2 Copyright (c) 2003-2014, Troy D. Hanson http://troydhanson.github.com/uthash/
5 Redistribution and use in source and binary forms, with or without
6 modification, are permitted provided that the following conditions are met:
8 * Redistributions of source code must retain the above copyright
9 notice, this list of conditions and the following disclaimer.
11 THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS
12 IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
13 TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A
14 PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER
15 OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
16 EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
17 PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
18 PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
19 LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
20 NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
21 SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27 #include <string.h> /* memcmp,strlen */
28 #include <stddef.h> /* ptrdiff_t */
29 #include <stdlib.h> /* exit() */
31 /* These macros use decltype or the earlier __typeof GNU extension.
32 As decltype is only available in newer compilers (VS2010 or gcc 4.3+
33 when compiling c++ source) this code uses whatever method is needed
34 or, for VS2008 where neither is available, uses casting workarounds. */
35 #if defined(_MSC_VER) /* MS compiler */
36 #if _MSC_VER >= 1600 && defined(__cplusplus) /* VS2010 or newer in C++ mode */
37 #define DECLTYPE(x) (decltype(x))
38 #else /* VS2008 or older (or VS2010 in C mode) */
42 #elif defined(__BORLANDC__) || defined(__LCC__) || defined(__WATCOMC__)
45 #else /* GNU, Sun and other compilers */
46 #define DECLTYPE(x) (__typeof(x))
50 #define DECLTYPE_ASSIGN(dst,src) \
52 char **_da_dst = (char**)(&(dst)); \
53 *_da_dst = (char*)(src); \
56 #define DECLTYPE_ASSIGN(dst,src) \
58 (dst) = DECLTYPE(dst)(src); \
62 /* a number of the hash function use uint32_t which isn't defined on Pre VS2010 */
64 #if defined(_MSC_VER) && _MSC_VER >= 1600
66 #elif defined(__WATCOMC__)
69 typedef unsigned int uint32_t;
70 typedef unsigned char uint8_t;
76 #define UTHASH_VERSION 1.9.9
79 #define uthash_fatal(msg) exit(-1) /* fatal error (out of memory,etc) */
82 #define uthash_malloc(sz) malloc(sz) /* malloc fcn */
85 #define uthash_free(ptr,sz) free(ptr) /* free fcn */
88 #ifndef uthash_noexpand_fyi
89 #define uthash_noexpand_fyi(tbl) /* can be defined to log noexpand */
91 #ifndef uthash_expand_fyi
92 #define uthash_expand_fyi(tbl) /* can be defined to log expands */
95 /* initial number of buckets */
96 #define HASH_INITIAL_NUM_BUCKETS 32 /* initial number of buckets */
97 #define HASH_INITIAL_NUM_BUCKETS_LOG2 5 /* lg2 of initial number of buckets */
98 #define HASH_BKT_CAPACITY_THRESH 10 /* expand when bucket count reaches */
100 /* calculate the element whose hash handle address is hhe */
101 #define ELMT_FROM_HH(tbl,hhp) ((void*)(((char*)(hhp)) - ((tbl)->hho)))
103 #define HASH_FIND(hh,head,keyptr,keylen,out) \
105 unsigned _hf_bkt,_hf_hashv; \
108 HASH_FCN(keyptr,keylen, (head)->hh.tbl->num_buckets, _hf_hashv, _hf_bkt); \
109 if (HASH_BLOOM_TEST((head)->hh.tbl, _hf_hashv)) { \
110 HASH_FIND_IN_BKT((head)->hh.tbl, hh, (head)->hh.tbl->buckets[ _hf_bkt ], \
111 keyptr,keylen,out); \
117 #define HASH_BLOOM_BITLEN (1ULL << HASH_BLOOM)
118 #define HASH_BLOOM_BYTELEN (HASH_BLOOM_BITLEN/8) + ((HASH_BLOOM_BITLEN%8) ? 1:0)
119 #define HASH_BLOOM_MAKE(tbl) \
121 (tbl)->bloom_nbits = HASH_BLOOM; \
122 (tbl)->bloom_bv = (uint8_t*)uthash_malloc(HASH_BLOOM_BYTELEN); \
123 if (!((tbl)->bloom_bv)) { uthash_fatal( "out of memory"); } \
124 memset((tbl)->bloom_bv, 0, HASH_BLOOM_BYTELEN); \
125 (tbl)->bloom_sig = HASH_BLOOM_SIGNATURE; \
128 #define HASH_BLOOM_FREE(tbl) \
130 uthash_free((tbl)->bloom_bv, HASH_BLOOM_BYTELEN); \
133 #define HASH_BLOOM_BITSET(bv,idx) (bv[(idx)/8] |= (1U << ((idx)%8)))
134 #define HASH_BLOOM_BITTEST(bv,idx) (bv[(idx)/8] & (1U << ((idx)%8)))
136 #define HASH_BLOOM_ADD(tbl,hashv) \
137 HASH_BLOOM_BITSET((tbl)->bloom_bv, (hashv & (uint32_t)((1ULL << (tbl)->bloom_nbits) - 1)))
139 #define HASH_BLOOM_TEST(tbl,hashv) \
140 HASH_BLOOM_BITTEST((tbl)->bloom_bv, (hashv & (uint32_t)((1ULL << (tbl)->bloom_nbits) - 1)))
143 #define HASH_BLOOM_MAKE(tbl)
144 #define HASH_BLOOM_FREE(tbl)
145 #define HASH_BLOOM_ADD(tbl,hashv)
146 #define HASH_BLOOM_TEST(tbl,hashv) (1)
147 #define HASH_BLOOM_BYTELEN 0
150 #define HASH_MAKE_TABLE(hh,head) \
152 (head)->hh.tbl = (UT_hash_table*)uthash_malloc( \
153 sizeof(UT_hash_table)); \
154 if (!((head)->hh.tbl)) { uthash_fatal( "out of memory"); } \
155 memset((head)->hh.tbl, 0, sizeof(UT_hash_table)); \
156 (head)->hh.tbl->tail = &((head)->hh); \
157 (head)->hh.tbl->num_buckets = HASH_INITIAL_NUM_BUCKETS; \
158 (head)->hh.tbl->log2_num_buckets = HASH_INITIAL_NUM_BUCKETS_LOG2; \
159 (head)->hh.tbl->hho = (char*)(&(head)->hh) - (char*)(head); \
160 (head)->hh.tbl->buckets = (UT_hash_bucket*)uthash_malloc( \
161 HASH_INITIAL_NUM_BUCKETS*sizeof(struct UT_hash_bucket)); \
162 if (! (head)->hh.tbl->buckets) { uthash_fatal( "out of memory"); } \
163 memset((head)->hh.tbl->buckets, 0, \
164 HASH_INITIAL_NUM_BUCKETS*sizeof(struct UT_hash_bucket)); \
165 HASH_BLOOM_MAKE((head)->hh.tbl); \
166 (head)->hh.tbl->signature = HASH_SIGNATURE; \
169 #define HASH_ADD(hh,head,fieldname,keylen_in,add) \
170 HASH_ADD_KEYPTR(hh,head,&((add)->fieldname),keylen_in,add)
172 #define HASH_REPLACE(hh,head,fieldname,keylen_in,add,replaced) \
175 HASH_FIND(hh,head,&((add)->fieldname),keylen_in,replaced); \
176 if (replaced!=NULL) { \
177 HASH_DELETE(hh,head,replaced); \
179 HASH_ADD(hh,head,fieldname,keylen_in,add); \
182 #define HASH_ADD_KEYPTR(hh,head,keyptr,keylen_in,add) \
185 (add)->hh.next = NULL; \
186 (add)->hh.key = (char*)(keyptr); \
187 (add)->hh.keylen = (unsigned)(keylen_in); \
190 (head)->hh.prev = NULL; \
191 HASH_MAKE_TABLE(hh,head); \
193 (head)->hh.tbl->tail->next = (add); \
194 (add)->hh.prev = ELMT_FROM_HH((head)->hh.tbl, (head)->hh.tbl->tail); \
195 (head)->hh.tbl->tail = &((add)->hh); \
197 (head)->hh.tbl->num_items++; \
198 (add)->hh.tbl = (head)->hh.tbl; \
199 HASH_FCN(keyptr,keylen_in, (head)->hh.tbl->num_buckets, \
200 (add)->hh.hashv, _ha_bkt); \
201 HASH_ADD_TO_BKT((head)->hh.tbl->buckets[_ha_bkt],&(add)->hh); \
202 HASH_BLOOM_ADD((head)->hh.tbl,(add)->hh.hashv); \
203 HASH_EMIT_KEY(hh,head,keyptr,keylen_in); \
204 HASH_FSCK(hh,head); \
207 #define HASH_TO_BKT( hashv, num_bkts, bkt ) \
209 bkt = ((hashv) & ((num_bkts) - 1)); \
212 /* delete "delptr" from the hash table.
213 * "the usual" patch-up process for the app-order doubly-linked-list.
214 * The use of _hd_hh_del below deserves special explanation.
215 * These used to be expressed using (delptr) but that led to a bug
216 * if someone used the same symbol for the head and deletee, like
217 * HASH_DELETE(hh,users,users);
218 * We want that to work, but by changing the head (users) below
219 * we were forfeiting our ability to further refer to the deletee (users)
220 * in the patch-up process. Solution: use scratch space to
221 * copy the deletee pointer, then the latter references are via that
222 * scratch pointer rather than through the repointed (users) symbol.
224 #define HASH_DELETE(hh,head,delptr) \
227 struct UT_hash_handle *_hd_hh_del; \
228 if ( ((delptr)->hh.prev == NULL) && ((delptr)->hh.next == NULL) ) { \
229 uthash_free((head)->hh.tbl->buckets, \
230 (head)->hh.tbl->num_buckets*sizeof(struct UT_hash_bucket) ); \
231 HASH_BLOOM_FREE((head)->hh.tbl); \
232 uthash_free((head)->hh.tbl, sizeof(UT_hash_table)); \
235 _hd_hh_del = &((delptr)->hh); \
236 if ((delptr) == ELMT_FROM_HH((head)->hh.tbl,(head)->hh.tbl->tail)) { \
237 (head)->hh.tbl->tail = \
238 (UT_hash_handle*)((ptrdiff_t)((delptr)->hh.prev) + \
239 (head)->hh.tbl->hho); \
241 if ((delptr)->hh.prev) { \
242 ((UT_hash_handle*)((ptrdiff_t)((delptr)->hh.prev) + \
243 (head)->hh.tbl->hho))->next = (delptr)->hh.next; \
245 DECLTYPE_ASSIGN(head,(delptr)->hh.next); \
247 if (_hd_hh_del->next) { \
248 ((UT_hash_handle*)((ptrdiff_t)_hd_hh_del->next + \
249 (head)->hh.tbl->hho))->prev = \
252 HASH_TO_BKT( _hd_hh_del->hashv, (head)->hh.tbl->num_buckets, _hd_bkt); \
253 HASH_DEL_IN_BKT(hh,(head)->hh.tbl->buckets[_hd_bkt], _hd_hh_del); \
254 (head)->hh.tbl->num_items--; \
256 HASH_FSCK(hh,head); \
260 /* convenience forms of HASH_FIND/HASH_ADD/HASH_DEL */
261 #define HASH_FIND_STR(head,findstr,out) \
262 HASH_FIND(hh,head,findstr,strlen(findstr),out)
263 #define HASH_ADD_STR(head,strfield,add) \
264 HASH_ADD(hh,head,strfield[0],strlen(add->strfield),add)
265 #define HASH_REPLACE_STR(head,strfield,add,replaced) \
266 HASH_REPLACE(hh,head,strfield[0],strlen(add->strfield),add,replaced)
267 #define HASH_FIND_INT(head,findint,out) \
268 HASH_FIND(hh,head,findint,sizeof(int),out)
269 #define HASH_ADD_INT(head,intfield,add) \
270 HASH_ADD(hh,head,intfield,sizeof(int),add)
271 #define HASH_REPLACE_INT(head,intfield,add,replaced) \
272 HASH_REPLACE(hh,head,intfield,sizeof(int),add,replaced)
273 #define HASH_FIND_PTR(head,findptr,out) \
274 HASH_FIND(hh,head,findptr,sizeof(void *),out)
275 #define HASH_ADD_PTR(head,ptrfield,add) \
276 HASH_ADD(hh,head,ptrfield,sizeof(void *),add)
277 #define HASH_REPLACE_PTR(head,ptrfield,add,replaced) \
278 HASH_REPLACE(hh,head,ptrfield,sizeof(void *),add,replaced)
279 #define HASH_DEL(head,delptr) \
280 HASH_DELETE(hh,head,delptr)
282 /* HASH_FSCK checks hash integrity on every add/delete when HASH_DEBUG is defined.
283 * This is for uthash developer only; it compiles away if HASH_DEBUG isn't defined.
286 #define HASH_OOPS(...) do { fprintf(stderr,__VA_ARGS__); exit(-1); } while (0)
287 #define HASH_FSCK(hh,head) \
290 unsigned _count, _bkt_count; \
292 struct UT_hash_handle *_thh; \
295 for( _bkt_i = 0; _bkt_i < (head)->hh.tbl->num_buckets; _bkt_i++) { \
297 _thh = (head)->hh.tbl->buckets[_bkt_i].hh_head; \
300 if (_prev != (char*)(_thh->hh_prev)) { \
301 HASH_OOPS("invalid hh_prev %p, actual %p\n", \
302 _thh->hh_prev, _prev ); \
305 _prev = (char*)(_thh); \
306 _thh = _thh->hh_next; \
308 _count += _bkt_count; \
309 if ((head)->hh.tbl->buckets[_bkt_i].count != _bkt_count) { \
310 HASH_OOPS("invalid bucket count %d, actual %d\n", \
311 (head)->hh.tbl->buckets[_bkt_i].count, _bkt_count); \
314 if (_count != (head)->hh.tbl->num_items) { \
315 HASH_OOPS("invalid hh item count %d, actual %d\n", \
316 (head)->hh.tbl->num_items, _count ); \
318 /* traverse hh in app order; check next/prev integrity, count */ \
321 _thh = &(head)->hh; \
324 if (_prev !=(char*)(_thh->prev)) { \
325 HASH_OOPS("invalid prev %p, actual %p\n", \
326 _thh->prev, _prev ); \
328 _prev = (char*)ELMT_FROM_HH((head)->hh.tbl, _thh); \
329 _thh = ( _thh->next ? (UT_hash_handle*)((char*)(_thh->next) + \
330 (head)->hh.tbl->hho) : NULL ); \
332 if (_count != (head)->hh.tbl->num_items) { \
333 HASH_OOPS("invalid app item count %d, actual %d\n", \
334 (head)->hh.tbl->num_items, _count ); \
339 #define HASH_FSCK(hh,head)
342 /* When compiled with -DHASH_EMIT_KEYS, length-prefixed keys are emitted to
343 * the descriptor to which this macro is defined for tuning the hash function.
344 * The app can #include <unistd.h> to get the prototype for write(2). */
345 #ifdef HASH_EMIT_KEYS
346 #define HASH_EMIT_KEY(hh,head,keyptr,fieldlen) \
348 unsigned _klen = fieldlen; \
349 write(HASH_EMIT_KEYS, &_klen, sizeof(_klen)); \
350 write(HASH_EMIT_KEYS, keyptr, fieldlen); \
353 #define HASH_EMIT_KEY(hh,head,keyptr,fieldlen)
356 /* default to Jenkin's hash unless overridden e.g. DHASH_FUNCTION=HASH_SAX */
358 #define HASH_FCN HASH_FUNCTION
360 #define HASH_FCN HASH_JEN
363 /* The Bernstein hash function, used in Perl prior to v5.6. Note (x<<5+x)=x*33. */
364 #define HASH_BER(key,keylen,num_bkts,hashv,bkt) \
366 unsigned _hb_keylen=keylen; \
367 char *_hb_key=(char*)(key); \
369 while (_hb_keylen--) { (hashv) = (((hashv) << 5) + (hashv)) + *_hb_key++; } \
370 bkt = (hashv) & (num_bkts-1); \
374 /* SAX/FNV/OAT/JEN hash functions are macro variants of those listed at
375 * http://eternallyconfuzzled.com/tuts/algorithms/jsw_tut_hashing.aspx */
376 #define HASH_SAX(key,keylen,num_bkts,hashv,bkt) \
379 char *_hs_key=(char*)(key); \
381 for(_sx_i=0; _sx_i < keylen; _sx_i++) \
382 hashv ^= (hashv << 5) + (hashv >> 2) + _hs_key[_sx_i]; \
383 bkt = hashv & (num_bkts-1); \
385 /* FNV-1a variation */
386 #define HASH_FNV(key,keylen,num_bkts,hashv,bkt) \
389 char *_hf_key=(char*)(key); \
390 hashv = 2166136261UL; \
391 for(_fn_i=0; _fn_i < keylen; _fn_i++) \
392 hashv = hashv ^ _hf_key[_fn_i]; \
393 hashv = hashv * 16777619; \
394 bkt = hashv & (num_bkts-1); \
397 #define HASH_OAT(key,keylen,num_bkts,hashv,bkt) \
400 char *_ho_key=(char*)(key); \
402 for(_ho_i=0; _ho_i < keylen; _ho_i++) { \
403 hashv += _ho_key[_ho_i]; \
404 hashv += (hashv << 10); \
405 hashv ^= (hashv >> 6); \
407 hashv += (hashv << 3); \
408 hashv ^= (hashv >> 11); \
409 hashv += (hashv << 15); \
410 bkt = hashv & (num_bkts-1); \
413 #define HASH_JEN_MIX(a,b,c) \
415 a -= b; a -= c; a ^= ( c >> 13 ); \
416 b -= c; b -= a; b ^= ( a << 8 ); \
417 c -= a; c -= b; c ^= ( b >> 13 ); \
418 a -= b; a -= c; a ^= ( c >> 12 ); \
419 b -= c; b -= a; b ^= ( a << 16 ); \
420 c -= a; c -= b; c ^= ( b >> 5 ); \
421 a -= b; a -= c; a ^= ( c >> 3 ); \
422 b -= c; b -= a; b ^= ( a << 10 ); \
423 c -= a; c -= b; c ^= ( b >> 15 ); \
426 #define HASH_JEN(key,keylen,num_bkts,hashv,bkt) \
428 unsigned _hj_i,_hj_j,_hj_k; \
429 unsigned char *_hj_key=(unsigned char*)(key); \
430 hashv = 0xfeedbeef; \
431 _hj_i = _hj_j = 0x9e3779b9; \
432 _hj_k = (unsigned)(keylen); \
433 while (_hj_k >= 12) { \
434 _hj_i += (_hj_key[0] + ( (unsigned)_hj_key[1] << 8 ) \
435 + ( (unsigned)_hj_key[2] << 16 ) \
436 + ( (unsigned)_hj_key[3] << 24 ) ); \
437 _hj_j += (_hj_key[4] + ( (unsigned)_hj_key[5] << 8 ) \
438 + ( (unsigned)_hj_key[6] << 16 ) \
439 + ( (unsigned)_hj_key[7] << 24 ) ); \
440 hashv += (_hj_key[8] + ( (unsigned)_hj_key[9] << 8 ) \
441 + ( (unsigned)_hj_key[10] << 16 ) \
442 + ( (unsigned)_hj_key[11] << 24 ) ); \
444 HASH_JEN_MIX(_hj_i, _hj_j, hashv); \
451 case 11: hashv += ( (unsigned)_hj_key[10] << 24 ); \
452 case 10: hashv += ( (unsigned)_hj_key[9] << 16 ); \
453 case 9: hashv += ( (unsigned)_hj_key[8] << 8 ); \
454 case 8: _hj_j += ( (unsigned)_hj_key[7] << 24 ); \
455 case 7: _hj_j += ( (unsigned)_hj_key[6] << 16 ); \
456 case 6: _hj_j += ( (unsigned)_hj_key[5] << 8 ); \
457 case 5: _hj_j += _hj_key[4]; \
458 case 4: _hj_i += ( (unsigned)_hj_key[3] << 24 ); \
459 case 3: _hj_i += ( (unsigned)_hj_key[2] << 16 ); \
460 case 2: _hj_i += ( (unsigned)_hj_key[1] << 8 ); \
461 case 1: _hj_i += _hj_key[0]; \
463 HASH_JEN_MIX(_hj_i, _hj_j, hashv); \
464 bkt = hashv & (num_bkts-1); \
467 /* The Paul Hsieh hash function */
469 #if (defined(__GNUC__) && defined(__i386__)) || defined(__WATCOMC__) \
470 || defined(_MSC_VER) || defined (__BORLANDC__) || defined (__TURBOC__)
471 #define get16bits(d) (*((const uint16_t *) (d)))
474 #if !defined (get16bits)
475 #define get16bits(d) ((((uint32_t)(((const uint8_t *)(d))[1])) << 8) \
476 +(uint32_t)(((const uint8_t *)(d))[0]) )
478 #define HASH_SFH(key,keylen,num_bkts,hashv,bkt) \
480 unsigned char *_sfh_key=(unsigned char*)(key); \
481 uint32_t _sfh_tmp, _sfh_len = keylen; \
483 int _sfh_rem = _sfh_len & 3; \
485 hashv = 0xcafebabe; \
488 for (;_sfh_len > 0; _sfh_len--) { \
489 hashv += get16bits (_sfh_key); \
490 _sfh_tmp = (uint32_t)(get16bits (_sfh_key+2)) << 11 ^ hashv; \
491 hashv = (hashv << 16) ^ _sfh_tmp; \
492 _sfh_key += 2*sizeof (uint16_t); \
493 hashv += hashv >> 11; \
496 /* Handle end cases */ \
497 switch (_sfh_rem) { \
498 case 3: hashv += get16bits (_sfh_key); \
499 hashv ^= hashv << 16; \
500 hashv ^= (uint32_t)(_sfh_key[sizeof (uint16_t)] << 18); \
501 hashv += hashv >> 11; \
503 case 2: hashv += get16bits (_sfh_key); \
504 hashv ^= hashv << 11; \
505 hashv += hashv >> 17; \
507 case 1: hashv += *_sfh_key; \
508 hashv ^= hashv << 10; \
509 hashv += hashv >> 1; \
512 /* Force "avalanching" of final 127 bits */ \
513 hashv ^= hashv << 3; \
514 hashv += hashv >> 5; \
515 hashv ^= hashv << 4; \
516 hashv += hashv >> 17; \
517 hashv ^= hashv << 25; \
518 hashv += hashv >> 6; \
519 bkt = hashv & (num_bkts-1); \
522 #ifdef HASH_USING_NO_STRICT_ALIASING
523 /* The MurmurHash exploits some CPU's (x86,x86_64) tolerance for unaligned reads.
524 * For other types of CPU's (e.g. Sparc) an unaligned read causes a bus error.
525 * MurmurHash uses the faster approach only on CPU's where we know it's safe.
527 * Note the preprocessor built-in defines can be emitted using:
529 * gcc -m64 -dM -E - < /dev/null (on gcc)
530 * cc -## a.c (where a.c is a simple test file) (Sun Studio)
532 #if (defined(__i386__) || defined(__x86_64__) || defined(_M_IX86))
533 #define MUR_GETBLOCK(p,i) p[i]
534 #else /* non intel */
535 #define MUR_PLUS0_ALIGNED(p) (((unsigned long)p & 0x3) == 0)
536 #define MUR_PLUS1_ALIGNED(p) (((unsigned long)p & 0x3) == 1)
537 #define MUR_PLUS2_ALIGNED(p) (((unsigned long)p & 0x3) == 2)
538 #define MUR_PLUS3_ALIGNED(p) (((unsigned long)p & 0x3) == 3)
539 #define WP(p) ((uint32_t*)((unsigned long)(p) & ~3UL))
540 #if (defined(__BIG_ENDIAN__) || defined(SPARC) || defined(__ppc__) || defined(__ppc64__))
541 #define MUR_THREE_ONE(p) ((((*WP(p))&0x00ffffff) << 8) | (((*(WP(p)+1))&0xff000000) >> 24))
542 #define MUR_TWO_TWO(p) ((((*WP(p))&0x0000ffff) <<16) | (((*(WP(p)+1))&0xffff0000) >> 16))
543 #define MUR_ONE_THREE(p) ((((*WP(p))&0x000000ff) <<24) | (((*(WP(p)+1))&0xffffff00) >> 8))
544 #else /* assume little endian non-intel */
545 #define MUR_THREE_ONE(p) ((((*WP(p))&0xffffff00) >> 8) | (((*(WP(p)+1))&0x000000ff) << 24))
546 #define MUR_TWO_TWO(p) ((((*WP(p))&0xffff0000) >>16) | (((*(WP(p)+1))&0x0000ffff) << 16))
547 #define MUR_ONE_THREE(p) ((((*WP(p))&0xff000000) >>24) | (((*(WP(p)+1))&0x00ffffff) << 8))
549 #define MUR_GETBLOCK(p,i) (MUR_PLUS0_ALIGNED(p) ? ((p)[i]) : \
550 (MUR_PLUS1_ALIGNED(p) ? MUR_THREE_ONE(p) : \
551 (MUR_PLUS2_ALIGNED(p) ? MUR_TWO_TWO(p) : \
554 #define MUR_ROTL32(x,r) (((x) << (r)) | ((x) >> (32 - (r))))
555 #define MUR_FMIX(_h) \
564 #define HASH_MUR(key,keylen,num_bkts,hashv,bkt) \
566 const uint8_t *_mur_data = (const uint8_t*)(key); \
567 const int _mur_nblocks = (keylen) / 4; \
568 uint32_t _mur_h1 = 0xf88D5353; \
569 uint32_t _mur_c1 = 0xcc9e2d51; \
570 uint32_t _mur_c2 = 0x1b873593; \
571 uint32_t _mur_k1 = 0; \
572 const uint8_t *_mur_tail; \
573 const uint32_t *_mur_blocks = (const uint32_t*)(_mur_data+_mur_nblocks*4); \
575 for(_mur_i = -_mur_nblocks; _mur_i; _mur_i++) { \
576 _mur_k1 = MUR_GETBLOCK(_mur_blocks,_mur_i); \
577 _mur_k1 *= _mur_c1; \
578 _mur_k1 = MUR_ROTL32(_mur_k1,15); \
579 _mur_k1 *= _mur_c2; \
581 _mur_h1 ^= _mur_k1; \
582 _mur_h1 = MUR_ROTL32(_mur_h1,13); \
583 _mur_h1 = _mur_h1*5+0xe6546b64; \
585 _mur_tail = (const uint8_t*)(_mur_data + _mur_nblocks*4); \
587 switch((keylen) & 3) { \
588 case 3: _mur_k1 ^= _mur_tail[2] << 16; \
589 case 2: _mur_k1 ^= _mur_tail[1] << 8; \
590 case 1: _mur_k1 ^= _mur_tail[0]; \
591 _mur_k1 *= _mur_c1; \
592 _mur_k1 = MUR_ROTL32(_mur_k1,15); \
593 _mur_k1 *= _mur_c2; \
594 _mur_h1 ^= _mur_k1; \
596 _mur_h1 ^= (keylen); \
599 bkt = hashv & (num_bkts-1); \
601 #endif /* HASH_USING_NO_STRICT_ALIASING */
603 /* key comparison function; return 0 if keys equal */
604 #define HASH_KEYCMP(a,b,len) memcmp(a,b,len)
606 /* iterate over items in a known bucket to find desired item */
607 #define HASH_FIND_IN_BKT(tbl,hh,head,keyptr,keylen_in,out) \
609 if (head.hh_head) DECLTYPE_ASSIGN(out,ELMT_FROM_HH(tbl,head.hh_head)); \
612 if ((out)->hh.keylen == keylen_in) { \
613 if ((HASH_KEYCMP((out)->hh.key,keyptr,keylen_in)) == 0) break; \
615 if ((out)->hh.hh_next) DECLTYPE_ASSIGN(out,ELMT_FROM_HH(tbl,(out)->hh.hh_next)); \
620 /* add an item to a bucket */
621 #define HASH_ADD_TO_BKT(head,addhh) \
624 (addhh)->hh_next = head.hh_head; \
625 (addhh)->hh_prev = NULL; \
626 if (head.hh_head) { (head).hh_head->hh_prev = (addhh); } \
627 (head).hh_head=addhh; \
628 if (head.count >= ((head.expand_mult+1) * HASH_BKT_CAPACITY_THRESH) \
629 && (addhh)->tbl->noexpand != 1) { \
630 HASH_EXPAND_BUCKETS((addhh)->tbl); \
634 /* remove an item from a given bucket */
635 #define HASH_DEL_IN_BKT(hh,head,hh_del) \
637 if ((head).hh_head == hh_del) { \
638 (head).hh_head = hh_del->hh_next; \
640 if (hh_del->hh_prev) { \
641 hh_del->hh_prev->hh_next = hh_del->hh_next; \
643 if (hh_del->hh_next) { \
644 hh_del->hh_next->hh_prev = hh_del->hh_prev; \
647 /* Bucket expansion has the effect of doubling the number of buckets
648 * and redistributing the items into the new buckets. Ideally the
649 * items will distribute more or less evenly into the new buckets
650 * (the extent to which this is true is a measure of the quality of
651 * the hash function as it applies to the key domain).
653 * With the items distributed into more buckets, the chain length
654 * (item count) in each bucket is reduced. Thus by expanding buckets
655 * the hash keeps a bound on the chain length. This bounded chain
656 * length is the essence of how a hash provides constant time lookup.
658 * The calculation of tbl->ideal_chain_maxlen below deserves some
659 * explanation. First, keep in mind that we're calculating the ideal
660 * maximum chain length based on the *new* (doubled) bucket count.
661 * In fractions this is just n/b (n=number of items,b=new num buckets).
662 * Since the ideal chain length is an integer, we want to calculate
663 * ceil(n/b). We don't depend on floating point arithmetic in this
664 * hash, so to calculate ceil(n/b) with integers we could write
666 * ceil(n/b) = (n/b) + ((n%b)?1:0)
668 * and in fact a previous version of this hash did just that.
669 * But now we have improved things a bit by recognizing that b is
670 * always a power of two. We keep its base 2 log handy (call it lb),
671 * so now we can write this with a bit shift and logical AND:
673 * ceil(n/b) = (n>>lb) + ( (n & (b-1)) ? 1:0)
676 #define HASH_EXPAND_BUCKETS(tbl) \
679 unsigned _he_bkt_i; \
680 struct UT_hash_handle *_he_thh, *_he_hh_nxt; \
681 UT_hash_bucket *_he_new_buckets, *_he_newbkt; \
682 _he_new_buckets = (UT_hash_bucket*)uthash_malloc( \
683 2 * tbl->num_buckets * sizeof(struct UT_hash_bucket)); \
684 if (!_he_new_buckets) { uthash_fatal( "out of memory"); } \
685 memset(_he_new_buckets, 0, \
686 2 * tbl->num_buckets * sizeof(struct UT_hash_bucket)); \
687 tbl->ideal_chain_maxlen = \
688 (tbl->num_items >> (tbl->log2_num_buckets+1)) + \
689 ((tbl->num_items & ((tbl->num_buckets*2)-1)) ? 1 : 0); \
690 tbl->nonideal_items = 0; \
691 for(_he_bkt_i = 0; _he_bkt_i < tbl->num_buckets; _he_bkt_i++) \
693 _he_thh = tbl->buckets[ _he_bkt_i ].hh_head; \
695 _he_hh_nxt = _he_thh->hh_next; \
696 HASH_TO_BKT( _he_thh->hashv, tbl->num_buckets*2, _he_bkt); \
697 _he_newbkt = &(_he_new_buckets[ _he_bkt ]); \
698 if (++(_he_newbkt->count) > tbl->ideal_chain_maxlen) { \
699 tbl->nonideal_items++; \
700 _he_newbkt->expand_mult = _he_newbkt->count / \
701 tbl->ideal_chain_maxlen; \
703 _he_thh->hh_prev = NULL; \
704 _he_thh->hh_next = _he_newbkt->hh_head; \
705 if (_he_newbkt->hh_head) _he_newbkt->hh_head->hh_prev = \
707 _he_newbkt->hh_head = _he_thh; \
708 _he_thh = _he_hh_nxt; \
711 uthash_free( tbl->buckets, tbl->num_buckets*sizeof(struct UT_hash_bucket) ); \
712 tbl->num_buckets *= 2; \
713 tbl->log2_num_buckets++; \
714 tbl->buckets = _he_new_buckets; \
715 tbl->ineff_expands = (tbl->nonideal_items > (tbl->num_items >> 1)) ? \
716 (tbl->ineff_expands+1) : 0; \
717 if (tbl->ineff_expands > 1) { \
719 uthash_noexpand_fyi(tbl); \
721 uthash_expand_fyi(tbl); \
725 /* This is an adaptation of Simon Tatham's O(n log(n)) mergesort */
726 /* Note that HASH_SORT assumes the hash handle name to be hh.
727 * HASH_SRT was added to allow the hash handle name to be passed in. */
728 #define HASH_SORT(head,cmpfcn) HASH_SRT(hh,head,cmpfcn)
729 #define HASH_SRT(hh,head,cmpfcn) \
732 unsigned _hs_looping,_hs_nmerges,_hs_insize,_hs_psize,_hs_qsize; \
733 struct UT_hash_handle *_hs_p, *_hs_q, *_hs_e, *_hs_list, *_hs_tail; \
737 _hs_list = &((head)->hh); \
738 while (_hs_looping) { \
747 for ( _hs_i = 0; _hs_i < _hs_insize; _hs_i++ ) { \
749 _hs_q = (UT_hash_handle*)((_hs_q->next) ? \
750 ((void*)((char*)(_hs_q->next) + \
751 (head)->hh.tbl->hho)) : NULL); \
752 if (! (_hs_q) ) break; \
754 _hs_qsize = _hs_insize; \
755 while ((_hs_psize > 0) || ((_hs_qsize > 0) && _hs_q )) { \
756 if (_hs_psize == 0) { \
758 _hs_q = (UT_hash_handle*)((_hs_q->next) ? \
759 ((void*)((char*)(_hs_q->next) + \
760 (head)->hh.tbl->hho)) : NULL); \
762 } else if ( (_hs_qsize == 0) || !(_hs_q) ) { \
765 _hs_p = (UT_hash_handle*)((_hs_p->next) ? \
766 ((void*)((char*)(_hs_p->next) + \
767 (head)->hh.tbl->hho)) : NULL); \
771 cmpfcn(DECLTYPE(head)(ELMT_FROM_HH((head)->hh.tbl,_hs_p)), \
772 DECLTYPE(head)(ELMT_FROM_HH((head)->hh.tbl,_hs_q))) \
776 _hs_p = (UT_hash_handle*)((_hs_p->next) ? \
777 ((void*)((char*)(_hs_p->next) + \
778 (head)->hh.tbl->hho)) : NULL); \
783 _hs_q = (UT_hash_handle*)((_hs_q->next) ? \
784 ((void*)((char*)(_hs_q->next) + \
785 (head)->hh.tbl->hho)) : NULL); \
789 _hs_tail->next = ((_hs_e) ? \
790 ELMT_FROM_HH((head)->hh.tbl,_hs_e) : NULL); \
795 _hs_e->prev = ((_hs_tail) ? \
796 ELMT_FROM_HH((head)->hh.tbl,_hs_tail) : NULL); \
803 _hs_tail->next = NULL; \
805 if ( _hs_nmerges <= 1 ) { \
807 (head)->hh.tbl->tail = _hs_tail; \
808 DECLTYPE_ASSIGN(head,ELMT_FROM_HH((head)->hh.tbl, _hs_list)); \
812 HASH_FSCK(hh,head); \
816 /* This function selects items from one hash into another hash.
817 * The end result is that the selected items have dual presence
818 * in both hashes. There is no copy of the items made; rather
819 * they are added into the new hash through a secondary hash
820 * hash handle that must be present in the structure. */
821 #define HASH_SELECT(hh_dst, dst, hh_src, src, cond) \
823 unsigned _src_bkt, _dst_bkt; \
824 void *_last_elt=NULL, *_elt; \
825 UT_hash_handle *_src_hh, *_dst_hh, *_last_elt_hh=NULL; \
826 ptrdiff_t _dst_hho = ((char*)(&(dst)->hh_dst) - (char*)(dst)); \
828 for(_src_bkt=0; _src_bkt < (src)->hh_src.tbl->num_buckets; _src_bkt++) { \
829 for(_src_hh = (src)->hh_src.tbl->buckets[_src_bkt].hh_head; \
831 _src_hh = _src_hh->hh_next) { \
832 _elt = ELMT_FROM_HH((src)->hh_src.tbl, _src_hh); \
834 _dst_hh = (UT_hash_handle*)(((char*)_elt) + _dst_hho); \
835 _dst_hh->key = _src_hh->key; \
836 _dst_hh->keylen = _src_hh->keylen; \
837 _dst_hh->hashv = _src_hh->hashv; \
838 _dst_hh->prev = _last_elt; \
839 _dst_hh->next = NULL; \
840 if (_last_elt_hh) { _last_elt_hh->next = _elt; } \
842 DECLTYPE_ASSIGN(dst,_elt); \
843 HASH_MAKE_TABLE(hh_dst,dst); \
845 _dst_hh->tbl = (dst)->hh_dst.tbl; \
847 HASH_TO_BKT(_dst_hh->hashv, _dst_hh->tbl->num_buckets, _dst_bkt); \
848 HASH_ADD_TO_BKT(_dst_hh->tbl->buckets[_dst_bkt],_dst_hh); \
849 (dst)->hh_dst.tbl->num_items++; \
851 _last_elt_hh = _dst_hh; \
856 HASH_FSCK(hh_dst,dst); \
859 #define HASH_CLEAR(hh,head) \
862 uthash_free((head)->hh.tbl->buckets, \
863 (head)->hh.tbl->num_buckets*sizeof(struct UT_hash_bucket)); \
864 HASH_BLOOM_FREE((head)->hh.tbl); \
865 uthash_free((head)->hh.tbl, sizeof(UT_hash_table)); \
870 #define HASH_OVERHEAD(hh,head) \
871 (size_t)((((head)->hh.tbl->num_items * sizeof(UT_hash_handle)) + \
872 ((head)->hh.tbl->num_buckets * sizeof(UT_hash_bucket)) + \
873 (sizeof(UT_hash_table)) + \
874 (HASH_BLOOM_BYTELEN)))
877 #define HASH_ITER(hh,head,el,tmp) \
878 for((el)=(head), (*(char**)(&(tmp)))=(char*)((head)?(head)->hh.next:NULL); \
879 el; (el)=(tmp),(*(char**)(&(tmp)))=(char*)((tmp)?(tmp)->hh.next:NULL))
881 #define HASH_ITER(hh,head,el,tmp) \
882 for((el)=(head),(tmp)=DECLTYPE(el)((head)?(head)->hh.next:NULL); \
883 el; (el)=(tmp),(tmp)=DECLTYPE(el)((tmp)?(tmp)->hh.next:NULL))
886 /* obtain a count of items in the hash */
887 #define HASH_COUNT(head) HASH_CNT(hh,head)
888 #define HASH_CNT(hh,head) ((head)?((head)->hh.tbl->num_items):0)
890 typedef struct UT_hash_bucket
{
891 struct UT_hash_handle
*hh_head
;
894 /* expand_mult is normally set to 0. In this situation, the max chain length
895 * threshold is enforced at its default value, HASH_BKT_CAPACITY_THRESH. (If
896 * the bucket's chain exceeds this length, bucket expansion is triggered).
897 * However, setting expand_mult to a non-zero value delays bucket expansion
898 * (that would be triggered by additions to this particular bucket)
899 * until its chain length reaches a *multiple* of HASH_BKT_CAPACITY_THRESH.
900 * (The multiplier is simply expand_mult+1). The whole idea of this
901 * multiplier is to reduce bucket expansions, since they are expensive, in
902 * situations where we know that a particular bucket tends to be overused.
903 * It is better to let its chain length grow to a longer yet-still-bounded
904 * value, than to do an O(n) bucket expansion too often.
906 unsigned expand_mult
;
910 /* random signature used only to find hash tables in external analysis */
911 #define HASH_SIGNATURE 0xa0111fe1
912 #define HASH_BLOOM_SIGNATURE 0xb12220f2
914 typedef struct UT_hash_table
{
915 UT_hash_bucket
*buckets
;
916 unsigned num_buckets
, log2_num_buckets
;
918 struct UT_hash_handle
*tail
; /* tail hh in app order, for fast append */
919 ptrdiff_t hho
; /* hash handle offset (byte pos of hash handle in element */
921 /* in an ideal situation (all buckets used equally), no bucket would have
922 * more than ceil(#items/#buckets) items. that's the ideal chain length. */
923 unsigned ideal_chain_maxlen
;
925 /* nonideal_items is the number of items in the hash whose chain position
926 * exceeds the ideal chain maxlen. these items pay the penalty for an uneven
927 * hash distribution; reaching them in a chain traversal takes >ideal steps */
928 unsigned nonideal_items
;
930 /* ineffective expands occur when a bucket doubling was performed, but
931 * afterward, more than half the items in the hash had nonideal chain
932 * positions. If this happens on two consecutive expansions we inhibit any
933 * further expansion, as it's not helping; this happens when the hash
934 * function isn't a good fit for the key domain. When expansion is inhibited
935 * the hash will still work, albeit no longer in constant time. */
936 unsigned ineff_expands
, noexpand
;
938 uint32_t signature
; /* used only to find hash tables in external analysis */
940 uint32_t bloom_sig
; /* used only to test bloom exists in external analysis */
947 typedef struct UT_hash_handle
{
948 struct UT_hash_table
*tbl
;
949 void *prev
; /* prev element in app order */
950 void *next
; /* next element in app order */
951 struct UT_hash_handle
*hh_prev
; /* previous hh in bucket order */
952 struct UT_hash_handle
*hh_next
; /* next hh in bucket order */
953 void *key
; /* ptr to enclosing struct's key */
954 unsigned keylen
; /* enclosing struct's key len */
955 unsigned hashv
; /* result of hash-fcn(key) */
958 #endif /* UTHASH_H */