10] << 24);
222  /* fall through */
223  case 10:
224  c += ((uint32) k[9] << 16);
225  /* fall through */
226  case 9:
227  c += ((uint32) k[8] << 8);
228  /* fall through */
229  case 8:
230  /* the lowest byte of c is reserved for the length */
231  b += ka[1];
232  a += ka[0];
233  break;
234  case 7:
235  b += ((uint32) k[6] << 16);
236  /* fall through */
237  case 6:
238  b += ((uint32) k[5] << 8);
239  /* fall through */
240  case 5:
241  b += k[4];
242  /* fall through */
243  case 4:
244  a += ka[0];
245  break;
246  case 3:
247  a += ((uint32) k[2] << 16);
248  /* fall through */
249  case 2:
250  a += ((uint32) k[1] << 8);
251  /* fall through */
252  case 1:
253  a += k[0];
254  /* case 0: nothing left to add */
255  }
256 #endif /* WORDS_BIGENDIAN */
257  }
258  else
259  {
260  /* Code path for non-aligned source data */
261 
262  /* handle most of the key */
263  while (len >= 12)
264  {
265 #ifdef WORDS_BIGENDIAN
266  a += (k[3] + ((uint32) k[2] << 8) + ((uint32) k[1] << 16) + ((uint32) k[0] << 24));
267  b += (k[7] + ((uint32) k[6] << 8) + ((uint32) k[5] << 16) + ((uint32) k[4] << 24));
268  c += (k[11] + ((uint32) k[10] << 8) + ((uint32) k[9] << 16) + ((uint32) k[8] << 24));
269 #else /* !WORDS_BIGENDIAN */
270  a += (k[0] + ((uint32) k[1] << 8) + ((uint32) k[2] << 16) + ((uint32) k[3] << 24));
271  b += (k[4] + ((uint32) k[5] << 8) + ((uint32) k[6] << 16) + ((uint32) k[7] << 24));
272  c += (k[8] + ((uint32) k[9] << 8) + ((uint32) k[10] << 16) + ((uint32) k[11] << 24));
273 #endif /* WORDS_BIGENDIAN */
274  mix(a, b, c);
275  k += 12;
276  len -= 12;
277  }
278 
279  /* handle the last 11 bytes */
280 #ifdef WORDS_BIGENDIAN
281  switch (len)
282  {
283  case 11:
284  c += ((uint32) k[10] << 8);
285  /* fall through */
286  case 10:
287  c += ((uint32) k[9] << 16);
288  /* fall through */
289  case 9:
290  c += ((uint32) k[8] << 24);
291  /* fall through */
292  case 8:
293  /* the lowest byte of c is reserved for the length */
294  b += k[7];
295  /* fall through */
296  case 7:
297  b += ((uint32) k[6] << 8);
298  /* fall through */
299  case 6:
300  b += ((uint32) k[5] << 16);
301  /* fall through */
302  case 5:
303  b += ((uint32) k[4] << 24);
304  /* fall through */
305  case 4:
306  a += k[3];
307  /* fall through */
308  case 3:
309  a += ((uint32) k[2] << 8);
310  /* fall through */
311  case 2:
312  a += ((uint32) k[1] << 16);
313  /* fall through */
314  case 1:
315  a += ((uint32) k[0] << 24);
316  /* case 0: nothing left to add */
317  }
318 #else /* !WORDS_BIGENDIAN */
319  switch (len)
320  {
321  case 11:
322  c += ((uint32) k[10] << 24);
323  /* fall through */
324  case 10:
325  c += ((uint32) k[9] << 16);
326  /* fall through */
327  case 9:
328  c += ((uint32) k[8] << 8);
329  /* fall through */
330  case 8:
331  /* the lowest byte of c is reserved for the length */
332  b += ((uint32) k[7] << 24);
333  /* fall through */
334  case 7:
335  b += ((uint32) k[6] << 16);
336  /* fall through */
337  case 6:
338  b += ((uint32) k[5] << 8);
339  /* fall through */
340  case 5:
341  b += k[4];
342  /* fall through */
343  case 4:
344  a += ((uint32) k[3] << 24);
345  /* fall through */
346  case 3:
347  a += ((uint32) k[2] << 16);
348  /* fall through */
349  case 2:
350  a += ((uint32) k[1] << 8);
351  /* fall through */
352  case 1:
353  a += k[0];
354  /* case 0: nothing left to add */
355  }
356 #endif /* WORDS_BIGENDIAN */
357  }
358 
359  final(a, b, c);
360 
361  /* report the result */
362  return UInt32GetDatum(c);
363 }
364 
365 /*
366  * hash_any_extended() -- hash into a 64-bit value, using an optional seed
367  * k : the key (the unaligned variable-length array of bytes)
368  * len : the length of the key, counting by bytes
369  * seed : a 64-bit seed (0 means no seed)
370  *
371  * Returns a uint64 value. Otherwise similar to hash_any.
372  */
373 Datum
374 hash_any_extended(const unsigned char *k, int keylen,
375  uint64 seed)
376 {
377  uint32 a,
378  b,
379  c,
380  len;
381 
382  /* Set up the internal state */
383  len = keylen;
384  a = b = c = 0x9e3779b9 + len + 3923095;
385 
386  /* If the seed is non-zero, use it to perturb the internal state. */
387  if (seed != 0)
388  {
389  /*
390  * In essence, the seed is treated as part of the data being hashed,
391  * but for simplicity, we pretend that it's padded with four bytes of
392  * zeroes so that the seed constitutes a 12-byte chunk.
393  */
394  a += (uint32) (seed >> 32);
395  b += (uint32) seed;
396  mix(a, b, c);
397  }
398 
399  /* If the source pointer is word-aligned, we use word-wide fetches */
400  if (((uintptr_t) k & UINT32_ALIGN_MASK) == 0)
401  {
402  /* Code path for aligned source data */
403  const uint32 *ka = (const uint32 *) k;
404 
405  /* handle most of the key */
406  while (len >= 12)
407  {
408  a += ka[0];
409  b += ka[1];
410  c += ka[2];
411  mix(a, b, c);
412  ka += 3;
413  len -= 12;
414  }
415 
416  /* handle the last 11 bytes */
417  k = (const unsigned char *) ka;
418 #ifdef WORDS_BIGENDIAN
419  switch (len)
420  {
421  case 11:
422  c += ((uint32) k[10] << 8);
423  /* fall through */
424  case 10:
425  c += ((uint32) k[9] << 16);
426  /* fall through */
427  case 9:
428  c += ((uint32) k[8] << 24);
429  /* fall through */
430  case 8:
431  /* the lowest byte of c is reserved for the length */
432  b += ka[1];
433  a += ka[0];
434  break;
435  case 7:
436  b += ((uint32) k[6] << 8);
437  /* fall through */
438  case 6:
439  b += ((uint32) k[5] << 16);
440  /* fall through */
441  case 5:
442  b += ((uint32) k[4] << 24);
443  /* fall through */
444  case 4:
445  a += ka[0];
446  break;
447  case 3:
448  a += ((uint32) k[2] << 8);
449  /* fall through */
450  case 2:
451  a += ((uint32) k[1] << 16);
452  /* fall through */
453  case 1:
454  a += ((uint32) k[0] << 24);
455  /* case 0: nothing left to add */
456  }
457 #else /* !WORDS_BIGENDIAN */
458  switch (len)
459  {
460  case 11:
461  c += ((uint32) k[10] << 24);
462  /* fall through */
463  case 10:
464  c += ((uint32) k[9] << 16);
465  /* fall through */
466  case 9:
467  c += ((uint32) k[8] << 8);
468  /* fall through */
469  case 8:
470  /* the lowest byte of c is reserved for the length */
471  b += ka[1];
472  a += ka[0];
473  break;
474  case 7:
475  b += ((uint32) k[6] << 16);
476  /* fall through */
477  case 6:
478  b += ((uint32) k[5] << 8);
479  /* fall through */
480  case 5:
481  b += k[4];
482  /* fall through */
483  case 4:
484  a += ka[0];
485  break;
486  case 3:
487  a += ((uint32) k[2] << 16);
488  /* fall through */
489  case 2:
490  a += ((uint32) k[1] << 8);
491  /* fall through */
492  case 1:
493  a += k[0];
494  /* case 0: nothing left to add */
495  }
496 #endif /* WORDS_BIGENDIAN */
497  }
498  else
499  {
500  /* Code path for non-aligned source data */
501 
502  /* handle most of the key */
503  while (len >= 12)
504  {
505 #ifdef WORDS_BIGENDIAN
506  a += (k[3] + ((uint32) k[2] << 8) + ((uint32) k[1] << 16) + ((uint32) k[0] << 24));
507  b += (k[7] + ((uint32) k[6] << 8) + ((uint32) k[5] << 16) + ((uint32) k[4] << 24));
508  c += (k[11] + ((uint32) k[10] << 8) + ((uint32) k[9] << 16) + ((uint32) k[8] << 24));
509 #else /* !WORDS_BIGENDIAN */
510  a += (k[0] + ((uint32) k[1] << 8) + ((uint32) k[2] << 16) + ((uint32) k[3] << 24));
511  b += (k[4] + ((uint32) k[5] << 8) + ((uint32) k[6] << 16) + ((uint32) k[7] << 24));
512  c += (k[8] + ((uint32) k[9] << 8) + ((uint32) k[10] << 16) + ((uint32) k[11] << 24));
513 #endif /* WORDS_BIGENDIAN */
514  mix(a, b, c);
515  k += 12;
516  len -= 12;
517  }
518 
519  /* handle the last 11 bytes */
520 #ifdef WORDS_BIGENDIAN
521  switch (len)
522  {
523  case 11:
524  c += ((uint32) k[10] << 8);
525  /* fall through */
526  case 10:
527  c += ((uint32) k[9] << 16);
528  /* fall through */
529  case 9:
530  c += ((uint32) k[8] << 24);
531  /* fall through */
532  case 8:
533  /* the lowest byte of c is reserved for the length */
534  b += k[7];
535  /* fall through */
536  case 7:
537  b += ((uint32) k[6] << 8);
538  /* fall through */
539  case 6:
540  b += ((uint32) k[5] << 16);
541  /* fall through */
542  case 5:
543  b += ((uint32) k[4] << 24);
544  /* fall through */
545  case 4:
546  a += k[3];
547  /* fall through */
548  case 3:
549  a += ((uint32) k[2] << 8);
550  /* fall through */
551  case 2:
552  a += ((uint32) k[1] << 16);
553  /* fall through */
554  case 1:
555  a += ((uint32) k[0] << 24);
556  /* case 0: nothing left to add */
557  }
558 #else /* !WORDS_BIGENDIAN */
559  switch (len)
560  {
561  case 11:
562  c += ((uint32) k[10] << 24);
563  /* fall through */
564  case 10:
565  c += ((uint32) k[9] << 16);
566  /* fall through */
567  case 9:
568  c += ((uint32) k[8] << 8);
569  /* fall through */
570  case 8:
571  /* the lowest byte of c is reserved for the length */
572  b += ((uint32) k[7] << 24);
573  /* fall through */
574  case 7:
575  b += ((uint32) k[6] << 16);
576  /* fall through */
577  case 6:
578  b += ((uint32) k[5] << 8);
579  /* fall through */
580  case 5:
581  b += k[4];
582  /* fall through */
583  case 4:
584  a += ((uint32) k[3] << 24);
585  /* fall through */
586  case 3:
587  a += ((uint32) k[2] << 16);
588  /* fall through */
589  case 2:
590  a += ((uint32) k[1] << 8);
591  /* fall through */
592  case 1:
593  a += k[0];
594  /* case 0: nothing left to add */
595  }
596 #endif /* WORDS_BIGENDIAN */
597  }
598 
599  final(a, b, c);
600 
601  /* report the result */
602  PG_RETURN_UINT64(((uint64) b << 32) | c);
603 }
604 
605 /*
606  * hash_uint32() -- hash a 32-bit value to a 32-bit value
607  *
608  * This has the same result as
609  * hash_any(&k, sizeof(uint32))
610  * but is faster and doesn't force the caller to store k into memory.
611  */
612 Datum
613 hash_uint32(uint32 k)
614 {
615  uint32 a,
616  b,
617  c;
618 
619  a = b = c = 0x9e3779b9 + (uint32) sizeof(uint32) + 3923095;
620  a += k;
621 
622  final(a, b, c);
623 
624  /* report the result */
625  return UInt32GetDatum(c);
626 }
627 
628 /*
629  * hash_uint32_extended() -- hash a 32-bit value to a 64-bit value, with a seed
630  *
631  * Like hash_uint32, this is a convenience function.
632  */
633 Datum
634 hash_uint32_extended(uint32 k, uint64 seed)
635 {
636  uint32 a,
637  b,
638  c;
639 
640  a = b = c = 0x9e3779b9 + (uint32) sizeof(uint32) + 3923095;
641 
642  if (seed != 0)
643  {
644  a += (uint32) (seed >> 32);
645  b += (uint32) seed;
646  mix(a, b, c);
647  }
648 
649  a += k;
650 
651  final(a, b, c);
652 
653  /* report the result */
654  PG_RETURN_UINT64(((uint64) b << 32) | c);
655 }
656 
657 /*
658  * string_hash: hash function for keys that are NUL-terminated strings.
659  *
660  * NOTE: this is the default hash function if none is specified.
661  */
662 uint32
663 string_hash(const void *key, Size keysize)
664 {
665  /*
666  * If the string exceeds keysize-1 bytes, we want to hash only that many,
667  * because when it is copied into the hash table it will be truncated at
668  * that length.
669  */
670  Size s_len = strlen((const char *) key);
671 
672  s_len = Min(s_len, keysize - 1);
673  return DatumGetUInt32(hash_any((const unsigned char *) key,
674  (int) s_len));
675 }
676 
677 /*
678  * tag_hash: hash function for fixed-size tag values
679  */
680 uint32
681 tag_hash(const void *key, Size keysize)
682 {
683  return DatumGetUInt32(hash_any((const unsigned char *) key,
684  (int) keysize));
685 }
686 
687 /*
688  * uint32_hash: hash function for keys that are uint32 or int32
689  *
690  * (tag_hash works for this case too, but is slower)
691  */
692 uint32
693 uint32_hash(const void *key, Size keysize)
694 {
695  Assert(keysize == sizeof(uint32));
696  return DatumGetUInt32(hash_uint32(*((const uint32 *) key)));
697 }
698 
699 /*
700  * bitmap_hash: hash function for keys that are (pointers to) Bitmapsets
701  *
702  * Note: don't forget to specify bitmap_match as the match function!
703  */
704 uint32
705 bitmap_hash(const void *key, Size keysize)
706 {
707  Assert(keysize == sizeof(Bitmapset *));
708  return bms_hash_value(*((const Bitmapset *const *) key));
709 }
710 
711 /*
712  * bitmap_match: match function to use with bitmap_hash
713  */
714 int
715 bitmap_match(const void *key1, const void *key2, Size keysize)
716 {
717  Assert(keysize == sizeof(Bitmapset *));
718  return !bms_equal(*((const Bitmapset *const *) key1),
719  *((const Bitmapset *const *) key2));
720 }
DatumGetUInt32
#define DatumGetUInt32(X)
Definition: postgres.h:486
hash_any
Datum hash_any(const unsigned char *k, int keylen)
Definition: hashfn.c:148
string_hash
uint32 string_hash(const void *key, Size keysize)
Definition: hashfn.c:663
mix
#define mix(a, b, c)
Definition: hashfn.c:84
Min
#define Min(x, y)
Definition: c.h:911
hash_any_extended
Datum hash_any_extended(const unsigned char *k, int keylen, uint64 seed)
Definition: hashfn.c:374
PG_RETURN_UINT64
#define PG_RETURN_UINT64(x)
Definition: fmgr.h:358
uint32_hash
uint32 uint32_hash(const void *key, Size keysize)
Definition: hashfn.c:693
UINT32_ALIGN_MASK
#define UINT32_ALIGN_MASK
Definition: hashfn.c:47
bitmap_hash
uint32 bitmap_hash(const void *key, Size keysize)
Definition: hashfn.c:705
hash_uint32
Datum hash_uint32(uint32 k)
Definition: hashfn.c:613
bms_hash_value
uint32 bms_hash_value(const Bitmapset *a)
Definition: bitmapset.c:1154
c
char * c
Definition: preproc-cursor.c:31
uint32
unsigned int uint32
Definition: c.h:359
UInt32GetDatum
#define UInt32GetDatum(X)
Definition: postgres.h:493
bitmap_match
int bitmap_match(const void *key1, const void *key2, Size keysize)
Definition: hashfn.c:715
tag_hash
uint32 tag_hash(const void *key, Size keysize)
Definition: hashfn.c:681
Datum
uintptr_t Datum
Definition: postgres.h:367
Assert
#define Assert(condition)
Definition: c.h:739
Size
size_t Size
Definition: c.h:467
hash_uint32_extended
Datum hash_uint32_extended(uint32 k, uint64 seed)
Definition: hashfn.c:634
bms_equal
bool bms_equal(const Bitmapset *a, const Bitmapset *b)
Definition: bitmapset.c:94