1 | /* An expandable hash tables datatype. |
2 | Copyright (C) 1999-2024 Free Software Foundation, Inc. |
3 | Contributed by Vladimir Makarov (vmakarov@cygnus.com). |
4 | |
5 | This file is part of the libiberty library. |
6 | Libiberty is free software; you can redistribute it and/or |
7 | modify it under the terms of the GNU Library General Public |
8 | License as published by the Free Software Foundation; either |
9 | version 2 of the License, or (at your option) any later version. |
10 | |
11 | Libiberty 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 GNU |
14 | Library General Public License for more details. |
15 | |
16 | You should have received a copy of the GNU Library General Public |
17 | License along with libiberty; see the file COPYING.LIB. If |
18 | not, write to the Free Software Foundation, Inc., 51 Franklin Street - Fifth Floor, |
19 | Boston, MA 02110-1301, USA. */ |
20 | |
21 | /* This package implements basic hash table functionality. It is possible |
22 | to search for an entry, create an entry and destroy an entry. |
23 | |
24 | Elements in the table are generic pointers. |
25 | |
26 | The size of the table is not fixed; if the occupancy of the table |
27 | grows too high the hash table will be expanded. |
28 | |
29 | The abstract data implementation is based on generalized Algorithm D |
30 | from Knuth's book "The art of computer programming". Hash table is |
31 | expanded by creation of new hash table and transferring elements from |
32 | the old table to the new table. */ |
33 | |
34 | #ifdef HAVE_CONFIG_H |
35 | #include "config.h" |
36 | #endif |
37 | |
38 | #include <sys/types.h> |
39 | |
40 | #ifdef HAVE_STDLIB_H |
41 | #include <stdlib.h> |
42 | #endif |
43 | #ifdef HAVE_STRING_H |
44 | #include <string.h> |
45 | #endif |
46 | #ifdef HAVE_MALLOC_H |
47 | #include <malloc.h> |
48 | #endif |
49 | #ifdef HAVE_LIMITS_H |
50 | #include <limits.h> |
51 | #endif |
52 | #ifdef HAVE_INTTYPES_H |
53 | #include <inttypes.h> |
54 | #endif |
55 | #ifdef HAVE_STDINT_H |
56 | #include <stdint.h> |
57 | #endif |
58 | |
59 | #include <stdio.h> |
60 | |
61 | #include "libiberty.h" |
62 | #include "ansidecl.h" |
63 | #include "hashtab.h" |
64 | |
65 | #ifndef CHAR_BIT |
66 | #define CHAR_BIT 8 |
67 | #endif |
68 | |
69 | static unsigned int higher_prime_index (unsigned long); |
70 | static hashval_t htab_mod_1 (hashval_t, hashval_t, hashval_t, int); |
71 | static hashval_t htab_mod (hashval_t, htab_t); |
72 | static hashval_t htab_mod_m2 (hashval_t, htab_t); |
73 | static hashval_t hash_pointer (const void *); |
74 | static int eq_pointer (const void *, const void *); |
75 | static int htab_expand (htab_t); |
76 | static void **find_empty_slot_for_expand (htab_t, hashval_t); |
77 | |
78 | /* At some point, we could make these be NULL, and modify the |
79 | hash-table routines to handle NULL specially; that would avoid |
80 | function-call overhead for the common case of hashing pointers. */ |
81 | htab_hash htab_hash_pointer = hash_pointer; |
82 | htab_eq htab_eq_pointer = eq_pointer; |
83 | |
84 | /* Table of primes and multiplicative inverses. |
85 | |
86 | Note that these are not minimally reduced inverses. Unlike when generating |
87 | code to divide by a constant, we want to be able to use the same algorithm |
88 | all the time. All of these inverses (are implied to) have bit 32 set. |
89 | |
90 | For the record, here's the function that computed the table; it's a |
91 | vastly simplified version of the function of the same name from gcc. */ |
92 | |
93 | #if 0 |
94 | unsigned int |
95 | ceil_log2 (unsigned int x) |
96 | { |
97 | int i; |
98 | for (i = 31; i >= 0 ; --i) |
99 | if (x > (1u << i)) |
100 | return i+1; |
101 | abort (); |
102 | } |
103 | |
104 | unsigned int |
105 | choose_multiplier (unsigned int d, unsigned int *mlp, unsigned char *shiftp) |
106 | { |
107 | unsigned long long mhigh; |
108 | double nx; |
109 | int lgup, post_shift; |
110 | int pow, pow2; |
111 | int n = 32, precision = 32; |
112 | |
113 | lgup = ceil_log2 (d); |
114 | pow = n + lgup; |
115 | pow2 = n + lgup - precision; |
116 | |
117 | nx = ldexp (1.0, pow) + ldexp (1.0, pow2); |
118 | mhigh = nx / d; |
119 | |
120 | *shiftp = lgup - 1; |
121 | *mlp = mhigh; |
122 | return mhigh >> 32; |
123 | } |
124 | #endif |
125 | |
126 | struct prime_ent |
127 | { |
128 | hashval_t prime; |
129 | hashval_t inv; |
130 | hashval_t inv_m2; /* inverse of prime-2 */ |
131 | hashval_t shift; |
132 | }; |
133 | |
134 | static struct prime_ent const prime_tab[] = { |
135 | { .prime: 7, .inv: 0x24924925, .inv_m2: 0x9999999b, .shift: 2 }, |
136 | { .prime: 13, .inv: 0x3b13b13c, .inv_m2: 0x745d1747, .shift: 3 }, |
137 | { .prime: 31, .inv: 0x08421085, .inv_m2: 0x1a7b9612, .shift: 4 }, |
138 | { .prime: 61, .inv: 0x0c9714fc, .inv_m2: 0x15b1e5f8, .shift: 5 }, |
139 | { .prime: 127, .inv: 0x02040811, .inv_m2: 0x0624dd30, .shift: 6 }, |
140 | { .prime: 251, .inv: 0x05197f7e, .inv_m2: 0x073260a5, .shift: 7 }, |
141 | { .prime: 509, .inv: 0x01824366, .inv_m2: 0x02864fc8, .shift: 8 }, |
142 | { .prime: 1021, .inv: 0x00c0906d, .inv_m2: 0x014191f7, .shift: 9 }, |
143 | { .prime: 2039, .inv: 0x0121456f, .inv_m2: 0x0161e69e, .shift: 10 }, |
144 | { .prime: 4093, .inv: 0x00300902, .inv_m2: 0x00501908, .shift: 11 }, |
145 | { .prime: 8191, .inv: 0x00080041, .inv_m2: 0x00180241, .shift: 12 }, |
146 | { .prime: 16381, .inv: 0x000c0091, .inv_m2: 0x00140191, .shift: 13 }, |
147 | { .prime: 32749, .inv: 0x002605a5, .inv_m2: 0x002a06e6, .shift: 14 }, |
148 | { .prime: 65521, .inv: 0x000f00e2, .inv_m2: 0x00110122, .shift: 15 }, |
149 | { .prime: 131071, .inv: 0x00008001, .inv_m2: 0x00018003, .shift: 16 }, |
150 | { .prime: 262139, .inv: 0x00014002, .inv_m2: 0x0001c004, .shift: 17 }, |
151 | { .prime: 524287, .inv: 0x00002001, .inv_m2: 0x00006001, .shift: 18 }, |
152 | { .prime: 1048573, .inv: 0x00003001, .inv_m2: 0x00005001, .shift: 19 }, |
153 | { .prime: 2097143, .inv: 0x00004801, .inv_m2: 0x00005801, .shift: 20 }, |
154 | { .prime: 4194301, .inv: 0x00000c01, .inv_m2: 0x00001401, .shift: 21 }, |
155 | { .prime: 8388593, .inv: 0x00001e01, .inv_m2: 0x00002201, .shift: 22 }, |
156 | { .prime: 16777213, .inv: 0x00000301, .inv_m2: 0x00000501, .shift: 23 }, |
157 | { .prime: 33554393, .inv: 0x00001381, .inv_m2: 0x00001481, .shift: 24 }, |
158 | { .prime: 67108859, .inv: 0x00000141, .inv_m2: 0x000001c1, .shift: 25 }, |
159 | { .prime: 134217689, .inv: 0x000004e1, .inv_m2: 0x00000521, .shift: 26 }, |
160 | { .prime: 268435399, .inv: 0x00000391, .inv_m2: 0x000003b1, .shift: 27 }, |
161 | { .prime: 536870909, .inv: 0x00000019, .inv_m2: 0x00000029, .shift: 28 }, |
162 | { .prime: 1073741789, .inv: 0x0000008d, .inv_m2: 0x00000095, .shift: 29 }, |
163 | { .prime: 2147483647, .inv: 0x00000003, .inv_m2: 0x00000007, .shift: 30 }, |
164 | /* Avoid "decimal constant so large it is unsigned" for 4294967291. */ |
165 | { .prime: 0xfffffffb, .inv: 0x00000006, .inv_m2: 0x00000008, .shift: 31 } |
166 | }; |
167 | |
168 | /* The following function returns an index into the above table of the |
169 | nearest prime number which is greater than N, and near a power of two. */ |
170 | |
171 | static unsigned int |
172 | higher_prime_index (unsigned long n) |
173 | { |
174 | unsigned int low = 0; |
175 | unsigned int high = sizeof(prime_tab) / sizeof(prime_tab[0]); |
176 | |
177 | while (low != high) |
178 | { |
179 | unsigned int mid = low + (high - low) / 2; |
180 | if (n > prime_tab[mid].prime) |
181 | low = mid + 1; |
182 | else |
183 | high = mid; |
184 | } |
185 | |
186 | /* If we've run out of primes, abort. */ |
187 | if (n > prime_tab[low].prime) |
188 | { |
189 | fprintf (stderr, format: "Cannot find prime bigger than %lu\n" , n); |
190 | abort (); |
191 | } |
192 | |
193 | return low; |
194 | } |
195 | |
196 | /* Returns non-zero if P1 and P2 are equal. */ |
197 | |
198 | static int |
199 | eq_pointer (const void *p1, const void *p2) |
200 | { |
201 | return p1 == p2; |
202 | } |
203 | |
204 | |
205 | /* The parens around the function names in the next two definitions |
206 | are essential in order to prevent macro expansions of the name. |
207 | The bodies, however, are expanded as expected, so they are not |
208 | recursive definitions. */ |
209 | |
210 | /* Return the current size of given hash table. */ |
211 | |
212 | #define htab_size(htab) ((htab)->size) |
213 | |
214 | size_t |
215 | (htab_size) (htab_t htab) |
216 | { |
217 | return htab_size (htab); |
218 | } |
219 | |
220 | /* Return the current number of elements in given hash table. */ |
221 | |
222 | #define htab_elements(htab) ((htab)->n_elements - (htab)->n_deleted) |
223 | |
224 | size_t |
225 | (htab_elements) (htab_t htab) |
226 | { |
227 | return htab_elements (htab); |
228 | } |
229 | |
230 | /* Return X % Y. */ |
231 | |
232 | static inline hashval_t |
233 | htab_mod_1 (hashval_t x, hashval_t y, hashval_t inv, int shift) |
234 | { |
235 | /* The multiplicative inverses computed above are for 32-bit types, and |
236 | requires that we be able to compute a highpart multiply. */ |
237 | #ifdef UNSIGNED_64BIT_TYPE |
238 | __extension__ typedef UNSIGNED_64BIT_TYPE ull; |
239 | if (sizeof (hashval_t) * CHAR_BIT <= 32) |
240 | { |
241 | hashval_t t1, t2, t3, t4, q, r; |
242 | |
243 | t1 = ((ull)x * inv) >> 32; |
244 | t2 = x - t1; |
245 | t3 = t2 >> 1; |
246 | t4 = t1 + t3; |
247 | q = t4 >> shift; |
248 | r = x - (q * y); |
249 | |
250 | return r; |
251 | } |
252 | #endif |
253 | |
254 | /* Otherwise just use the native division routines. */ |
255 | return x % y; |
256 | } |
257 | |
258 | /* Compute the primary hash for HASH given HTAB's current size. */ |
259 | |
260 | static inline hashval_t |
261 | htab_mod (hashval_t hash, htab_t htab) |
262 | { |
263 | const struct prime_ent *p = &prime_tab[htab->size_prime_index]; |
264 | return htab_mod_1 (x: hash, y: p->prime, inv: p->inv, shift: p->shift); |
265 | } |
266 | |
267 | /* Compute the secondary hash for HASH given HTAB's current size. */ |
268 | |
269 | static inline hashval_t |
270 | htab_mod_m2 (hashval_t hash, htab_t htab) |
271 | { |
272 | const struct prime_ent *p = &prime_tab[htab->size_prime_index]; |
273 | return 1 + htab_mod_1 (x: hash, y: p->prime - 2, inv: p->inv_m2, shift: p->shift); |
274 | } |
275 | |
276 | /* This function creates table with length slightly longer than given |
277 | source length. Created hash table is initiated as empty (all the |
278 | hash table entries are HTAB_EMPTY_ENTRY). The function returns the |
279 | created hash table, or NULL if memory allocation fails. */ |
280 | |
281 | htab_t |
282 | htab_create_alloc (size_t size, htab_hash hash_f, htab_eq eq_f, |
283 | htab_del del_f, htab_alloc alloc_f, htab_free free_f) |
284 | { |
285 | return htab_create_typed_alloc (size, hash_f, eq_f, del_f, alloc_f, alloc_f, |
286 | free_f); |
287 | } |
288 | |
289 | /* As above, but uses the variants of ALLOC_F and FREE_F which accept |
290 | an extra argument. */ |
291 | |
292 | htab_t |
293 | htab_create_alloc_ex (size_t size, htab_hash hash_f, htab_eq eq_f, |
294 | htab_del del_f, void *alloc_arg, |
295 | htab_alloc_with_arg alloc_f, |
296 | htab_free_with_arg free_f) |
297 | { |
298 | htab_t result; |
299 | unsigned int size_prime_index; |
300 | |
301 | size_prime_index = higher_prime_index (n: size); |
302 | size = prime_tab[size_prime_index].prime; |
303 | |
304 | result = (htab_t) (*alloc_f) (alloc_arg, 1, sizeof (struct htab)); |
305 | if (result == NULL) |
306 | return NULL; |
307 | result->entries = (void **) (*alloc_f) (alloc_arg, size, sizeof (void *)); |
308 | if (result->entries == NULL) |
309 | { |
310 | if (free_f != NULL) |
311 | (*free_f) (alloc_arg, result); |
312 | return NULL; |
313 | } |
314 | result->size = size; |
315 | result->size_prime_index = size_prime_index; |
316 | result->hash_f = hash_f; |
317 | result->eq_f = eq_f; |
318 | result->del_f = del_f; |
319 | result->alloc_arg = alloc_arg; |
320 | result->alloc_with_arg_f = alloc_f; |
321 | result->free_with_arg_f = free_f; |
322 | return result; |
323 | } |
324 | |
325 | /* |
326 | |
327 | @deftypefn Supplemental htab_t htab_create_typed_alloc (size_t @var{size}, @ |
328 | htab_hash @var{hash_f}, htab_eq @var{eq_f}, htab_del @var{del_f}, @ |
329 | htab_alloc @var{alloc_tab_f}, htab_alloc @var{alloc_f}, @ |
330 | htab_free @var{free_f}) |
331 | |
332 | This function creates a hash table that uses two different allocators |
333 | @var{alloc_tab_f} and @var{alloc_f} to use for allocating the table itself |
334 | and its entries respectively. This is useful when variables of different |
335 | types need to be allocated with different allocators. |
336 | |
337 | The created hash table is slightly larger than @var{size} and it is |
338 | initially empty (all the hash table entries are @code{HTAB_EMPTY_ENTRY}). |
339 | The function returns the created hash table, or @code{NULL} if memory |
340 | allocation fails. |
341 | |
342 | @end deftypefn |
343 | |
344 | */ |
345 | |
346 | htab_t |
347 | htab_create_typed_alloc (size_t size, htab_hash hash_f, htab_eq eq_f, |
348 | htab_del del_f, htab_alloc alloc_tab_f, |
349 | htab_alloc alloc_f, htab_free free_f) |
350 | { |
351 | htab_t result; |
352 | unsigned int size_prime_index; |
353 | |
354 | size_prime_index = higher_prime_index (n: size); |
355 | size = prime_tab[size_prime_index].prime; |
356 | |
357 | result = (htab_t) (*alloc_tab_f) (1, sizeof (struct htab)); |
358 | if (result == NULL) |
359 | return NULL; |
360 | result->entries = (void **) (*alloc_f) (size, sizeof (void *)); |
361 | if (result->entries == NULL) |
362 | { |
363 | if (free_f != NULL) |
364 | (*free_f) (result); |
365 | return NULL; |
366 | } |
367 | result->size = size; |
368 | result->size_prime_index = size_prime_index; |
369 | result->hash_f = hash_f; |
370 | result->eq_f = eq_f; |
371 | result->del_f = del_f; |
372 | result->alloc_f = alloc_f; |
373 | result->free_f = free_f; |
374 | return result; |
375 | } |
376 | |
377 | |
378 | /* Update the function pointers and allocation parameter in the htab_t. */ |
379 | |
380 | void |
381 | htab_set_functions_ex (htab_t htab, htab_hash hash_f, htab_eq eq_f, |
382 | htab_del del_f, void *alloc_arg, |
383 | htab_alloc_with_arg alloc_f, htab_free_with_arg free_f) |
384 | { |
385 | htab->hash_f = hash_f; |
386 | htab->eq_f = eq_f; |
387 | htab->del_f = del_f; |
388 | htab->alloc_arg = alloc_arg; |
389 | htab->alloc_with_arg_f = alloc_f; |
390 | htab->free_with_arg_f = free_f; |
391 | } |
392 | |
393 | /* These functions exist solely for backward compatibility. */ |
394 | |
395 | #undef htab_create |
396 | htab_t |
397 | htab_create (size_t size, htab_hash hash_f, htab_eq eq_f, htab_del del_f) |
398 | { |
399 | return htab_create_alloc (size, hash_f, eq_f, del_f, alloc_f: xcalloc, free_f: free); |
400 | } |
401 | |
402 | htab_t |
403 | htab_try_create (size_t size, htab_hash hash_f, htab_eq eq_f, htab_del del_f) |
404 | { |
405 | return htab_create_alloc (size, hash_f, eq_f, del_f, alloc_f: calloc, free_f: free); |
406 | } |
407 | |
408 | /* This function frees all memory allocated for given hash table. |
409 | Naturally the hash table must already exist. */ |
410 | |
411 | void |
412 | htab_delete (htab_t htab) |
413 | { |
414 | size_t size = htab_size (htab); |
415 | void **entries = htab->entries; |
416 | int i; |
417 | |
418 | if (htab->del_f) |
419 | for (i = size - 1; i >= 0; i--) |
420 | if (entries[i] != HTAB_EMPTY_ENTRY && entries[i] != HTAB_DELETED_ENTRY) |
421 | (*htab->del_f) (entries[i]); |
422 | |
423 | if (htab->free_f != NULL) |
424 | { |
425 | (*htab->free_f) (entries); |
426 | (*htab->free_f) (htab); |
427 | } |
428 | else if (htab->free_with_arg_f != NULL) |
429 | { |
430 | (*htab->free_with_arg_f) (htab->alloc_arg, entries); |
431 | (*htab->free_with_arg_f) (htab->alloc_arg, htab); |
432 | } |
433 | } |
434 | |
435 | /* This function clears all entries in the given hash table. */ |
436 | |
437 | void |
438 | htab_empty (htab_t htab) |
439 | { |
440 | size_t size = htab_size (htab); |
441 | void **entries = htab->entries; |
442 | int i; |
443 | |
444 | if (htab->del_f) |
445 | for (i = size - 1; i >= 0; i--) |
446 | if (entries[i] != HTAB_EMPTY_ENTRY && entries[i] != HTAB_DELETED_ENTRY) |
447 | (*htab->del_f) (entries[i]); |
448 | |
449 | /* Instead of clearing megabyte, downsize the table. */ |
450 | if (size > 1024*1024 / sizeof (void *)) |
451 | { |
452 | int nindex = higher_prime_index (n: 1024 / sizeof (void *)); |
453 | int nsize = prime_tab[nindex].prime; |
454 | |
455 | if (htab->free_f != NULL) |
456 | (*htab->free_f) (htab->entries); |
457 | else if (htab->free_with_arg_f != NULL) |
458 | (*htab->free_with_arg_f) (htab->alloc_arg, htab->entries); |
459 | if (htab->alloc_with_arg_f != NULL) |
460 | htab->entries = (void **) (*htab->alloc_with_arg_f) (htab->alloc_arg, nsize, |
461 | sizeof (void *)); |
462 | else |
463 | htab->entries = (void **) (*htab->alloc_f) (nsize, sizeof (void *)); |
464 | htab->size = nsize; |
465 | htab->size_prime_index = nindex; |
466 | } |
467 | else |
468 | memset (s: entries, c: 0, n: size * sizeof (void *)); |
469 | htab->n_deleted = 0; |
470 | htab->n_elements = 0; |
471 | } |
472 | |
473 | /* Similar to htab_find_slot, but without several unwanted side effects: |
474 | - Does not call htab->eq_f when it finds an existing entry. |
475 | - Does not change the count of elements/searches/collisions in the |
476 | hash table. |
477 | This function also assumes there are no deleted entries in the table. |
478 | HASH is the hash value for the element to be inserted. */ |
479 | |
480 | static void ** |
481 | find_empty_slot_for_expand (htab_t htab, hashval_t hash) |
482 | { |
483 | hashval_t index = htab_mod (hash, htab); |
484 | size_t size = htab_size (htab); |
485 | void **slot = htab->entries + index; |
486 | hashval_t hash2; |
487 | |
488 | if (*slot == HTAB_EMPTY_ENTRY) |
489 | return slot; |
490 | else if (*slot == HTAB_DELETED_ENTRY) |
491 | abort (); |
492 | |
493 | hash2 = htab_mod_m2 (hash, htab); |
494 | for (;;) |
495 | { |
496 | index += hash2; |
497 | if (index >= size) |
498 | index -= size; |
499 | |
500 | slot = htab->entries + index; |
501 | if (*slot == HTAB_EMPTY_ENTRY) |
502 | return slot; |
503 | else if (*slot == HTAB_DELETED_ENTRY) |
504 | abort (); |
505 | } |
506 | } |
507 | |
508 | /* The following function changes size of memory allocated for the |
509 | entries and repeatedly inserts the table elements. The occupancy |
510 | of the table after the call will be about 50%. Naturally the hash |
511 | table must already exist. Remember also that the place of the |
512 | table entries is changed. If memory allocation failures are allowed, |
513 | this function will return zero, indicating that the table could not be |
514 | expanded. If all goes well, it will return a non-zero value. */ |
515 | |
516 | static int |
517 | htab_expand (htab_t htab) |
518 | { |
519 | void **oentries; |
520 | void **olimit; |
521 | void **p; |
522 | void **nentries; |
523 | size_t nsize, osize, elts; |
524 | unsigned int oindex, nindex; |
525 | |
526 | oentries = htab->entries; |
527 | oindex = htab->size_prime_index; |
528 | osize = htab->size; |
529 | olimit = oentries + osize; |
530 | elts = htab_elements (htab); |
531 | |
532 | /* Resize only when table after removal of unused elements is either |
533 | too full or too empty. */ |
534 | if (elts * 2 > osize || (elts * 8 < osize && osize > 32)) |
535 | { |
536 | nindex = higher_prime_index (n: elts * 2); |
537 | nsize = prime_tab[nindex].prime; |
538 | } |
539 | else |
540 | { |
541 | nindex = oindex; |
542 | nsize = osize; |
543 | } |
544 | |
545 | if (htab->alloc_with_arg_f != NULL) |
546 | nentries = (void **) (*htab->alloc_with_arg_f) (htab->alloc_arg, nsize, |
547 | sizeof (void *)); |
548 | else |
549 | nentries = (void **) (*htab->alloc_f) (nsize, sizeof (void *)); |
550 | if (nentries == NULL) |
551 | return 0; |
552 | htab->entries = nentries; |
553 | htab->size = nsize; |
554 | htab->size_prime_index = nindex; |
555 | htab->n_elements -= htab->n_deleted; |
556 | htab->n_deleted = 0; |
557 | |
558 | p = oentries; |
559 | do |
560 | { |
561 | void *x = *p; |
562 | |
563 | if (x != HTAB_EMPTY_ENTRY && x != HTAB_DELETED_ENTRY) |
564 | { |
565 | void **q = find_empty_slot_for_expand (htab, hash: (*htab->hash_f) (x)); |
566 | |
567 | *q = x; |
568 | } |
569 | |
570 | p++; |
571 | } |
572 | while (p < olimit); |
573 | |
574 | if (htab->free_f != NULL) |
575 | (*htab->free_f) (oentries); |
576 | else if (htab->free_with_arg_f != NULL) |
577 | (*htab->free_with_arg_f) (htab->alloc_arg, oentries); |
578 | return 1; |
579 | } |
580 | |
581 | /* This function searches for a hash table entry equal to the given |
582 | element. It cannot be used to insert or delete an element. */ |
583 | |
584 | void * |
585 | htab_find_with_hash (htab_t htab, const void *element, hashval_t hash) |
586 | { |
587 | hashval_t index, hash2; |
588 | size_t size; |
589 | void *entry; |
590 | |
591 | htab->searches++; |
592 | size = htab_size (htab); |
593 | index = htab_mod (hash, htab); |
594 | |
595 | entry = htab->entries[index]; |
596 | if (entry == HTAB_EMPTY_ENTRY |
597 | || (entry != HTAB_DELETED_ENTRY && (*htab->eq_f) (entry, element))) |
598 | return entry; |
599 | |
600 | hash2 = htab_mod_m2 (hash, htab); |
601 | for (;;) |
602 | { |
603 | htab->collisions++; |
604 | index += hash2; |
605 | if (index >= size) |
606 | index -= size; |
607 | |
608 | entry = htab->entries[index]; |
609 | if (entry == HTAB_EMPTY_ENTRY |
610 | || (entry != HTAB_DELETED_ENTRY && (*htab->eq_f) (entry, element))) |
611 | return entry; |
612 | } |
613 | } |
614 | |
615 | /* Like htab_find_slot_with_hash, but compute the hash value from the |
616 | element. */ |
617 | |
618 | void * |
619 | htab_find (htab_t htab, const void *element) |
620 | { |
621 | return htab_find_with_hash (htab, element, hash: (*htab->hash_f) (element)); |
622 | } |
623 | |
624 | /* This function searches for a hash table slot containing an entry |
625 | equal to the given element. To delete an entry, call this with |
626 | insert=NO_INSERT, then call htab_clear_slot on the slot returned |
627 | (possibly after doing some checks). To insert an entry, call this |
628 | with insert=INSERT, then write the value you want into the returned |
629 | slot. When inserting an entry, NULL may be returned if memory |
630 | allocation fails. */ |
631 | |
632 | void ** |
633 | htab_find_slot_with_hash (htab_t htab, const void *element, |
634 | hashval_t hash, enum insert_option insert) |
635 | { |
636 | void **first_deleted_slot; |
637 | hashval_t index, hash2; |
638 | size_t size; |
639 | void *entry; |
640 | |
641 | size = htab_size (htab); |
642 | if (insert == INSERT && size * 3 <= htab->n_elements * 4) |
643 | { |
644 | if (htab_expand (htab) == 0) |
645 | return NULL; |
646 | size = htab_size (htab); |
647 | } |
648 | |
649 | index = htab_mod (hash, htab); |
650 | |
651 | htab->searches++; |
652 | first_deleted_slot = NULL; |
653 | |
654 | entry = htab->entries[index]; |
655 | if (entry == HTAB_EMPTY_ENTRY) |
656 | goto empty_entry; |
657 | else if (entry == HTAB_DELETED_ENTRY) |
658 | first_deleted_slot = &htab->entries[index]; |
659 | else if ((*htab->eq_f) (entry, element)) |
660 | return &htab->entries[index]; |
661 | |
662 | hash2 = htab_mod_m2 (hash, htab); |
663 | for (;;) |
664 | { |
665 | htab->collisions++; |
666 | index += hash2; |
667 | if (index >= size) |
668 | index -= size; |
669 | |
670 | entry = htab->entries[index]; |
671 | if (entry == HTAB_EMPTY_ENTRY) |
672 | goto empty_entry; |
673 | else if (entry == HTAB_DELETED_ENTRY) |
674 | { |
675 | if (!first_deleted_slot) |
676 | first_deleted_slot = &htab->entries[index]; |
677 | } |
678 | else if ((*htab->eq_f) (entry, element)) |
679 | return &htab->entries[index]; |
680 | } |
681 | |
682 | empty_entry: |
683 | if (insert == NO_INSERT) |
684 | return NULL; |
685 | |
686 | if (first_deleted_slot) |
687 | { |
688 | htab->n_deleted--; |
689 | *first_deleted_slot = HTAB_EMPTY_ENTRY; |
690 | return first_deleted_slot; |
691 | } |
692 | |
693 | htab->n_elements++; |
694 | return &htab->entries[index]; |
695 | } |
696 | |
697 | /* Like htab_find_slot_with_hash, but compute the hash value from the |
698 | element. */ |
699 | |
700 | void ** |
701 | htab_find_slot (htab_t htab, const void *element, enum insert_option insert) |
702 | { |
703 | return htab_find_slot_with_hash (htab, element, hash: (*htab->hash_f) (element), |
704 | insert); |
705 | } |
706 | |
707 | /* This function deletes an element with the given value from hash |
708 | table (the hash is computed from the element). If there is no matching |
709 | element in the hash table, this function does nothing. */ |
710 | |
711 | void |
712 | htab_remove_elt (htab_t htab, const void *element) |
713 | { |
714 | htab_remove_elt_with_hash (htab, element, (*htab->hash_f) (element)); |
715 | } |
716 | |
717 | |
718 | /* This function deletes an element with the given value from hash |
719 | table. If there is no matching element in the hash table, this |
720 | function does nothing. */ |
721 | |
722 | void |
723 | htab_remove_elt_with_hash (htab_t htab, const void *element, hashval_t hash) |
724 | { |
725 | void **slot; |
726 | |
727 | slot = htab_find_slot_with_hash (htab, element, hash, insert: NO_INSERT); |
728 | if (slot == NULL) |
729 | return; |
730 | |
731 | if (htab->del_f) |
732 | (*htab->del_f) (*slot); |
733 | |
734 | *slot = HTAB_DELETED_ENTRY; |
735 | htab->n_deleted++; |
736 | } |
737 | |
738 | /* This function clears a specified slot in a hash table. It is |
739 | useful when you've already done the lookup and don't want to do it |
740 | again. */ |
741 | |
742 | void |
743 | htab_clear_slot (htab_t htab, void **slot) |
744 | { |
745 | if (slot < htab->entries || slot >= htab->entries + htab_size (htab) |
746 | || *slot == HTAB_EMPTY_ENTRY || *slot == HTAB_DELETED_ENTRY) |
747 | abort (); |
748 | |
749 | if (htab->del_f) |
750 | (*htab->del_f) (*slot); |
751 | |
752 | *slot = HTAB_DELETED_ENTRY; |
753 | htab->n_deleted++; |
754 | } |
755 | |
756 | /* This function scans over the entire hash table calling |
757 | CALLBACK for each live entry. If CALLBACK returns false, |
758 | the iteration stops. INFO is passed as CALLBACK's second |
759 | argument. */ |
760 | |
761 | void |
762 | htab_traverse_noresize (htab_t htab, htab_trav callback, void *info) |
763 | { |
764 | void **slot; |
765 | void **limit; |
766 | |
767 | slot = htab->entries; |
768 | limit = slot + htab_size (htab); |
769 | |
770 | do |
771 | { |
772 | void *x = *slot; |
773 | |
774 | if (x != HTAB_EMPTY_ENTRY && x != HTAB_DELETED_ENTRY) |
775 | if (!(*callback) (slot, info)) |
776 | break; |
777 | } |
778 | while (++slot < limit); |
779 | } |
780 | |
781 | /* Like htab_traverse_noresize, but does resize the table when it is |
782 | too empty to improve effectivity of subsequent calls. */ |
783 | |
784 | void |
785 | htab_traverse (htab_t htab, htab_trav callback, void *info) |
786 | { |
787 | size_t size = htab_size (htab); |
788 | if (htab_elements (htab) * 8 < size && size > 32) |
789 | htab_expand (htab); |
790 | |
791 | htab_traverse_noresize (htab, callback, info); |
792 | } |
793 | |
794 | /* Return the fraction of fixed collisions during all work with given |
795 | hash table. */ |
796 | |
797 | double |
798 | htab_collisions (htab_t htab) |
799 | { |
800 | if (htab->searches == 0) |
801 | return 0.0; |
802 | |
803 | return (double) htab->collisions / (double) htab->searches; |
804 | } |
805 | |
806 | /* Hash P as a null-terminated string. |
807 | |
808 | Copied from gcc/hashtable.c. Zack had the following to say with respect |
809 | to applicability, though note that unlike hashtable.c, this hash table |
810 | implementation re-hashes rather than chain buckets. |
811 | |
812 | http://gcc.gnu.org/ml/gcc-patches/2001-08/msg01021.html |
813 | From: Zack Weinberg <zackw@panix.com> |
814 | Date: Fri, 17 Aug 2001 02:15:56 -0400 |
815 | |
816 | I got it by extracting all the identifiers from all the source code |
817 | I had lying around in mid-1999, and testing many recurrences of |
818 | the form "H_n = H_{n-1} * K + c_n * L + M" where K, L, M were either |
819 | prime numbers or the appropriate identity. This was the best one. |
820 | I don't remember exactly what constituted "best", except I was |
821 | looking at bucket-length distributions mostly. |
822 | |
823 | So it should be very good at hashing identifiers, but might not be |
824 | as good at arbitrary strings. |
825 | |
826 | I'll add that it thoroughly trounces the hash functions recommended |
827 | for this use at http://burtleburtle.net/bob/hash/index.html, both |
828 | on speed and bucket distribution. I haven't tried it against the |
829 | function they just started using for Perl's hashes. */ |
830 | |
831 | hashval_t |
832 | htab_hash_string (const void *p) |
833 | { |
834 | const unsigned char *str = (const unsigned char *) p; |
835 | hashval_t r = 0; |
836 | unsigned char c; |
837 | |
838 | while ((c = *str++) != 0) |
839 | r = r * 67 + c - 113; |
840 | |
841 | return r; |
842 | } |
843 | |
844 | /* An equality function for null-terminated strings. */ |
845 | int |
846 | htab_eq_string (const void *a, const void *b) |
847 | { |
848 | return strcmp (s1: (const char *) a, s2: (const char *) b) == 0; |
849 | } |
850 | |
851 | /* DERIVED FROM: |
852 | -------------------------------------------------------------------- |
853 | lookup2.c, by Bob Jenkins, December 1996, Public Domain. |
854 | hash(), hash2(), hash3, and mix() are externally useful functions. |
855 | Routines to test the hash are included if SELF_TEST is defined. |
856 | You can use this free for any purpose. It has no warranty. |
857 | -------------------------------------------------------------------- |
858 | */ |
859 | |
860 | /* |
861 | -------------------------------------------------------------------- |
862 | mix -- mix 3 32-bit values reversibly. |
863 | For every delta with one or two bit set, and the deltas of all three |
864 | high bits or all three low bits, whether the original value of a,b,c |
865 | is almost all zero or is uniformly distributed, |
866 | * If mix() is run forward or backward, at least 32 bits in a,b,c |
867 | have at least 1/4 probability of changing. |
868 | * If mix() is run forward, every bit of c will change between 1/3 and |
869 | 2/3 of the time. (Well, 22/100 and 78/100 for some 2-bit deltas.) |
870 | mix() was built out of 36 single-cycle latency instructions in a |
871 | structure that could supported 2x parallelism, like so: |
872 | a -= b; |
873 | a -= c; x = (c>>13); |
874 | b -= c; a ^= x; |
875 | b -= a; x = (a<<8); |
876 | c -= a; b ^= x; |
877 | c -= b; x = (b>>13); |
878 | ... |
879 | Unfortunately, superscalar Pentiums and Sparcs can't take advantage |
880 | of that parallelism. They've also turned some of those single-cycle |
881 | latency instructions into multi-cycle latency instructions. Still, |
882 | this is the fastest good hash I could find. There were about 2^^68 |
883 | to choose from. I only looked at a billion or so. |
884 | -------------------------------------------------------------------- |
885 | */ |
886 | /* same, but slower, works on systems that might have 8 byte hashval_t's */ |
887 | #define mix(a,b,c) \ |
888 | { \ |
889 | a -= b; a -= c; a ^= (c>>13); \ |
890 | b -= c; b -= a; b ^= (a<< 8); \ |
891 | c -= a; c -= b; c ^= ((b&0xffffffff)>>13); \ |
892 | a -= b; a -= c; a ^= ((c&0xffffffff)>>12); \ |
893 | b -= c; b -= a; b = (b ^ (a<<16)) & 0xffffffff; \ |
894 | c -= a; c -= b; c = (c ^ (b>> 5)) & 0xffffffff; \ |
895 | a -= b; a -= c; a = (a ^ (c>> 3)) & 0xffffffff; \ |
896 | b -= c; b -= a; b = (b ^ (a<<10)) & 0xffffffff; \ |
897 | c -= a; c -= b; c = (c ^ (b>>15)) & 0xffffffff; \ |
898 | } |
899 | |
900 | /* |
901 | -------------------------------------------------------------------- |
902 | hash() -- hash a variable-length key into a 32-bit value |
903 | k : the key (the unaligned variable-length array of bytes) |
904 | len : the length of the key, counting by bytes |
905 | level : can be any 4-byte value |
906 | Returns a 32-bit value. Every bit of the key affects every bit of |
907 | the return value. Every 1-bit and 2-bit delta achieves avalanche. |
908 | About 36+6len instructions. |
909 | |
910 | The best hash table sizes are powers of 2. There is no need to do |
911 | mod a prime (mod is sooo slow!). If you need less than 32 bits, |
912 | use a bitmask. For example, if you need only 10 bits, do |
913 | h = (h & hashmask(10)); |
914 | In which case, the hash table should have hashsize(10) elements. |
915 | |
916 | If you are hashing n strings (ub1 **)k, do it like this: |
917 | for (i=0, h=0; i<n; ++i) h = hash( k[i], len[i], h); |
918 | |
919 | By Bob Jenkins, 1996. bob_jenkins@burtleburtle.net. You may use this |
920 | code any way you wish, private, educational, or commercial. It's free. |
921 | |
922 | See http://burtleburtle.net/bob/hash/evahash.html |
923 | Use for hash table lookup, or anything where one collision in 2^32 is |
924 | acceptable. Do NOT use for cryptographic purposes. |
925 | -------------------------------------------------------------------- |
926 | */ |
927 | |
928 | hashval_t |
929 | iterative_hash (const void *k_in /* the key */, |
930 | register size_t length /* the length of the key */, |
931 | register hashval_t initval /* the previous hash, or |
932 | an arbitrary value */) |
933 | { |
934 | register const unsigned char *k = (const unsigned char *)k_in; |
935 | register hashval_t a,b,c,len; |
936 | |
937 | /* Set up the internal state */ |
938 | len = length; |
939 | a = b = 0x9e3779b9; /* the golden ratio; an arbitrary value */ |
940 | c = initval; /* the previous hash value */ |
941 | |
942 | /*---------------------------------------- handle most of the key */ |
943 | /* Provide specialization for the aligned case for targets that cannot |
944 | efficiently perform misaligned loads of a merged access. */ |
945 | if ((((size_t)k)&3) == 0) |
946 | while (len >= 12) |
947 | { |
948 | a += (k[0] | ((hashval_t)k[1]<<8) | ((hashval_t)k[2]<<16) | ((hashval_t)k[3]<<24)); |
949 | b += (k[4] | ((hashval_t)k[5]<<8) | ((hashval_t)k[6]<<16) | ((hashval_t)k[7]<<24)); |
950 | c += (k[8] | ((hashval_t)k[9]<<8) | ((hashval_t)k[10]<<16)| ((hashval_t)k[11]<<24)); |
951 | mix(a,b,c); |
952 | k += 12; len -= 12; |
953 | } |
954 | else /* unaligned */ |
955 | while (len >= 12) |
956 | { |
957 | a += (k[0] | ((hashval_t)k[1]<<8) | ((hashval_t)k[2]<<16) | ((hashval_t)k[3]<<24)); |
958 | b += (k[4] | ((hashval_t)k[5]<<8) | ((hashval_t)k[6]<<16) | ((hashval_t)k[7]<<24)); |
959 | c += (k[8] | ((hashval_t)k[9]<<8) | ((hashval_t)k[10]<<16)| ((hashval_t)k[11]<<24)); |
960 | mix(a,b,c); |
961 | k += 12; len -= 12; |
962 | } |
963 | |
964 | /*------------------------------------- handle the last 11 bytes */ |
965 | c += length; |
966 | switch(len) /* all the case statements fall through */ |
967 | { |
968 | case 11: c+=((hashval_t)k[10]<<24); /* fall through */ |
969 | case 10: c+=((hashval_t)k[9]<<16); /* fall through */ |
970 | case 9 : c+=((hashval_t)k[8]<<8); /* fall through */ |
971 | /* the first byte of c is reserved for the length */ |
972 | case 8 : b+=((hashval_t)k[7]<<24); /* fall through */ |
973 | case 7 : b+=((hashval_t)k[6]<<16); /* fall through */ |
974 | case 6 : b+=((hashval_t)k[5]<<8); /* fall through */ |
975 | case 5 : b+=k[4]; /* fall through */ |
976 | case 4 : a+=((hashval_t)k[3]<<24); /* fall through */ |
977 | case 3 : a+=((hashval_t)k[2]<<16); /* fall through */ |
978 | case 2 : a+=((hashval_t)k[1]<<8); /* fall through */ |
979 | case 1 : a+=k[0]; |
980 | /* case 0: nothing left to add */ |
981 | } |
982 | mix(a,b,c); |
983 | /*-------------------------------------------- report the result */ |
984 | return c; |
985 | } |
986 | |
987 | /* Returns a hash code for pointer P. Simplified version of evahash */ |
988 | |
989 | static hashval_t |
990 | hash_pointer (const void *p) |
991 | { |
992 | intptr_t v = (intptr_t) p; |
993 | unsigned a, b, c; |
994 | |
995 | a = b = 0x9e3779b9; |
996 | a += v >> (sizeof (intptr_t) * CHAR_BIT / 2); |
997 | b += v & (((intptr_t) 1 << (sizeof (intptr_t) * CHAR_BIT / 2)) - 1); |
998 | c = 0x42135234; |
999 | mix (a, b, c); |
1000 | return c; |
1001 | } |
1002 | |