Commit | Line | Data |
---|---|---|
e2eaf477 | 1 | /* An expandable hash tables datatype. |
bb6a587d DD |
2 | Copyright (C) 1999, 2000, 2001, 2002, 2003, 2004 |
3 | Free Software Foundation, Inc. | |
e2eaf477 ILT |
4 | Contributed by Vladimir Makarov (vmakarov@cygnus.com). |
5 | ||
6 | This file is part of the libiberty library. | |
7 | Libiberty is free software; you can redistribute it and/or | |
8 | modify it under the terms of the GNU Library General Public | |
9 | License as published by the Free Software Foundation; either | |
10 | version 2 of the License, or (at your option) any later version. | |
11 | ||
12 | Libiberty is distributed in the hope that it will be useful, | |
13 | but WITHOUT ANY WARRANTY; without even the implied warranty of | |
14 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
15 | Library General Public License for more details. | |
16 | ||
17 | You should have received a copy of the GNU Library General Public | |
18 | License along with libiberty; see the file COPYING.LIB. If | |
979c05d3 NC |
19 | not, write to the Free Software Foundation, Inc., 51 Franklin Street - Fifth Floor, |
20 | Boston, MA 02110-1301, USA. */ | |
e2eaf477 ILT |
21 | |
22 | /* This package implements basic hash table functionality. It is possible | |
23 | to search for an entry, create an entry and destroy an entry. | |
24 | ||
25 | Elements in the table are generic pointers. | |
26 | ||
27 | The size of the table is not fixed; if the occupancy of the table | |
28 | grows too high the hash table will be expanded. | |
29 | ||
30 | The abstract data implementation is based on generalized Algorithm D | |
31 | from Knuth's book "The art of computer programming". Hash table is | |
32 | expanded by creation of new hash table and transferring elements from | |
33 | the old table to the new table. */ | |
34 | ||
35 | #ifdef HAVE_CONFIG_H | |
36 | #include "config.h" | |
37 | #endif | |
38 | ||
39 | #include <sys/types.h> | |
40 | ||
41 | #ifdef HAVE_STDLIB_H | |
42 | #include <stdlib.h> | |
43 | #endif | |
5c82d20a ZW |
44 | #ifdef HAVE_STRING_H |
45 | #include <string.h> | |
46 | #endif | |
5f73c378 DD |
47 | #ifdef HAVE_MALLOC_H |
48 | #include <malloc.h> | |
49 | #endif | |
bb6a587d DD |
50 | #ifdef HAVE_LIMITS_H |
51 | #include <limits.h> | |
52 | #endif | |
53 | #ifdef HAVE_STDINT_H | |
54 | #include <stdint.h> | |
55 | #endif | |
5f73c378 | 56 | |
e2eaf477 ILT |
57 | #include <stdio.h> |
58 | ||
59 | #include "libiberty.h" | |
bb6a587d | 60 | #include "ansidecl.h" |
e2eaf477 ILT |
61 | #include "hashtab.h" |
62 | ||
bb6a587d DD |
63 | #ifndef CHAR_BIT |
64 | #define CHAR_BIT 8 | |
65 | #endif | |
66 | ||
49b1fae4 DD |
67 | static unsigned int higher_prime_index (unsigned long); |
68 | static hashval_t htab_mod_1 (hashval_t, hashval_t, hashval_t, int); | |
69 | static hashval_t htab_mod (hashval_t, htab_t); | |
70 | static hashval_t htab_mod_m2 (hashval_t, htab_t); | |
71 | static hashval_t hash_pointer (const void *); | |
72 | static int eq_pointer (const void *, const void *); | |
73 | static int htab_expand (htab_t); | |
74 | static PTR *find_empty_slot_for_expand (htab_t, hashval_t); | |
eb383413 L |
75 | |
76 | /* At some point, we could make these be NULL, and modify the | |
77 | hash-table routines to handle NULL specially; that would avoid | |
78 | function-call overhead for the common case of hashing pointers. */ | |
79 | htab_hash htab_hash_pointer = hash_pointer; | |
80 | htab_eq htab_eq_pointer = eq_pointer; | |
81 | ||
bb6a587d DD |
82 | /* Table of primes and multiplicative inverses. |
83 | ||
84 | Note that these are not minimally reduced inverses. Unlike when generating | |
85 | code to divide by a constant, we want to be able to use the same algorithm | |
86 | all the time. All of these inverses (are implied to) have bit 32 set. | |
87 | ||
88 | For the record, here's the function that computed the table; it's a | |
89 | vastly simplified version of the function of the same name from gcc. */ | |
90 | ||
91 | #if 0 | |
92 | unsigned int | |
93 | ceil_log2 (unsigned int x) | |
94 | { | |
95 | int i; | |
96 | for (i = 31; i >= 0 ; --i) | |
97 | if (x > (1u << i)) | |
98 | return i+1; | |
99 | abort (); | |
100 | } | |
e2eaf477 | 101 | |
bb6a587d DD |
102 | unsigned int |
103 | choose_multiplier (unsigned int d, unsigned int *mlp, unsigned char *shiftp) | |
104 | { | |
105 | unsigned long long mhigh; | |
106 | double nx; | |
107 | int lgup, post_shift; | |
108 | int pow, pow2; | |
109 | int n = 32, precision = 32; | |
110 | ||
111 | lgup = ceil_log2 (d); | |
112 | pow = n + lgup; | |
113 | pow2 = n + lgup - precision; | |
114 | ||
115 | nx = ldexp (1.0, pow) + ldexp (1.0, pow2); | |
116 | mhigh = nx / d; | |
117 | ||
118 | *shiftp = lgup - 1; | |
119 | *mlp = mhigh; | |
120 | return mhigh >> 32; | |
121 | } | |
122 | #endif | |
123 | ||
124 | struct prime_ent | |
125 | { | |
126 | hashval_t prime; | |
127 | hashval_t inv; | |
128 | hashval_t inv_m2; /* inverse of prime-2 */ | |
129 | hashval_t shift; | |
130 | }; | |
131 | ||
132 | static struct prime_ent const prime_tab[] = { | |
133 | { 7, 0x24924925, 0x9999999b, 2 }, | |
134 | { 13, 0x3b13b13c, 0x745d1747, 3 }, | |
135 | { 31, 0x08421085, 0x1a7b9612, 4 }, | |
136 | { 61, 0x0c9714fc, 0x15b1e5f8, 5 }, | |
137 | { 127, 0x02040811, 0x0624dd30, 6 }, | |
138 | { 251, 0x05197f7e, 0x073260a5, 7 }, | |
139 | { 509, 0x01824366, 0x02864fc8, 8 }, | |
140 | { 1021, 0x00c0906d, 0x014191f7, 9 }, | |
141 | { 2039, 0x0121456f, 0x0161e69e, 10 }, | |
142 | { 4093, 0x00300902, 0x00501908, 11 }, | |
143 | { 8191, 0x00080041, 0x00180241, 12 }, | |
144 | { 16381, 0x000c0091, 0x00140191, 13 }, | |
145 | { 32749, 0x002605a5, 0x002a06e6, 14 }, | |
146 | { 65521, 0x000f00e2, 0x00110122, 15 }, | |
147 | { 131071, 0x00008001, 0x00018003, 16 }, | |
148 | { 262139, 0x00014002, 0x0001c004, 17 }, | |
149 | { 524287, 0x00002001, 0x00006001, 18 }, | |
150 | { 1048573, 0x00003001, 0x00005001, 19 }, | |
151 | { 2097143, 0x00004801, 0x00005801, 20 }, | |
152 | { 4194301, 0x00000c01, 0x00001401, 21 }, | |
153 | { 8388593, 0x00001e01, 0x00002201, 22 }, | |
154 | { 16777213, 0x00000301, 0x00000501, 23 }, | |
155 | { 33554393, 0x00001381, 0x00001481, 24 }, | |
156 | { 67108859, 0x00000141, 0x000001c1, 25 }, | |
157 | { 134217689, 0x000004e1, 0x00000521, 26 }, | |
158 | { 268435399, 0x00000391, 0x000003b1, 27 }, | |
159 | { 536870909, 0x00000019, 0x00000029, 28 }, | |
160 | { 1073741789, 0x0000008d, 0x00000095, 29 }, | |
161 | { 2147483647, 0x00000003, 0x00000007, 30 }, | |
162 | /* Avoid "decimal constant so large it is unsigned" for 4294967291. */ | |
163 | { 0xfffffffb, 0x00000006, 0x00000008, 31 } | |
164 | }; | |
165 | ||
166 | /* The following function returns an index into the above table of the | |
167 | nearest prime number which is greater than N, and near a power of two. */ | |
168 | ||
169 | static unsigned int | |
49b1fae4 | 170 | higher_prime_index (unsigned long n) |
e2eaf477 | 171 | { |
bb6a587d DD |
172 | unsigned int low = 0; |
173 | unsigned int high = sizeof(prime_tab) / sizeof(prime_tab[0]); | |
5ca0f83d DD |
174 | |
175 | while (low != high) | |
176 | { | |
bb6a587d DD |
177 | unsigned int mid = low + (high - low) / 2; |
178 | if (n > prime_tab[mid].prime) | |
5ca0f83d DD |
179 | low = mid + 1; |
180 | else | |
181 | high = mid; | |
182 | } | |
183 | ||
184 | /* If we've run out of primes, abort. */ | |
bb6a587d | 185 | if (n > prime_tab[low].prime) |
5ca0f83d DD |
186 | { |
187 | fprintf (stderr, "Cannot find prime bigger than %lu\n", n); | |
188 | abort (); | |
189 | } | |
190 | ||
bb6a587d | 191 | return low; |
e2eaf477 ILT |
192 | } |
193 | ||
eb383413 L |
194 | /* Returns a hash code for P. */ |
195 | ||
196 | static hashval_t | |
49b1fae4 | 197 | hash_pointer (const PTR p) |
eb383413 L |
198 | { |
199 | return (hashval_t) ((long)p >> 3); | |
200 | } | |
201 | ||
202 | /* Returns non-zero if P1 and P2 are equal. */ | |
203 | ||
204 | static int | |
49b1fae4 | 205 | eq_pointer (const PTR p1, const PTR p2) |
eb383413 L |
206 | { |
207 | return p1 == p2; | |
208 | } | |
209 | ||
fe046a17 | 210 | |
abf6a75b DD |
211 | /* The parens around the function names in the next two definitions |
212 | are essential in order to prevent macro expansions of the name. | |
213 | The bodies, however, are expanded as expected, so they are not | |
214 | recursive definitions. */ | |
215 | ||
216 | /* Return the current size of given hash table. */ | |
217 | ||
218 | #define htab_size(htab) ((htab)->size) | |
219 | ||
220 | size_t | |
221 | (htab_size) (htab_t htab) | |
fe046a17 | 222 | { |
abf6a75b | 223 | return htab_size (htab); |
fe046a17 DD |
224 | } |
225 | ||
226 | /* Return the current number of elements in given hash table. */ | |
227 | ||
abf6a75b DD |
228 | #define htab_elements(htab) ((htab)->n_elements - (htab)->n_deleted) |
229 | ||
230 | size_t | |
231 | (htab_elements) (htab_t htab) | |
fe046a17 | 232 | { |
abf6a75b | 233 | return htab_elements (htab); |
fe046a17 DD |
234 | } |
235 | ||
bb6a587d DD |
236 | /* Return X % Y. */ |
237 | ||
238 | static inline hashval_t | |
49b1fae4 | 239 | htab_mod_1 (hashval_t x, hashval_t y, hashval_t inv, int shift) |
bb6a587d DD |
240 | { |
241 | /* The multiplicative inverses computed above are for 32-bit types, and | |
242 | requires that we be able to compute a highpart multiply. */ | |
243 | #ifdef UNSIGNED_64BIT_TYPE | |
244 | __extension__ typedef UNSIGNED_64BIT_TYPE ull; | |
245 | if (sizeof (hashval_t) * CHAR_BIT <= 32) | |
246 | { | |
247 | hashval_t t1, t2, t3, t4, q, r; | |
248 | ||
249 | t1 = ((ull)x * inv) >> 32; | |
250 | t2 = x - t1; | |
251 | t3 = t2 >> 1; | |
252 | t4 = t1 + t3; | |
253 | q = t4 >> shift; | |
254 | r = x - (q * y); | |
255 | ||
256 | return r; | |
257 | } | |
258 | #endif | |
259 | ||
260 | /* Otherwise just use the native division routines. */ | |
261 | return x % y; | |
262 | } | |
263 | ||
fe046a17 DD |
264 | /* Compute the primary hash for HASH given HTAB's current size. */ |
265 | ||
266 | static inline hashval_t | |
49b1fae4 | 267 | htab_mod (hashval_t hash, htab_t htab) |
fe046a17 | 268 | { |
bb6a587d DD |
269 | const struct prime_ent *p = &prime_tab[htab->size_prime_index]; |
270 | return htab_mod_1 (hash, p->prime, p->inv, p->shift); | |
fe046a17 DD |
271 | } |
272 | ||
273 | /* Compute the secondary hash for HASH given HTAB's current size. */ | |
274 | ||
275 | static inline hashval_t | |
49b1fae4 | 276 | htab_mod_m2 (hashval_t hash, htab_t htab) |
fe046a17 | 277 | { |
bb6a587d DD |
278 | const struct prime_ent *p = &prime_tab[htab->size_prime_index]; |
279 | return 1 + htab_mod_1 (hash, p->prime - 2, p->inv_m2, p->shift); | |
fe046a17 DD |
280 | } |
281 | ||
e2eaf477 ILT |
282 | /* This function creates table with length slightly longer than given |
283 | source length. Created hash table is initiated as empty (all the | |
c3cca4c9 | 284 | hash table entries are HTAB_EMPTY_ENTRY). The function returns the |
18893690 | 285 | created hash table, or NULL if memory allocation fails. */ |
e2eaf477 | 286 | |
b4fe2683 | 287 | htab_t |
49b1fae4 DD |
288 | htab_create_alloc (size_t size, htab_hash hash_f, htab_eq eq_f, |
289 | htab_del del_f, htab_alloc alloc_f, htab_free free_f) | |
e2eaf477 | 290 | { |
b4fe2683 | 291 | htab_t result; |
bb6a587d DD |
292 | unsigned int size_prime_index; |
293 | ||
294 | size_prime_index = higher_prime_index (size); | |
295 | size = prime_tab[size_prime_index].prime; | |
e2eaf477 | 296 | |
18893690 DD |
297 | result = (htab_t) (*alloc_f) (1, sizeof (struct htab)); |
298 | if (result == NULL) | |
299 | return NULL; | |
300 | result->entries = (PTR *) (*alloc_f) (size, sizeof (PTR)); | |
301 | if (result->entries == NULL) | |
302 | { | |
303 | if (free_f != NULL) | |
304 | (*free_f) (result); | |
305 | return NULL; | |
306 | } | |
e2eaf477 | 307 | result->size = size; |
bb6a587d | 308 | result->size_prime_index = size_prime_index; |
b4fe2683 JM |
309 | result->hash_f = hash_f; |
310 | result->eq_f = eq_f; | |
311 | result->del_f = del_f; | |
18893690 DD |
312 | result->alloc_f = alloc_f; |
313 | result->free_f = free_f; | |
99a4c1bd HPN |
314 | return result; |
315 | } | |
316 | ||
5f9624e3 DJ |
317 | /* As above, but use the variants of alloc_f and free_f which accept |
318 | an extra argument. */ | |
319 | ||
320 | htab_t | |
abf6a75b DD |
321 | htab_create_alloc_ex (size_t size, htab_hash hash_f, htab_eq eq_f, |
322 | htab_del del_f, void *alloc_arg, | |
323 | htab_alloc_with_arg alloc_f, | |
324 | htab_free_with_arg free_f) | |
5f9624e3 DJ |
325 | { |
326 | htab_t result; | |
bb6a587d DD |
327 | unsigned int size_prime_index; |
328 | ||
329 | size_prime_index = higher_prime_index (size); | |
330 | size = prime_tab[size_prime_index].prime; | |
5f9624e3 | 331 | |
5f9624e3 DJ |
332 | result = (htab_t) (*alloc_f) (alloc_arg, 1, sizeof (struct htab)); |
333 | if (result == NULL) | |
334 | return NULL; | |
335 | result->entries = (PTR *) (*alloc_f) (alloc_arg, size, sizeof (PTR)); | |
336 | if (result->entries == NULL) | |
337 | { | |
338 | if (free_f != NULL) | |
339 | (*free_f) (alloc_arg, result); | |
340 | return NULL; | |
341 | } | |
342 | result->size = size; | |
bb6a587d | 343 | result->size_prime_index = size_prime_index; |
5f9624e3 DJ |
344 | result->hash_f = hash_f; |
345 | result->eq_f = eq_f; | |
346 | result->del_f = del_f; | |
347 | result->alloc_arg = alloc_arg; | |
348 | result->alloc_with_arg_f = alloc_f; | |
349 | result->free_with_arg_f = free_f; | |
350 | return result; | |
351 | } | |
352 | ||
353 | /* Update the function pointers and allocation parameter in the htab_t. */ | |
354 | ||
355 | void | |
49b1fae4 DD |
356 | htab_set_functions_ex (htab_t htab, htab_hash hash_f, htab_eq eq_f, |
357 | htab_del del_f, PTR alloc_arg, | |
358 | htab_alloc_with_arg alloc_f, htab_free_with_arg free_f) | |
5f9624e3 DJ |
359 | { |
360 | htab->hash_f = hash_f; | |
361 | htab->eq_f = eq_f; | |
362 | htab->del_f = del_f; | |
363 | htab->alloc_arg = alloc_arg; | |
364 | htab->alloc_with_arg_f = alloc_f; | |
365 | htab->free_with_arg_f = free_f; | |
366 | } | |
367 | ||
18893690 | 368 | /* These functions exist solely for backward compatibility. */ |
99a4c1bd | 369 | |
18893690 | 370 | #undef htab_create |
99a4c1bd | 371 | htab_t |
49b1fae4 | 372 | htab_create (size_t size, htab_hash hash_f, htab_eq eq_f, htab_del del_f) |
99a4c1bd | 373 | { |
18893690 DD |
374 | return htab_create_alloc (size, hash_f, eq_f, del_f, xcalloc, free); |
375 | } | |
99a4c1bd | 376 | |
18893690 | 377 | htab_t |
49b1fae4 | 378 | htab_try_create (size_t size, htab_hash hash_f, htab_eq eq_f, htab_del del_f) |
18893690 DD |
379 | { |
380 | return htab_create_alloc (size, hash_f, eq_f, del_f, calloc, free); | |
e2eaf477 ILT |
381 | } |
382 | ||
383 | /* This function frees all memory allocated for given hash table. | |
384 | Naturally the hash table must already exist. */ | |
385 | ||
386 | void | |
49b1fae4 | 387 | htab_delete (htab_t htab) |
e2eaf477 | 388 | { |
fe046a17 DD |
389 | size_t size = htab_size (htab); |
390 | PTR *entries = htab->entries; | |
b4fe2683 | 391 | int i; |
eb383413 | 392 | |
b4fe2683 | 393 | if (htab->del_f) |
fe046a17 | 394 | for (i = size - 1; i >= 0; i--) |
c3cca4c9 | 395 | if (entries[i] != HTAB_EMPTY_ENTRY && entries[i] != HTAB_DELETED_ENTRY) |
fe046a17 | 396 | (*htab->del_f) (entries[i]); |
b4fe2683 | 397 | |
18893690 DD |
398 | if (htab->free_f != NULL) |
399 | { | |
fe046a17 | 400 | (*htab->free_f) (entries); |
18893690 DD |
401 | (*htab->free_f) (htab); |
402 | } | |
5f9624e3 DJ |
403 | else if (htab->free_with_arg_f != NULL) |
404 | { | |
fe046a17 | 405 | (*htab->free_with_arg_f) (htab->alloc_arg, entries); |
5f9624e3 DJ |
406 | (*htab->free_with_arg_f) (htab->alloc_arg, htab); |
407 | } | |
e2eaf477 ILT |
408 | } |
409 | ||
410 | /* This function clears all entries in the given hash table. */ | |
411 | ||
412 | void | |
49b1fae4 | 413 | htab_empty (htab_t htab) |
b4fe2683 | 414 | { |
fe046a17 DD |
415 | size_t size = htab_size (htab); |
416 | PTR *entries = htab->entries; | |
b4fe2683 | 417 | int i; |
eb383413 | 418 | |
b4fe2683 | 419 | if (htab->del_f) |
fe046a17 | 420 | for (i = size - 1; i >= 0; i--) |
c3cca4c9 | 421 | if (entries[i] != HTAB_EMPTY_ENTRY && entries[i] != HTAB_DELETED_ENTRY) |
fe046a17 | 422 | (*htab->del_f) (entries[i]); |
b4fe2683 | 423 | |
a7d421b8 DD |
424 | /* Instead of clearing megabyte, downsize the table. */ |
425 | if (size > 1024*1024 / sizeof (PTR)) | |
426 | { | |
427 | int nindex = higher_prime_index (1024 / sizeof (PTR)); | |
428 | int nsize = prime_tab[nindex].prime; | |
429 | ||
430 | if (htab->free_f != NULL) | |
431 | (*htab->free_f) (htab->entries); | |
432 | else if (htab->free_with_arg_f != NULL) | |
433 | (*htab->free_with_arg_f) (htab->alloc_arg, htab->entries); | |
434 | if (htab->alloc_with_arg_f != NULL) | |
435 | htab->entries = (PTR *) (*htab->alloc_with_arg_f) (htab->alloc_arg, nsize, | |
436 | sizeof (PTR *)); | |
437 | else | |
438 | htab->entries = (PTR *) (*htab->alloc_f) (nsize, sizeof (PTR *)); | |
439 | htab->size = nsize; | |
440 | htab->size_prime_index = nindex; | |
441 | } | |
442 | else | |
443 | memset (entries, 0, size * sizeof (PTR)); | |
444 | htab->n_deleted = 0; | |
445 | htab->n_elements = 0; | |
b4fe2683 JM |
446 | } |
447 | ||
448 | /* Similar to htab_find_slot, but without several unwanted side effects: | |
449 | - Does not call htab->eq_f when it finds an existing entry. | |
450 | - Does not change the count of elements/searches/collisions in the | |
451 | hash table. | |
452 | This function also assumes there are no deleted entries in the table. | |
453 | HASH is the hash value for the element to be inserted. */ | |
eb383413 | 454 | |
e0f3df8f | 455 | static PTR * |
49b1fae4 | 456 | find_empty_slot_for_expand (htab_t htab, hashval_t hash) |
e2eaf477 | 457 | { |
fe046a17 DD |
458 | hashval_t index = htab_mod (hash, htab); |
459 | size_t size = htab_size (htab); | |
b1c933fc RH |
460 | PTR *slot = htab->entries + index; |
461 | hashval_t hash2; | |
462 | ||
c3cca4c9 | 463 | if (*slot == HTAB_EMPTY_ENTRY) |
b1c933fc | 464 | return slot; |
c3cca4c9 | 465 | else if (*slot == HTAB_DELETED_ENTRY) |
b1c933fc | 466 | abort (); |
b4fe2683 | 467 | |
fe046a17 | 468 | hash2 = htab_mod_m2 (hash, htab); |
b4fe2683 JM |
469 | for (;;) |
470 | { | |
b1c933fc RH |
471 | index += hash2; |
472 | if (index >= size) | |
473 | index -= size; | |
eb383413 | 474 | |
b1c933fc | 475 | slot = htab->entries + index; |
c3cca4c9 | 476 | if (*slot == HTAB_EMPTY_ENTRY) |
b4fe2683 | 477 | return slot; |
c3cca4c9 | 478 | else if (*slot == HTAB_DELETED_ENTRY) |
b4fe2683 | 479 | abort (); |
b4fe2683 | 480 | } |
e2eaf477 ILT |
481 | } |
482 | ||
483 | /* The following function changes size of memory allocated for the | |
484 | entries and repeatedly inserts the table elements. The occupancy | |
485 | of the table after the call will be about 50%. Naturally the hash | |
486 | table must already exist. Remember also that the place of the | |
99a4c1bd HPN |
487 | table entries is changed. If memory allocation failures are allowed, |
488 | this function will return zero, indicating that the table could not be | |
489 | expanded. If all goes well, it will return a non-zero value. */ | |
e2eaf477 | 490 | |
99a4c1bd | 491 | static int |
49b1fae4 | 492 | htab_expand (htab_t htab) |
e2eaf477 | 493 | { |
e0f3df8f HPN |
494 | PTR *oentries; |
495 | PTR *olimit; | |
496 | PTR *p; | |
18893690 | 497 | PTR *nentries; |
bb6a587d DD |
498 | size_t nsize, osize, elts; |
499 | unsigned int oindex, nindex; | |
b4fe2683 JM |
500 | |
501 | oentries = htab->entries; | |
bb6a587d DD |
502 | oindex = htab->size_prime_index; |
503 | osize = htab->size; | |
504 | olimit = oentries + osize; | |
505 | elts = htab_elements (htab); | |
b4fe2683 | 506 | |
c4d8feb2 DD |
507 | /* Resize only when table after removal of unused elements is either |
508 | too full or too empty. */ | |
bb6a587d DD |
509 | if (elts * 2 > osize || (elts * 8 < osize && osize > 32)) |
510 | { | |
511 | nindex = higher_prime_index (elts * 2); | |
512 | nsize = prime_tab[nindex].prime; | |
513 | } | |
c4d8feb2 | 514 | else |
bb6a587d DD |
515 | { |
516 | nindex = oindex; | |
517 | nsize = osize; | |
518 | } | |
99a4c1bd | 519 | |
5f9624e3 DJ |
520 | if (htab->alloc_with_arg_f != NULL) |
521 | nentries = (PTR *) (*htab->alloc_with_arg_f) (htab->alloc_arg, nsize, | |
522 | sizeof (PTR *)); | |
523 | else | |
524 | nentries = (PTR *) (*htab->alloc_f) (nsize, sizeof (PTR *)); | |
18893690 DD |
525 | if (nentries == NULL) |
526 | return 0; | |
527 | htab->entries = nentries; | |
eed2b28c | 528 | htab->size = nsize; |
bb6a587d | 529 | htab->size_prime_index = nindex; |
b4fe2683 JM |
530 | htab->n_elements -= htab->n_deleted; |
531 | htab->n_deleted = 0; | |
532 | ||
533 | p = oentries; | |
534 | do | |
535 | { | |
e0f3df8f | 536 | PTR x = *p; |
eb383413 | 537 | |
c3cca4c9 | 538 | if (x != HTAB_EMPTY_ENTRY && x != HTAB_DELETED_ENTRY) |
b4fe2683 | 539 | { |
e0f3df8f | 540 | PTR *q = find_empty_slot_for_expand (htab, (*htab->hash_f) (x)); |
eb383413 | 541 | |
b4fe2683 JM |
542 | *q = x; |
543 | } | |
eb383413 | 544 | |
b4fe2683 JM |
545 | p++; |
546 | } | |
547 | while (p < olimit); | |
eb383413 | 548 | |
18893690 DD |
549 | if (htab->free_f != NULL) |
550 | (*htab->free_f) (oentries); | |
5f9624e3 DJ |
551 | else if (htab->free_with_arg_f != NULL) |
552 | (*htab->free_with_arg_f) (htab->alloc_arg, oentries); | |
99a4c1bd | 553 | return 1; |
e2eaf477 ILT |
554 | } |
555 | ||
b4fe2683 JM |
556 | /* This function searches for a hash table entry equal to the given |
557 | element. It cannot be used to insert or delete an element. */ | |
558 | ||
e0f3df8f | 559 | PTR |
49b1fae4 | 560 | htab_find_with_hash (htab_t htab, const PTR element, hashval_t hash) |
e2eaf477 | 561 | { |
fe046a17 | 562 | hashval_t index, hash2; |
b4fe2683 | 563 | size_t size; |
e0f3df8f | 564 | PTR entry; |
e2eaf477 | 565 | |
b4fe2683 | 566 | htab->searches++; |
fe046a17 DD |
567 | size = htab_size (htab); |
568 | index = htab_mod (hash, htab); | |
b4fe2683 | 569 | |
eb383413 | 570 | entry = htab->entries[index]; |
c3cca4c9 DD |
571 | if (entry == HTAB_EMPTY_ENTRY |
572 | || (entry != HTAB_DELETED_ENTRY && (*htab->eq_f) (entry, element))) | |
eb383413 L |
573 | return entry; |
574 | ||
fe046a17 | 575 | hash2 = htab_mod_m2 (hash, htab); |
b4fe2683 | 576 | for (;;) |
e2eaf477 | 577 | { |
b4fe2683 JM |
578 | htab->collisions++; |
579 | index += hash2; | |
580 | if (index >= size) | |
581 | index -= size; | |
eb383413 L |
582 | |
583 | entry = htab->entries[index]; | |
c3cca4c9 DD |
584 | if (entry == HTAB_EMPTY_ENTRY |
585 | || (entry != HTAB_DELETED_ENTRY && (*htab->eq_f) (entry, element))) | |
eb383413 | 586 | return entry; |
e2eaf477 | 587 | } |
b4fe2683 JM |
588 | } |
589 | ||
590 | /* Like htab_find_slot_with_hash, but compute the hash value from the | |
591 | element. */ | |
eb383413 | 592 | |
e0f3df8f | 593 | PTR |
49b1fae4 | 594 | htab_find (htab_t htab, const PTR element) |
b4fe2683 JM |
595 | { |
596 | return htab_find_with_hash (htab, element, (*htab->hash_f) (element)); | |
597 | } | |
598 | ||
599 | /* This function searches for a hash table slot containing an entry | |
600 | equal to the given element. To delete an entry, call this with | |
bac7199c DD |
601 | insert=NO_INSERT, then call htab_clear_slot on the slot returned |
602 | (possibly after doing some checks). To insert an entry, call this | |
603 | with insert=INSERT, then write the value you want into the returned | |
604 | slot. When inserting an entry, NULL may be returned if memory | |
605 | allocation fails. */ | |
b4fe2683 | 606 | |
e0f3df8f | 607 | PTR * |
49b1fae4 DD |
608 | htab_find_slot_with_hash (htab_t htab, const PTR element, |
609 | hashval_t hash, enum insert_option insert) | |
b4fe2683 | 610 | { |
e0f3df8f | 611 | PTR *first_deleted_slot; |
fe046a17 | 612 | hashval_t index, hash2; |
b4fe2683 | 613 | size_t size; |
b1c933fc | 614 | PTR entry; |
b4fe2683 | 615 | |
fe046a17 DD |
616 | size = htab_size (htab); |
617 | if (insert == INSERT && size * 3 <= htab->n_elements * 4) | |
618 | { | |
619 | if (htab_expand (htab) == 0) | |
620 | return NULL; | |
621 | size = htab_size (htab); | |
622 | } | |
b4fe2683 | 623 | |
fe046a17 | 624 | index = htab_mod (hash, htab); |
b4fe2683 | 625 | |
e2eaf477 | 626 | htab->searches++; |
b4fe2683 JM |
627 | first_deleted_slot = NULL; |
628 | ||
b1c933fc | 629 | entry = htab->entries[index]; |
c3cca4c9 | 630 | if (entry == HTAB_EMPTY_ENTRY) |
b1c933fc | 631 | goto empty_entry; |
c3cca4c9 | 632 | else if (entry == HTAB_DELETED_ENTRY) |
b1c933fc RH |
633 | first_deleted_slot = &htab->entries[index]; |
634 | else if ((*htab->eq_f) (entry, element)) | |
635 | return &htab->entries[index]; | |
636 | ||
fe046a17 | 637 | hash2 = htab_mod_m2 (hash, htab); |
b4fe2683 | 638 | for (;;) |
e2eaf477 | 639 | { |
b1c933fc RH |
640 | htab->collisions++; |
641 | index += hash2; | |
642 | if (index >= size) | |
643 | index -= size; | |
644 | ||
645 | entry = htab->entries[index]; | |
c3cca4c9 | 646 | if (entry == HTAB_EMPTY_ENTRY) |
b1c933fc | 647 | goto empty_entry; |
c3cca4c9 | 648 | else if (entry == HTAB_DELETED_ENTRY) |
b4fe2683 JM |
649 | { |
650 | if (!first_deleted_slot) | |
651 | first_deleted_slot = &htab->entries[index]; | |
652 | } | |
b1c933fc | 653 | else if ((*htab->eq_f) (entry, element)) |
eb383413 | 654 | return &htab->entries[index]; |
e2eaf477 | 655 | } |
b1c933fc RH |
656 | |
657 | empty_entry: | |
658 | if (insert == NO_INSERT) | |
659 | return NULL; | |
660 | ||
b1c933fc RH |
661 | if (first_deleted_slot) |
662 | { | |
686e72d7 | 663 | htab->n_deleted--; |
c3cca4c9 | 664 | *first_deleted_slot = HTAB_EMPTY_ENTRY; |
b1c933fc RH |
665 | return first_deleted_slot; |
666 | } | |
667 | ||
686e72d7 | 668 | htab->n_elements++; |
b1c933fc | 669 | return &htab->entries[index]; |
e2eaf477 ILT |
670 | } |
671 | ||
b4fe2683 JM |
672 | /* Like htab_find_slot_with_hash, but compute the hash value from the |
673 | element. */ | |
eb383413 | 674 | |
e0f3df8f | 675 | PTR * |
49b1fae4 | 676 | htab_find_slot (htab_t htab, const PTR element, enum insert_option insert) |
b4fe2683 JM |
677 | { |
678 | return htab_find_slot_with_hash (htab, element, (*htab->hash_f) (element), | |
679 | insert); | |
680 | } | |
681 | ||
d6ea4e80 DD |
682 | /* This function deletes an element with the given value from hash |
683 | table (the hash is computed from the element). If there is no matching | |
684 | element in the hash table, this function does nothing. */ | |
685 | ||
686 | void | |
49b1fae4 | 687 | htab_remove_elt (htab_t htab, PTR element) |
d6ea4e80 DD |
688 | { |
689 | htab_remove_elt_with_hash (htab, element, (*htab->hash_f) (element)); | |
690 | } | |
691 | ||
692 | ||
b4fe2683 JM |
693 | /* This function deletes an element with the given value from hash |
694 | table. If there is no matching element in the hash table, this | |
695 | function does nothing. */ | |
e2eaf477 ILT |
696 | |
697 | void | |
49b1fae4 | 698 | htab_remove_elt_with_hash (htab_t htab, PTR element, hashval_t hash) |
e2eaf477 | 699 | { |
e0f3df8f | 700 | PTR *slot; |
b4fe2683 | 701 | |
d6ea4e80 | 702 | slot = htab_find_slot_with_hash (htab, element, hash, NO_INSERT); |
c3cca4c9 | 703 | if (*slot == HTAB_EMPTY_ENTRY) |
b4fe2683 JM |
704 | return; |
705 | ||
706 | if (htab->del_f) | |
707 | (*htab->del_f) (*slot); | |
e2eaf477 | 708 | |
c3cca4c9 | 709 | *slot = HTAB_DELETED_ENTRY; |
b4fe2683 | 710 | htab->n_deleted++; |
e2eaf477 ILT |
711 | } |
712 | ||
b4fe2683 JM |
713 | /* This function clears a specified slot in a hash table. It is |
714 | useful when you've already done the lookup and don't want to do it | |
715 | again. */ | |
e2eaf477 ILT |
716 | |
717 | void | |
49b1fae4 | 718 | htab_clear_slot (htab_t htab, PTR *slot) |
e2eaf477 | 719 | { |
fe046a17 | 720 | if (slot < htab->entries || slot >= htab->entries + htab_size (htab) |
c3cca4c9 | 721 | || *slot == HTAB_EMPTY_ENTRY || *slot == HTAB_DELETED_ENTRY) |
e2eaf477 | 722 | abort (); |
eb383413 | 723 | |
b4fe2683 JM |
724 | if (htab->del_f) |
725 | (*htab->del_f) (*slot); | |
eb383413 | 726 | |
c3cca4c9 | 727 | *slot = HTAB_DELETED_ENTRY; |
b4fe2683 | 728 | htab->n_deleted++; |
e2eaf477 ILT |
729 | } |
730 | ||
731 | /* This function scans over the entire hash table calling | |
732 | CALLBACK for each live entry. If CALLBACK returns false, | |
733 | the iteration stops. INFO is passed as CALLBACK's second | |
734 | argument. */ | |
735 | ||
736 | void | |
49b1fae4 | 737 | htab_traverse_noresize (htab_t htab, htab_trav callback, PTR info) |
e2eaf477 | 738 | { |
c4d8feb2 DD |
739 | PTR *slot; |
740 | PTR *limit; | |
c3cca4c9 | 741 | |
c4d8feb2 | 742 | slot = htab->entries; |
fe046a17 | 743 | limit = slot + htab_size (htab); |
eb383413 | 744 | |
b4fe2683 JM |
745 | do |
746 | { | |
e0f3df8f | 747 | PTR x = *slot; |
eb383413 | 748 | |
c3cca4c9 | 749 | if (x != HTAB_EMPTY_ENTRY && x != HTAB_DELETED_ENTRY) |
b4fe2683 JM |
750 | if (!(*callback) (slot, info)) |
751 | break; | |
752 | } | |
753 | while (++slot < limit); | |
e2eaf477 ILT |
754 | } |
755 | ||
f77ed96c DD |
756 | /* Like htab_traverse_noresize, but does resize the table when it is |
757 | too empty to improve effectivity of subsequent calls. */ | |
758 | ||
759 | void | |
49b1fae4 | 760 | htab_traverse (htab_t htab, htab_trav callback, PTR info) |
f77ed96c | 761 | { |
fe046a17 | 762 | if (htab_elements (htab) * 8 < htab_size (htab)) |
f77ed96c DD |
763 | htab_expand (htab); |
764 | ||
765 | htab_traverse_noresize (htab, callback, info); | |
766 | } | |
767 | ||
eb383413 L |
768 | /* Return the fraction of fixed collisions during all work with given |
769 | hash table. */ | |
e2eaf477 | 770 | |
b4fe2683 | 771 | double |
49b1fae4 | 772 | htab_collisions (htab_t htab) |
e2eaf477 | 773 | { |
eb383413 | 774 | if (htab->searches == 0) |
b4fe2683 | 775 | return 0.0; |
eb383413 L |
776 | |
777 | return (double) htab->collisions / (double) htab->searches; | |
e2eaf477 | 778 | } |
8fc34799 | 779 | |
68a41de7 DD |
780 | /* Hash P as a null-terminated string. |
781 | ||
782 | Copied from gcc/hashtable.c. Zack had the following to say with respect | |
783 | to applicability, though note that unlike hashtable.c, this hash table | |
784 | implementation re-hashes rather than chain buckets. | |
785 | ||
786 | http://gcc.gnu.org/ml/gcc-patches/2001-08/msg01021.html | |
787 | From: Zack Weinberg <zackw@panix.com> | |
788 | Date: Fri, 17 Aug 2001 02:15:56 -0400 | |
789 | ||
790 | I got it by extracting all the identifiers from all the source code | |
791 | I had lying around in mid-1999, and testing many recurrences of | |
792 | the form "H_n = H_{n-1} * K + c_n * L + M" where K, L, M were either | |
793 | prime numbers or the appropriate identity. This was the best one. | |
794 | I don't remember exactly what constituted "best", except I was | |
795 | looking at bucket-length distributions mostly. | |
796 | ||
797 | So it should be very good at hashing identifiers, but might not be | |
798 | as good at arbitrary strings. | |
799 | ||
800 | I'll add that it thoroughly trounces the hash functions recommended | |
801 | for this use at http://burtleburtle.net/bob/hash/index.html, both | |
802 | on speed and bucket distribution. I haven't tried it against the | |
803 | function they just started using for Perl's hashes. */ | |
8fc34799 DD |
804 | |
805 | hashval_t | |
49b1fae4 | 806 | htab_hash_string (const PTR p) |
8fc34799 DD |
807 | { |
808 | const unsigned char *str = (const unsigned char *) p; | |
809 | hashval_t r = 0; | |
810 | unsigned char c; | |
811 | ||
812 | while ((c = *str++) != 0) | |
813 | r = r * 67 + c - 113; | |
814 | ||
815 | return r; | |
816 | } | |
7108c5dc JM |
817 | |
818 | /* DERIVED FROM: | |
819 | -------------------------------------------------------------------- | |
820 | lookup2.c, by Bob Jenkins, December 1996, Public Domain. | |
821 | hash(), hash2(), hash3, and mix() are externally useful functions. | |
822 | Routines to test the hash are included if SELF_TEST is defined. | |
823 | You can use this free for any purpose. It has no warranty. | |
824 | -------------------------------------------------------------------- | |
825 | */ | |
826 | ||
827 | /* | |
828 | -------------------------------------------------------------------- | |
829 | mix -- mix 3 32-bit values reversibly. | |
830 | For every delta with one or two bit set, and the deltas of all three | |
831 | high bits or all three low bits, whether the original value of a,b,c | |
832 | is almost all zero or is uniformly distributed, | |
833 | * If mix() is run forward or backward, at least 32 bits in a,b,c | |
834 | have at least 1/4 probability of changing. | |
835 | * If mix() is run forward, every bit of c will change between 1/3 and | |
836 | 2/3 of the time. (Well, 22/100 and 78/100 for some 2-bit deltas.) | |
837 | mix() was built out of 36 single-cycle latency instructions in a | |
838 | structure that could supported 2x parallelism, like so: | |
839 | a -= b; | |
840 | a -= c; x = (c>>13); | |
841 | b -= c; a ^= x; | |
842 | b -= a; x = (a<<8); | |
843 | c -= a; b ^= x; | |
844 | c -= b; x = (b>>13); | |
845 | ... | |
846 | Unfortunately, superscalar Pentiums and Sparcs can't take advantage | |
847 | of that parallelism. They've also turned some of those single-cycle | |
848 | latency instructions into multi-cycle latency instructions. Still, | |
849 | this is the fastest good hash I could find. There were about 2^^68 | |
850 | to choose from. I only looked at a billion or so. | |
851 | -------------------------------------------------------------------- | |
852 | */ | |
853 | /* same, but slower, works on systems that might have 8 byte hashval_t's */ | |
854 | #define mix(a,b,c) \ | |
855 | { \ | |
856 | a -= b; a -= c; a ^= (c>>13); \ | |
857 | b -= c; b -= a; b ^= (a<< 8); \ | |
858 | c -= a; c -= b; c ^= ((b&0xffffffff)>>13); \ | |
859 | a -= b; a -= c; a ^= ((c&0xffffffff)>>12); \ | |
860 | b -= c; b -= a; b = (b ^ (a<<16)) & 0xffffffff; \ | |
861 | c -= a; c -= b; c = (c ^ (b>> 5)) & 0xffffffff; \ | |
862 | a -= b; a -= c; a = (a ^ (c>> 3)) & 0xffffffff; \ | |
863 | b -= c; b -= a; b = (b ^ (a<<10)) & 0xffffffff; \ | |
864 | c -= a; c -= b; c = (c ^ (b>>15)) & 0xffffffff; \ | |
865 | } | |
866 | ||
867 | /* | |
868 | -------------------------------------------------------------------- | |
869 | hash() -- hash a variable-length key into a 32-bit value | |
870 | k : the key (the unaligned variable-length array of bytes) | |
871 | len : the length of the key, counting by bytes | |
872 | level : can be any 4-byte value | |
873 | Returns a 32-bit value. Every bit of the key affects every bit of | |
874 | the return value. Every 1-bit and 2-bit delta achieves avalanche. | |
875 | About 36+6len instructions. | |
876 | ||
877 | The best hash table sizes are powers of 2. There is no need to do | |
878 | mod a prime (mod is sooo slow!). If you need less than 32 bits, | |
879 | use a bitmask. For example, if you need only 10 bits, do | |
880 | h = (h & hashmask(10)); | |
881 | In which case, the hash table should have hashsize(10) elements. | |
882 | ||
883 | If you are hashing n strings (ub1 **)k, do it like this: | |
884 | for (i=0, h=0; i<n; ++i) h = hash( k[i], len[i], h); | |
885 | ||
886 | By Bob Jenkins, 1996. bob_jenkins@burtleburtle.net. You may use this | |
887 | code any way you wish, private, educational, or commercial. It's free. | |
888 | ||
889 | See http://burtleburtle.net/bob/hash/evahash.html | |
890 | Use for hash table lookup, or anything where one collision in 2^32 is | |
891 | acceptable. Do NOT use for cryptographic purposes. | |
892 | -------------------------------------------------------------------- | |
893 | */ | |
894 | ||
49b1fae4 DD |
895 | hashval_t |
896 | iterative_hash (const PTR k_in /* the key */, | |
897 | register size_t length /* the length of the key */, | |
898 | register hashval_t initval /* the previous hash, or | |
899 | an arbitrary value */) | |
7108c5dc JM |
900 | { |
901 | register const unsigned char *k = (const unsigned char *)k_in; | |
902 | register hashval_t a,b,c,len; | |
903 | ||
904 | /* Set up the internal state */ | |
905 | len = length; | |
906 | a = b = 0x9e3779b9; /* the golden ratio; an arbitrary value */ | |
907 | c = initval; /* the previous hash value */ | |
908 | ||
909 | /*---------------------------------------- handle most of the key */ | |
910 | #ifndef WORDS_BIGENDIAN | |
911 | /* On a little-endian machine, if the data is 4-byte aligned we can hash | |
912 | by word for better speed. This gives nondeterministic results on | |
913 | big-endian machines. */ | |
914 | if (sizeof (hashval_t) == 4 && (((size_t)k)&3) == 0) | |
915 | while (len >= 12) /* aligned */ | |
916 | { | |
917 | a += *(hashval_t *)(k+0); | |
918 | b += *(hashval_t *)(k+4); | |
919 | c += *(hashval_t *)(k+8); | |
920 | mix(a,b,c); | |
921 | k += 12; len -= 12; | |
922 | } | |
923 | else /* unaligned */ | |
924 | #endif | |
925 | while (len >= 12) | |
926 | { | |
927 | a += (k[0] +((hashval_t)k[1]<<8) +((hashval_t)k[2]<<16) +((hashval_t)k[3]<<24)); | |
928 | b += (k[4] +((hashval_t)k[5]<<8) +((hashval_t)k[6]<<16) +((hashval_t)k[7]<<24)); | |
929 | c += (k[8] +((hashval_t)k[9]<<8) +((hashval_t)k[10]<<16)+((hashval_t)k[11]<<24)); | |
930 | mix(a,b,c); | |
931 | k += 12; len -= 12; | |
932 | } | |
933 | ||
934 | /*------------------------------------- handle the last 11 bytes */ | |
935 | c += length; | |
936 | switch(len) /* all the case statements fall through */ | |
937 | { | |
938 | case 11: c+=((hashval_t)k[10]<<24); | |
939 | case 10: c+=((hashval_t)k[9]<<16); | |
940 | case 9 : c+=((hashval_t)k[8]<<8); | |
941 | /* the first byte of c is reserved for the length */ | |
942 | case 8 : b+=((hashval_t)k[7]<<24); | |
943 | case 7 : b+=((hashval_t)k[6]<<16); | |
944 | case 6 : b+=((hashval_t)k[5]<<8); | |
945 | case 5 : b+=k[4]; | |
946 | case 4 : a+=((hashval_t)k[3]<<24); | |
947 | case 3 : a+=((hashval_t)k[2]<<16); | |
948 | case 2 : a+=((hashval_t)k[1]<<8); | |
949 | case 1 : a+=k[0]; | |
950 | /* case 0: nothing left to add */ | |
951 | } | |
952 | mix(a,b,c); | |
953 | /*-------------------------------------------- report the result */ | |
954 | return c; | |
955 | } |