| 1 | /* An expandable hash tables datatype. |
| 2 | Copyright (C) 1999, 2000, 2001, 2002, 2003, 2004 |
| 3 | Free Software Foundation, Inc. |
| 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 |
| 19 | not, write to the Free Software Foundation, Inc., 51 Franklin Street - Fifth Floor, |
| 20 | Boston, MA 02110-1301, USA. */ |
| 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 |
| 44 | #ifdef HAVE_STRING_H |
| 45 | #include <string.h> |
| 46 | #endif |
| 47 | #ifdef HAVE_MALLOC_H |
| 48 | #include <malloc.h> |
| 49 | #endif |
| 50 | #ifdef HAVE_LIMITS_H |
| 51 | #include <limits.h> |
| 52 | #endif |
| 53 | #ifdef HAVE_STDINT_H |
| 54 | #include <stdint.h> |
| 55 | #endif |
| 56 | |
| 57 | #include <stdio.h> |
| 58 | |
| 59 | #include "libiberty.h" |
| 60 | #include "ansidecl.h" |
| 61 | #include "hashtab.h" |
| 62 | |
| 63 | #ifndef CHAR_BIT |
| 64 | #define CHAR_BIT 8 |
| 65 | #endif |
| 66 | |
| 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); |
| 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 | |
| 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 | } |
| 101 | |
| 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 |
| 170 | higher_prime_index (unsigned long n) |
| 171 | { |
| 172 | unsigned int low = 0; |
| 173 | unsigned int high = sizeof(prime_tab) / sizeof(prime_tab[0]); |
| 174 | |
| 175 | while (low != high) |
| 176 | { |
| 177 | unsigned int mid = low + (high - low) / 2; |
| 178 | if (n > prime_tab[mid].prime) |
| 179 | low = mid + 1; |
| 180 | else |
| 181 | high = mid; |
| 182 | } |
| 183 | |
| 184 | /* If we've run out of primes, abort. */ |
| 185 | if (n > prime_tab[low].prime) |
| 186 | { |
| 187 | fprintf (stderr, "Cannot find prime bigger than %lu\n", n); |
| 188 | abort (); |
| 189 | } |
| 190 | |
| 191 | return low; |
| 192 | } |
| 193 | |
| 194 | /* Returns a hash code for P. */ |
| 195 | |
| 196 | static hashval_t |
| 197 | hash_pointer (const PTR p) |
| 198 | { |
| 199 | return (hashval_t) ((long)p >> 3); |
| 200 | } |
| 201 | |
| 202 | /* Returns non-zero if P1 and P2 are equal. */ |
| 203 | |
| 204 | static int |
| 205 | eq_pointer (const PTR p1, const PTR p2) |
| 206 | { |
| 207 | return p1 == p2; |
| 208 | } |
| 209 | |
| 210 | |
| 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) |
| 222 | { |
| 223 | return htab_size (htab); |
| 224 | } |
| 225 | |
| 226 | /* Return the current number of elements in given hash table. */ |
| 227 | |
| 228 | #define htab_elements(htab) ((htab)->n_elements - (htab)->n_deleted) |
| 229 | |
| 230 | size_t |
| 231 | (htab_elements) (htab_t htab) |
| 232 | { |
| 233 | return htab_elements (htab); |
| 234 | } |
| 235 | |
| 236 | /* Return X % Y. */ |
| 237 | |
| 238 | static inline hashval_t |
| 239 | htab_mod_1 (hashval_t x, hashval_t y, hashval_t inv, int shift) |
| 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 | |
| 264 | /* Compute the primary hash for HASH given HTAB's current size. */ |
| 265 | |
| 266 | static inline hashval_t |
| 267 | htab_mod (hashval_t hash, htab_t htab) |
| 268 | { |
| 269 | const struct prime_ent *p = &prime_tab[htab->size_prime_index]; |
| 270 | return htab_mod_1 (hash, p->prime, p->inv, p->shift); |
| 271 | } |
| 272 | |
| 273 | /* Compute the secondary hash for HASH given HTAB's current size. */ |
| 274 | |
| 275 | static inline hashval_t |
| 276 | htab_mod_m2 (hashval_t hash, htab_t htab) |
| 277 | { |
| 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); |
| 280 | } |
| 281 | |
| 282 | /* This function creates table with length slightly longer than given |
| 283 | source length. Created hash table is initiated as empty (all the |
| 284 | hash table entries are HTAB_EMPTY_ENTRY). The function returns the |
| 285 | created hash table, or NULL if memory allocation fails. */ |
| 286 | |
| 287 | htab_t |
| 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) |
| 290 | { |
| 291 | htab_t result; |
| 292 | unsigned int size_prime_index; |
| 293 | |
| 294 | size_prime_index = higher_prime_index (size); |
| 295 | size = prime_tab[size_prime_index].prime; |
| 296 | |
| 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 | } |
| 307 | result->size = size; |
| 308 | result->size_prime_index = size_prime_index; |
| 309 | result->hash_f = hash_f; |
| 310 | result->eq_f = eq_f; |
| 311 | result->del_f = del_f; |
| 312 | result->alloc_f = alloc_f; |
| 313 | result->free_f = free_f; |
| 314 | return result; |
| 315 | } |
| 316 | |
| 317 | /* As above, but use the variants of alloc_f and free_f which accept |
| 318 | an extra argument. */ |
| 319 | |
| 320 | htab_t |
| 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) |
| 325 | { |
| 326 | htab_t result; |
| 327 | unsigned int size_prime_index; |
| 328 | |
| 329 | size_prime_index = higher_prime_index (size); |
| 330 | size = prime_tab[size_prime_index].prime; |
| 331 | |
| 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; |
| 343 | result->size_prime_index = size_prime_index; |
| 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 |
| 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) |
| 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 | |
| 368 | /* These functions exist solely for backward compatibility. */ |
| 369 | |
| 370 | #undef htab_create |
| 371 | htab_t |
| 372 | htab_create (size_t size, htab_hash hash_f, htab_eq eq_f, htab_del del_f) |
| 373 | { |
| 374 | return htab_create_alloc (size, hash_f, eq_f, del_f, xcalloc, free); |
| 375 | } |
| 376 | |
| 377 | htab_t |
| 378 | htab_try_create (size_t size, htab_hash hash_f, htab_eq eq_f, htab_del del_f) |
| 379 | { |
| 380 | return htab_create_alloc (size, hash_f, eq_f, del_f, calloc, free); |
| 381 | } |
| 382 | |
| 383 | /* This function frees all memory allocated for given hash table. |
| 384 | Naturally the hash table must already exist. */ |
| 385 | |
| 386 | void |
| 387 | htab_delete (htab_t htab) |
| 388 | { |
| 389 | size_t size = htab_size (htab); |
| 390 | PTR *entries = htab->entries; |
| 391 | int i; |
| 392 | |
| 393 | if (htab->del_f) |
| 394 | for (i = size - 1; i >= 0; i--) |
| 395 | if (entries[i] != HTAB_EMPTY_ENTRY && entries[i] != HTAB_DELETED_ENTRY) |
| 396 | (*htab->del_f) (entries[i]); |
| 397 | |
| 398 | if (htab->free_f != NULL) |
| 399 | { |
| 400 | (*htab->free_f) (entries); |
| 401 | (*htab->free_f) (htab); |
| 402 | } |
| 403 | else if (htab->free_with_arg_f != NULL) |
| 404 | { |
| 405 | (*htab->free_with_arg_f) (htab->alloc_arg, entries); |
| 406 | (*htab->free_with_arg_f) (htab->alloc_arg, htab); |
| 407 | } |
| 408 | } |
| 409 | |
| 410 | /* This function clears all entries in the given hash table. */ |
| 411 | |
| 412 | void |
| 413 | htab_empty (htab_t htab) |
| 414 | { |
| 415 | size_t size = htab_size (htab); |
| 416 | PTR *entries = htab->entries; |
| 417 | int i; |
| 418 | |
| 419 | if (htab->del_f) |
| 420 | for (i = size - 1; i >= 0; i--) |
| 421 | if (entries[i] != HTAB_EMPTY_ENTRY && entries[i] != HTAB_DELETED_ENTRY) |
| 422 | (*htab->del_f) (entries[i]); |
| 423 | |
| 424 | memset (entries, 0, size * sizeof (PTR)); |
| 425 | } |
| 426 | |
| 427 | /* Similar to htab_find_slot, but without several unwanted side effects: |
| 428 | - Does not call htab->eq_f when it finds an existing entry. |
| 429 | - Does not change the count of elements/searches/collisions in the |
| 430 | hash table. |
| 431 | This function also assumes there are no deleted entries in the table. |
| 432 | HASH is the hash value for the element to be inserted. */ |
| 433 | |
| 434 | static PTR * |
| 435 | find_empty_slot_for_expand (htab_t htab, hashval_t hash) |
| 436 | { |
| 437 | hashval_t index = htab_mod (hash, htab); |
| 438 | size_t size = htab_size (htab); |
| 439 | PTR *slot = htab->entries + index; |
| 440 | hashval_t hash2; |
| 441 | |
| 442 | if (*slot == HTAB_EMPTY_ENTRY) |
| 443 | return slot; |
| 444 | else if (*slot == HTAB_DELETED_ENTRY) |
| 445 | abort (); |
| 446 | |
| 447 | hash2 = htab_mod_m2 (hash, htab); |
| 448 | for (;;) |
| 449 | { |
| 450 | index += hash2; |
| 451 | if (index >= size) |
| 452 | index -= size; |
| 453 | |
| 454 | slot = htab->entries + index; |
| 455 | if (*slot == HTAB_EMPTY_ENTRY) |
| 456 | return slot; |
| 457 | else if (*slot == HTAB_DELETED_ENTRY) |
| 458 | abort (); |
| 459 | } |
| 460 | } |
| 461 | |
| 462 | /* The following function changes size of memory allocated for the |
| 463 | entries and repeatedly inserts the table elements. The occupancy |
| 464 | of the table after the call will be about 50%. Naturally the hash |
| 465 | table must already exist. Remember also that the place of the |
| 466 | table entries is changed. If memory allocation failures are allowed, |
| 467 | this function will return zero, indicating that the table could not be |
| 468 | expanded. If all goes well, it will return a non-zero value. */ |
| 469 | |
| 470 | static int |
| 471 | htab_expand (htab_t htab) |
| 472 | { |
| 473 | PTR *oentries; |
| 474 | PTR *olimit; |
| 475 | PTR *p; |
| 476 | PTR *nentries; |
| 477 | size_t nsize, osize, elts; |
| 478 | unsigned int oindex, nindex; |
| 479 | |
| 480 | oentries = htab->entries; |
| 481 | oindex = htab->size_prime_index; |
| 482 | osize = htab->size; |
| 483 | olimit = oentries + osize; |
| 484 | elts = htab_elements (htab); |
| 485 | |
| 486 | /* Resize only when table after removal of unused elements is either |
| 487 | too full or too empty. */ |
| 488 | if (elts * 2 > osize || (elts * 8 < osize && osize > 32)) |
| 489 | { |
| 490 | nindex = higher_prime_index (elts * 2); |
| 491 | nsize = prime_tab[nindex].prime; |
| 492 | } |
| 493 | else |
| 494 | { |
| 495 | nindex = oindex; |
| 496 | nsize = osize; |
| 497 | } |
| 498 | |
| 499 | if (htab->alloc_with_arg_f != NULL) |
| 500 | nentries = (PTR *) (*htab->alloc_with_arg_f) (htab->alloc_arg, nsize, |
| 501 | sizeof (PTR *)); |
| 502 | else |
| 503 | nentries = (PTR *) (*htab->alloc_f) (nsize, sizeof (PTR *)); |
| 504 | if (nentries == NULL) |
| 505 | return 0; |
| 506 | htab->entries = nentries; |
| 507 | htab->size = nsize; |
| 508 | htab->size_prime_index = nindex; |
| 509 | htab->n_elements -= htab->n_deleted; |
| 510 | htab->n_deleted = 0; |
| 511 | |
| 512 | p = oentries; |
| 513 | do |
| 514 | { |
| 515 | PTR x = *p; |
| 516 | |
| 517 | if (x != HTAB_EMPTY_ENTRY && x != HTAB_DELETED_ENTRY) |
| 518 | { |
| 519 | PTR *q = find_empty_slot_for_expand (htab, (*htab->hash_f) (x)); |
| 520 | |
| 521 | *q = x; |
| 522 | } |
| 523 | |
| 524 | p++; |
| 525 | } |
| 526 | while (p < olimit); |
| 527 | |
| 528 | if (htab->free_f != NULL) |
| 529 | (*htab->free_f) (oentries); |
| 530 | else if (htab->free_with_arg_f != NULL) |
| 531 | (*htab->free_with_arg_f) (htab->alloc_arg, oentries); |
| 532 | return 1; |
| 533 | } |
| 534 | |
| 535 | /* This function searches for a hash table entry equal to the given |
| 536 | element. It cannot be used to insert or delete an element. */ |
| 537 | |
| 538 | PTR |
| 539 | htab_find_with_hash (htab_t htab, const PTR element, hashval_t hash) |
| 540 | { |
| 541 | hashval_t index, hash2; |
| 542 | size_t size; |
| 543 | PTR entry; |
| 544 | |
| 545 | htab->searches++; |
| 546 | size = htab_size (htab); |
| 547 | index = htab_mod (hash, htab); |
| 548 | |
| 549 | entry = htab->entries[index]; |
| 550 | if (entry == HTAB_EMPTY_ENTRY |
| 551 | || (entry != HTAB_DELETED_ENTRY && (*htab->eq_f) (entry, element))) |
| 552 | return entry; |
| 553 | |
| 554 | hash2 = htab_mod_m2 (hash, htab); |
| 555 | for (;;) |
| 556 | { |
| 557 | htab->collisions++; |
| 558 | index += hash2; |
| 559 | if (index >= size) |
| 560 | index -= size; |
| 561 | |
| 562 | entry = htab->entries[index]; |
| 563 | if (entry == HTAB_EMPTY_ENTRY |
| 564 | || (entry != HTAB_DELETED_ENTRY && (*htab->eq_f) (entry, element))) |
| 565 | return entry; |
| 566 | } |
| 567 | } |
| 568 | |
| 569 | /* Like htab_find_slot_with_hash, but compute the hash value from the |
| 570 | element. */ |
| 571 | |
| 572 | PTR |
| 573 | htab_find (htab_t htab, const PTR element) |
| 574 | { |
| 575 | return htab_find_with_hash (htab, element, (*htab->hash_f) (element)); |
| 576 | } |
| 577 | |
| 578 | /* This function searches for a hash table slot containing an entry |
| 579 | equal to the given element. To delete an entry, call this with |
| 580 | insert=NO_INSERT, then call htab_clear_slot on the slot returned |
| 581 | (possibly after doing some checks). To insert an entry, call this |
| 582 | with insert=INSERT, then write the value you want into the returned |
| 583 | slot. When inserting an entry, NULL may be returned if memory |
| 584 | allocation fails. */ |
| 585 | |
| 586 | PTR * |
| 587 | htab_find_slot_with_hash (htab_t htab, const PTR element, |
| 588 | hashval_t hash, enum insert_option insert) |
| 589 | { |
| 590 | PTR *first_deleted_slot; |
| 591 | hashval_t index, hash2; |
| 592 | size_t size; |
| 593 | PTR entry; |
| 594 | |
| 595 | size = htab_size (htab); |
| 596 | if (insert == INSERT && size * 3 <= htab->n_elements * 4) |
| 597 | { |
| 598 | if (htab_expand (htab) == 0) |
| 599 | return NULL; |
| 600 | size = htab_size (htab); |
| 601 | } |
| 602 | |
| 603 | index = htab_mod (hash, htab); |
| 604 | |
| 605 | htab->searches++; |
| 606 | first_deleted_slot = NULL; |
| 607 | |
| 608 | entry = htab->entries[index]; |
| 609 | if (entry == HTAB_EMPTY_ENTRY) |
| 610 | goto empty_entry; |
| 611 | else if (entry == HTAB_DELETED_ENTRY) |
| 612 | first_deleted_slot = &htab->entries[index]; |
| 613 | else if ((*htab->eq_f) (entry, element)) |
| 614 | return &htab->entries[index]; |
| 615 | |
| 616 | hash2 = htab_mod_m2 (hash, htab); |
| 617 | for (;;) |
| 618 | { |
| 619 | htab->collisions++; |
| 620 | index += hash2; |
| 621 | if (index >= size) |
| 622 | index -= size; |
| 623 | |
| 624 | entry = htab->entries[index]; |
| 625 | if (entry == HTAB_EMPTY_ENTRY) |
| 626 | goto empty_entry; |
| 627 | else if (entry == HTAB_DELETED_ENTRY) |
| 628 | { |
| 629 | if (!first_deleted_slot) |
| 630 | first_deleted_slot = &htab->entries[index]; |
| 631 | } |
| 632 | else if ((*htab->eq_f) (entry, element)) |
| 633 | return &htab->entries[index]; |
| 634 | } |
| 635 | |
| 636 | empty_entry: |
| 637 | if (insert == NO_INSERT) |
| 638 | return NULL; |
| 639 | |
| 640 | if (first_deleted_slot) |
| 641 | { |
| 642 | htab->n_deleted--; |
| 643 | *first_deleted_slot = HTAB_EMPTY_ENTRY; |
| 644 | return first_deleted_slot; |
| 645 | } |
| 646 | |
| 647 | htab->n_elements++; |
| 648 | return &htab->entries[index]; |
| 649 | } |
| 650 | |
| 651 | /* Like htab_find_slot_with_hash, but compute the hash value from the |
| 652 | element. */ |
| 653 | |
| 654 | PTR * |
| 655 | htab_find_slot (htab_t htab, const PTR element, enum insert_option insert) |
| 656 | { |
| 657 | return htab_find_slot_with_hash (htab, element, (*htab->hash_f) (element), |
| 658 | insert); |
| 659 | } |
| 660 | |
| 661 | /* This function deletes an element with the given value from hash |
| 662 | table (the hash is computed from the element). If there is no matching |
| 663 | element in the hash table, this function does nothing. */ |
| 664 | |
| 665 | void |
| 666 | htab_remove_elt (htab_t htab, PTR element) |
| 667 | { |
| 668 | htab_remove_elt_with_hash (htab, element, (*htab->hash_f) (element)); |
| 669 | } |
| 670 | |
| 671 | |
| 672 | /* This function deletes an element with the given value from hash |
| 673 | table. If there is no matching element in the hash table, this |
| 674 | function does nothing. */ |
| 675 | |
| 676 | void |
| 677 | htab_remove_elt_with_hash (htab_t htab, PTR element, hashval_t hash) |
| 678 | { |
| 679 | PTR *slot; |
| 680 | |
| 681 | slot = htab_find_slot_with_hash (htab, element, hash, NO_INSERT); |
| 682 | if (*slot == HTAB_EMPTY_ENTRY) |
| 683 | return; |
| 684 | |
| 685 | if (htab->del_f) |
| 686 | (*htab->del_f) (*slot); |
| 687 | |
| 688 | *slot = HTAB_DELETED_ENTRY; |
| 689 | htab->n_deleted++; |
| 690 | } |
| 691 | |
| 692 | /* This function clears a specified slot in a hash table. It is |
| 693 | useful when you've already done the lookup and don't want to do it |
| 694 | again. */ |
| 695 | |
| 696 | void |
| 697 | htab_clear_slot (htab_t htab, PTR *slot) |
| 698 | { |
| 699 | if (slot < htab->entries || slot >= htab->entries + htab_size (htab) |
| 700 | || *slot == HTAB_EMPTY_ENTRY || *slot == HTAB_DELETED_ENTRY) |
| 701 | abort (); |
| 702 | |
| 703 | if (htab->del_f) |
| 704 | (*htab->del_f) (*slot); |
| 705 | |
| 706 | *slot = HTAB_DELETED_ENTRY; |
| 707 | htab->n_deleted++; |
| 708 | } |
| 709 | |
| 710 | /* This function scans over the entire hash table calling |
| 711 | CALLBACK for each live entry. If CALLBACK returns false, |
| 712 | the iteration stops. INFO is passed as CALLBACK's second |
| 713 | argument. */ |
| 714 | |
| 715 | void |
| 716 | htab_traverse_noresize (htab_t htab, htab_trav callback, PTR info) |
| 717 | { |
| 718 | PTR *slot; |
| 719 | PTR *limit; |
| 720 | |
| 721 | slot = htab->entries; |
| 722 | limit = slot + htab_size (htab); |
| 723 | |
| 724 | do |
| 725 | { |
| 726 | PTR x = *slot; |
| 727 | |
| 728 | if (x != HTAB_EMPTY_ENTRY && x != HTAB_DELETED_ENTRY) |
| 729 | if (!(*callback) (slot, info)) |
| 730 | break; |
| 731 | } |
| 732 | while (++slot < limit); |
| 733 | } |
| 734 | |
| 735 | /* Like htab_traverse_noresize, but does resize the table when it is |
| 736 | too empty to improve effectivity of subsequent calls. */ |
| 737 | |
| 738 | void |
| 739 | htab_traverse (htab_t htab, htab_trav callback, PTR info) |
| 740 | { |
| 741 | if (htab_elements (htab) * 8 < htab_size (htab)) |
| 742 | htab_expand (htab); |
| 743 | |
| 744 | htab_traverse_noresize (htab, callback, info); |
| 745 | } |
| 746 | |
| 747 | /* Return the fraction of fixed collisions during all work with given |
| 748 | hash table. */ |
| 749 | |
| 750 | double |
| 751 | htab_collisions (htab_t htab) |
| 752 | { |
| 753 | if (htab->searches == 0) |
| 754 | return 0.0; |
| 755 | |
| 756 | return (double) htab->collisions / (double) htab->searches; |
| 757 | } |
| 758 | |
| 759 | /* Hash P as a null-terminated string. |
| 760 | |
| 761 | Copied from gcc/hashtable.c. Zack had the following to say with respect |
| 762 | to applicability, though note that unlike hashtable.c, this hash table |
| 763 | implementation re-hashes rather than chain buckets. |
| 764 | |
| 765 | http://gcc.gnu.org/ml/gcc-patches/2001-08/msg01021.html |
| 766 | From: Zack Weinberg <zackw@panix.com> |
| 767 | Date: Fri, 17 Aug 2001 02:15:56 -0400 |
| 768 | |
| 769 | I got it by extracting all the identifiers from all the source code |
| 770 | I had lying around in mid-1999, and testing many recurrences of |
| 771 | the form "H_n = H_{n-1} * K + c_n * L + M" where K, L, M were either |
| 772 | prime numbers or the appropriate identity. This was the best one. |
| 773 | I don't remember exactly what constituted "best", except I was |
| 774 | looking at bucket-length distributions mostly. |
| 775 | |
| 776 | So it should be very good at hashing identifiers, but might not be |
| 777 | as good at arbitrary strings. |
| 778 | |
| 779 | I'll add that it thoroughly trounces the hash functions recommended |
| 780 | for this use at http://burtleburtle.net/bob/hash/index.html, both |
| 781 | on speed and bucket distribution. I haven't tried it against the |
| 782 | function they just started using for Perl's hashes. */ |
| 783 | |
| 784 | hashval_t |
| 785 | htab_hash_string (const PTR p) |
| 786 | { |
| 787 | const unsigned char *str = (const unsigned char *) p; |
| 788 | hashval_t r = 0; |
| 789 | unsigned char c; |
| 790 | |
| 791 | while ((c = *str++) != 0) |
| 792 | r = r * 67 + c - 113; |
| 793 | |
| 794 | return r; |
| 795 | } |
| 796 | |
| 797 | /* DERIVED FROM: |
| 798 | -------------------------------------------------------------------- |
| 799 | lookup2.c, by Bob Jenkins, December 1996, Public Domain. |
| 800 | hash(), hash2(), hash3, and mix() are externally useful functions. |
| 801 | Routines to test the hash are included if SELF_TEST is defined. |
| 802 | You can use this free for any purpose. It has no warranty. |
| 803 | -------------------------------------------------------------------- |
| 804 | */ |
| 805 | |
| 806 | /* |
| 807 | -------------------------------------------------------------------- |
| 808 | mix -- mix 3 32-bit values reversibly. |
| 809 | For every delta with one or two bit set, and the deltas of all three |
| 810 | high bits or all three low bits, whether the original value of a,b,c |
| 811 | is almost all zero or is uniformly distributed, |
| 812 | * If mix() is run forward or backward, at least 32 bits in a,b,c |
| 813 | have at least 1/4 probability of changing. |
| 814 | * If mix() is run forward, every bit of c will change between 1/3 and |
| 815 | 2/3 of the time. (Well, 22/100 and 78/100 for some 2-bit deltas.) |
| 816 | mix() was built out of 36 single-cycle latency instructions in a |
| 817 | structure that could supported 2x parallelism, like so: |
| 818 | a -= b; |
| 819 | a -= c; x = (c>>13); |
| 820 | b -= c; a ^= x; |
| 821 | b -= a; x = (a<<8); |
| 822 | c -= a; b ^= x; |
| 823 | c -= b; x = (b>>13); |
| 824 | ... |
| 825 | Unfortunately, superscalar Pentiums and Sparcs can't take advantage |
| 826 | of that parallelism. They've also turned some of those single-cycle |
| 827 | latency instructions into multi-cycle latency instructions. Still, |
| 828 | this is the fastest good hash I could find. There were about 2^^68 |
| 829 | to choose from. I only looked at a billion or so. |
| 830 | -------------------------------------------------------------------- |
| 831 | */ |
| 832 | /* same, but slower, works on systems that might have 8 byte hashval_t's */ |
| 833 | #define mix(a,b,c) \ |
| 834 | { \ |
| 835 | a -= b; a -= c; a ^= (c>>13); \ |
| 836 | b -= c; b -= a; b ^= (a<< 8); \ |
| 837 | c -= a; c -= b; c ^= ((b&0xffffffff)>>13); \ |
| 838 | a -= b; a -= c; a ^= ((c&0xffffffff)>>12); \ |
| 839 | b -= c; b -= a; b = (b ^ (a<<16)) & 0xffffffff; \ |
| 840 | c -= a; c -= b; c = (c ^ (b>> 5)) & 0xffffffff; \ |
| 841 | a -= b; a -= c; a = (a ^ (c>> 3)) & 0xffffffff; \ |
| 842 | b -= c; b -= a; b = (b ^ (a<<10)) & 0xffffffff; \ |
| 843 | c -= a; c -= b; c = (c ^ (b>>15)) & 0xffffffff; \ |
| 844 | } |
| 845 | |
| 846 | /* |
| 847 | -------------------------------------------------------------------- |
| 848 | hash() -- hash a variable-length key into a 32-bit value |
| 849 | k : the key (the unaligned variable-length array of bytes) |
| 850 | len : the length of the key, counting by bytes |
| 851 | level : can be any 4-byte value |
| 852 | Returns a 32-bit value. Every bit of the key affects every bit of |
| 853 | the return value. Every 1-bit and 2-bit delta achieves avalanche. |
| 854 | About 36+6len instructions. |
| 855 | |
| 856 | The best hash table sizes are powers of 2. There is no need to do |
| 857 | mod a prime (mod is sooo slow!). If you need less than 32 bits, |
| 858 | use a bitmask. For example, if you need only 10 bits, do |
| 859 | h = (h & hashmask(10)); |
| 860 | In which case, the hash table should have hashsize(10) elements. |
| 861 | |
| 862 | If you are hashing n strings (ub1 **)k, do it like this: |
| 863 | for (i=0, h=0; i<n; ++i) h = hash( k[i], len[i], h); |
| 864 | |
| 865 | By Bob Jenkins, 1996. bob_jenkins@burtleburtle.net. You may use this |
| 866 | code any way you wish, private, educational, or commercial. It's free. |
| 867 | |
| 868 | See http://burtleburtle.net/bob/hash/evahash.html |
| 869 | Use for hash table lookup, or anything where one collision in 2^32 is |
| 870 | acceptable. Do NOT use for cryptographic purposes. |
| 871 | -------------------------------------------------------------------- |
| 872 | */ |
| 873 | |
| 874 | hashval_t |
| 875 | iterative_hash (const PTR k_in /* the key */, |
| 876 | register size_t length /* the length of the key */, |
| 877 | register hashval_t initval /* the previous hash, or |
| 878 | an arbitrary value */) |
| 879 | { |
| 880 | register const unsigned char *k = (const unsigned char *)k_in; |
| 881 | register hashval_t a,b,c,len; |
| 882 | |
| 883 | /* Set up the internal state */ |
| 884 | len = length; |
| 885 | a = b = 0x9e3779b9; /* the golden ratio; an arbitrary value */ |
| 886 | c = initval; /* the previous hash value */ |
| 887 | |
| 888 | /*---------------------------------------- handle most of the key */ |
| 889 | #ifndef WORDS_BIGENDIAN |
| 890 | /* On a little-endian machine, if the data is 4-byte aligned we can hash |
| 891 | by word for better speed. This gives nondeterministic results on |
| 892 | big-endian machines. */ |
| 893 | if (sizeof (hashval_t) == 4 && (((size_t)k)&3) == 0) |
| 894 | while (len >= 12) /* aligned */ |
| 895 | { |
| 896 | a += *(hashval_t *)(k+0); |
| 897 | b += *(hashval_t *)(k+4); |
| 898 | c += *(hashval_t *)(k+8); |
| 899 | mix(a,b,c); |
| 900 | k += 12; len -= 12; |
| 901 | } |
| 902 | else /* unaligned */ |
| 903 | #endif |
| 904 | while (len >= 12) |
| 905 | { |
| 906 | a += (k[0] +((hashval_t)k[1]<<8) +((hashval_t)k[2]<<16) +((hashval_t)k[3]<<24)); |
| 907 | b += (k[4] +((hashval_t)k[5]<<8) +((hashval_t)k[6]<<16) +((hashval_t)k[7]<<24)); |
| 908 | c += (k[8] +((hashval_t)k[9]<<8) +((hashval_t)k[10]<<16)+((hashval_t)k[11]<<24)); |
| 909 | mix(a,b,c); |
| 910 | k += 12; len -= 12; |
| 911 | } |
| 912 | |
| 913 | /*------------------------------------- handle the last 11 bytes */ |
| 914 | c += length; |
| 915 | switch(len) /* all the case statements fall through */ |
| 916 | { |
| 917 | case 11: c+=((hashval_t)k[10]<<24); |
| 918 | case 10: c+=((hashval_t)k[9]<<16); |
| 919 | case 9 : c+=((hashval_t)k[8]<<8); |
| 920 | /* the first byte of c is reserved for the length */ |
| 921 | case 8 : b+=((hashval_t)k[7]<<24); |
| 922 | case 7 : b+=((hashval_t)k[6]<<16); |
| 923 | case 6 : b+=((hashval_t)k[5]<<8); |
| 924 | case 5 : b+=k[4]; |
| 925 | case 4 : a+=((hashval_t)k[3]<<24); |
| 926 | case 3 : a+=((hashval_t)k[2]<<16); |
| 927 | case 2 : a+=((hashval_t)k[1]<<8); |
| 928 | case 1 : a+=k[0]; |
| 929 | /* case 0: nothing left to add */ |
| 930 | } |
| 931 | mix(a,b,c); |
| 932 | /*-------------------------------------------- report the result */ |
| 933 | return c; |
| 934 | } |