2 Copyright (c) 2003-2013, 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 #ifdef _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 #else /* GNU, Sun and other compilers */
43 #if defined(__HP_aCC) || (defined(__SUNPRO_CC) && (__SUNPRO_CC < 0x590))
47 #define DECLTYPE(x) (__typeof(x))
52 #define DECLTYPE_ASSIGN(dst,src) \
54 char **_da_dst = (char**)(&(dst)); \
55 *_da_dst = (char*)(src); \
58 #define DECLTYPE_ASSIGN(dst,src) \
60 (dst) = DECLTYPE(dst)(src); \
64 /* a number of the hash function use uint32_t which isn't defined on win32 */
66 typedef unsigned int uint32_t;
67 typedef unsigned char uint8_t;
69 #include <inttypes.h> /* uint32_t */
72 #define UTHASH_VERSION 1.9.8
75 #define uthash_fatal(msg) exit(-1) /* fatal error (out of memory,etc) */
78 #define uthash_malloc(sz) malloc(sz) /* malloc fcn */
81 #define uthash_free(ptr,sz) free(ptr) /* free fcn */
84 #ifndef uthash_noexpand_fyi
85 #define uthash_noexpand_fyi(tbl) /* can be defined to log noexpand */
87 #ifndef uthash_expand_fyi
88 #define uthash_expand_fyi(tbl) /* can be defined to log expands */
91 /* initial number of buckets */
92 #define HASH_INITIAL_NUM_BUCKETS 32 /* initial number of buckets */
93 #define HASH_INITIAL_NUM_BUCKETS_LOG2 5 /* lg2 of initial number of buckets */
94 #define HASH_BKT_CAPACITY_THRESH 10 /* expand when bucket count reaches */
96 /* calculate the element whose hash handle address is hhe */
97 #define ELMT_FROM_HH(tbl,hhp) ((void*)(((char*)(hhp)) - ((tbl)->hho)))
99 #define HASH_FIND(hh,head,keyptr,keylen,out) \
101 unsigned _hf_bkt,_hf_hashv; \
104 HASH_FCN(keyptr,keylen, (head)->hh.tbl->num_buckets, _hf_hashv, _hf_bkt); \
105 if (HASH_BLOOM_TEST((head)->hh.tbl, _hf_hashv)) { \
106 HASH_FIND_IN_BKT((head)->hh.tbl, hh, (head)->hh.tbl->buckets[ _hf_bkt ], \
107 keyptr,keylen,out); \
113 #define HASH_BLOOM_BITLEN (1ULL << HASH_BLOOM)
114 #define HASH_BLOOM_BYTELEN (HASH_BLOOM_BITLEN/8) + ((HASH_BLOOM_BITLEN%8) ? 1:0)
115 #define HASH_BLOOM_MAKE(tbl) \
117 (tbl)->bloom_nbits = HASH_BLOOM; \
118 (tbl)->bloom_bv = (uint8_t*)uthash_malloc(HASH_BLOOM_BYTELEN); \
119 if (!((tbl)->bloom_bv)) { uthash_fatal( "out of memory"); } \
120 memset((tbl)->bloom_bv, 0, HASH_BLOOM_BYTELEN); \
121 (tbl)->bloom_sig = HASH_BLOOM_SIGNATURE; \
124 #define HASH_BLOOM_FREE(tbl) \
126 uthash_free((tbl)->bloom_bv, HASH_BLOOM_BYTELEN); \
129 #define HASH_BLOOM_BITSET(bv,idx) (bv[(idx)/8] |= (1U << ((idx)%8)))
130 #define HASH_BLOOM_BITTEST(bv,idx) (bv[(idx)/8] & (1U << ((idx)%8)))
132 #define HASH_BLOOM_ADD(tbl,hashv) \
133 HASH_BLOOM_BITSET((tbl)->bloom_bv, (hashv & (uint32_t)((1ULL << (tbl)->bloom_nbits) - 1)))
135 #define HASH_BLOOM_TEST(tbl,hashv) \
136 HASH_BLOOM_BITTEST((tbl)->bloom_bv, (hashv & (uint32_t)((1ULL << (tbl)->bloom_nbits) - 1)))
139 #define HASH_BLOOM_MAKE(tbl)
140 #define HASH_BLOOM_FREE(tbl)
141 #define HASH_BLOOM_ADD(tbl,hashv)
142 #define HASH_BLOOM_TEST(tbl,hashv) (1)
143 #define HASH_BLOOM_BYTELEN 0
146 #define HASH_MAKE_TABLE(hh,head) \
148 (head)->hh.tbl = (UT_hash_table*)uthash_malloc( \
149 sizeof(UT_hash_table)); \
150 if (!((head)->hh.tbl)) { uthash_fatal( "out of memory"); } \
151 memset((head)->hh.tbl, 0, sizeof(UT_hash_table)); \
152 (head)->hh.tbl->tail = &((head)->hh); \
153 (head)->hh.tbl->num_buckets = HASH_INITIAL_NUM_BUCKETS; \
154 (head)->hh.tbl->log2_num_buckets = HASH_INITIAL_NUM_BUCKETS_LOG2; \
155 (head)->hh.tbl->hho = (char*)(&(head)->hh) - (char*)(head); \
156 (head)->hh.tbl->buckets = (UT_hash_bucket*)uthash_malloc( \
157 HASH_INITIAL_NUM_BUCKETS*sizeof(struct UT_hash_bucket)); \
158 if (! (head)->hh.tbl->buckets) { uthash_fatal( "out of memory"); } \
159 memset((head)->hh.tbl->buckets, 0, \
160 HASH_INITIAL_NUM_BUCKETS*sizeof(struct UT_hash_bucket)); \
161 HASH_BLOOM_MAKE((head)->hh.tbl); \
162 (head)->hh.tbl->signature = HASH_SIGNATURE; \
165 #define HASH_ADD(hh,head,fieldname,keylen_in,add) \
166 HASH_ADD_KEYPTR(hh,head,&((add)->fieldname),keylen_in,add)
168 #define HASH_REPLACE(hh,head,fieldname,keylen_in,add,replaced) \
171 HASH_FIND(hh,head,&((add)->fieldname),keylen_in,replaced); \
172 if (replaced!=NULL) { \
173 HASH_DELETE(hh,head,replaced); \
175 HASH_ADD(hh,head,fieldname,keylen_in,add); \
178 #define HASH_ADD_KEYPTR(hh,head,keyptr,keylen_in,add) \
181 (add)->hh.next = NULL; \
182 (add)->hh.key = (char*)keyptr; \
183 (add)->hh.keylen = (unsigned)keylen_in; \
186 (head)->hh.prev = NULL; \
187 HASH_MAKE_TABLE(hh,head); \
189 (head)->hh.tbl->tail->next = (add); \
190 (add)->hh.prev = ELMT_FROM_HH((head)->hh.tbl, (head)->hh.tbl->tail); \
191 (head)->hh.tbl->tail = &((add)->hh); \
193 (head)->hh.tbl->num_items++; \
194 (add)->hh.tbl = (head)->hh.tbl; \
195 HASH_FCN(keyptr,keylen_in, (head)->hh.tbl->num_buckets, \
196 (add)->hh.hashv, _ha_bkt); \
197 HASH_ADD_TO_BKT((head)->hh.tbl->buckets[_ha_bkt],&(add)->hh); \
198 HASH_BLOOM_ADD((head)->hh.tbl,(add)->hh.hashv); \
199 HASH_EMIT_KEY(hh,head,keyptr,keylen_in); \
200 HASH_FSCK(hh,head); \
203 #define HASH_TO_BKT( hashv, num_bkts, bkt ) \
205 bkt = ((hashv) & ((num_bkts) - 1)); \
208 /* delete "delptr" from the hash table.
209 * "the usual" patch-up process for the app-order doubly-linked-list.
210 * The use of _hd_hh_del below deserves special explanation.
211 * These used to be expressed using (delptr) but that led to a bug
212 * if someone used the same symbol for the head and deletee, like
213 * HASH_DELETE(hh,users,users);
214 * We want that to work, but by changing the head (users) below
215 * we were forfeiting our ability to further refer to the deletee (users)
216 * in the patch-up process. Solution: use scratch space to
217 * copy the deletee pointer, then the latter references are via that
218 * scratch pointer rather than through the repointed (users) symbol.
220 #define HASH_DELETE(hh,head,delptr) \
223 struct UT_hash_handle *_hd_hh_del; \
224 if ( ((delptr)->hh.prev == NULL) && ((delptr)->hh.next == NULL) ) { \
225 uthash_free((head)->hh.tbl->buckets, \
226 (head)->hh.tbl->num_buckets*sizeof(struct UT_hash_bucket) ); \
227 HASH_BLOOM_FREE((head)->hh.tbl); \
228 uthash_free((head)->hh.tbl, sizeof(UT_hash_table)); \
231 _hd_hh_del = &((delptr)->hh); \
232 if ((delptr) == ELMT_FROM_HH((head)->hh.tbl,(head)->hh.tbl->tail)) { \
233 (head)->hh.tbl->tail = \
234 (UT_hash_handle*)((ptrdiff_t)((delptr)->hh.prev) + \
235 (head)->hh.tbl->hho); \
237 if ((delptr)->hh.prev) { \
238 ((UT_hash_handle*)((ptrdiff_t)((delptr)->hh.prev) + \
239 (head)->hh.tbl->hho))->next = (delptr)->hh.next; \
241 DECLTYPE_ASSIGN(head,(delptr)->hh.next); \
243 if (_hd_hh_del->next) { \
244 ((UT_hash_handle*)((ptrdiff_t)_hd_hh_del->next + \
245 (head)->hh.tbl->hho))->prev = \
248 HASH_TO_BKT( _hd_hh_del->hashv, (head)->hh.tbl->num_buckets, _hd_bkt); \
249 HASH_DEL_IN_BKT(hh,(head)->hh.tbl->buckets[_hd_bkt], _hd_hh_del); \
250 (head)->hh.tbl->num_items--; \
252 HASH_FSCK(hh,head); \
256 /* convenience forms of HASH_FIND/HASH_ADD/HASH_DEL */
257 #define HASH_FIND_STR(head,findstr,out) \
258 HASH_FIND(hh,head,findstr,strlen(findstr),out)
259 #define HASH_ADD_STR(head,strfield,add) \
260 HASH_ADD(hh,head,strfield,strlen(add->strfield),add)
261 #define HASH_REPLACE_STR(head,strfield,add,replaced) \
262 HASH_REPLACE(hh,head,strfield,strlen(add->strfield),add,replaced)
263 #define HASH_FIND_INT(head,findint,out) \
264 HASH_FIND(hh,head,findint,sizeof(int),out)
265 #define HASH_ADD_INT(head,intfield,add) \
266 HASH_ADD(hh,head,intfield,sizeof(int),add)
267 #define HASH_REPLACE_INT(head,intfield,add,replaced) \
268 HASH_REPLACE(hh,head,intfield,sizeof(int),add,replaced)
269 #define HASH_FIND_PTR(head,findptr,out) \
270 HASH_FIND(hh,head,findptr,sizeof(void *),out)
271 #define HASH_ADD_PTR(head,ptrfield,add) \
272 HASH_ADD(hh,head,ptrfield,sizeof(void *),add)
273 #define HASH_REPLACE_PTR(head,ptrfield,add) \
274 HASH_REPLACE(hh,head,ptrfield,sizeof(void *),add,replaced)
275 #define HASH_DEL(head,delptr) \
276 HASH_DELETE(hh,head,delptr)
278 /* HASH_FSCK checks hash integrity on every add/delete when HASH_DEBUG is defined.
279 * This is for uthash developer only; it compiles away if HASH_DEBUG isn't defined.
282 #define HASH_OOPS(...) do { fprintf(stderr,__VA_ARGS__); exit(-1); } while (0)
283 #define HASH_FSCK(hh,head) \
286 unsigned _count, _bkt_count; \
288 struct UT_hash_handle *_thh; \
291 for( _bkt_i = 0; _bkt_i < (head)->hh.tbl->num_buckets; _bkt_i++) { \
293 _thh = (head)->hh.tbl->buckets[_bkt_i].hh_head; \
296 if (_prev != (char*)(_thh->hh_prev)) { \
297 HASH_OOPS("invalid hh_prev %p, actual %p\n", \
298 _thh->hh_prev, _prev ); \
301 _prev = (char*)(_thh); \
302 _thh = _thh->hh_next; \
304 _count += _bkt_count; \
305 if ((head)->hh.tbl->buckets[_bkt_i].count != _bkt_count) { \
306 HASH_OOPS("invalid bucket count %d, actual %d\n", \
307 (head)->hh.tbl->buckets[_bkt_i].count, _bkt_count); \
310 if (_count != (head)->hh.tbl->num_items) { \
311 HASH_OOPS("invalid hh item count %d, actual %d\n", \
312 (head)->hh.tbl->num_items, _count ); \
314 /* traverse hh in app order; check next/prev integrity, count */ \
317 _thh = &(head)->hh; \
320 if (_prev !=(char*)(_thh->prev)) { \
321 HASH_OOPS("invalid prev %p, actual %p\n", \
322 _thh->prev, _prev ); \
324 _prev = (char*)ELMT_FROM_HH((head)->hh.tbl, _thh); \
325 _thh = ( _thh->next ? (UT_hash_handle*)((char*)(_thh->next) + \
326 (head)->hh.tbl->hho) : NULL ); \
328 if (_count != (head)->hh.tbl->num_items) { \
329 HASH_OOPS("invalid app item count %d, actual %d\n", \
330 (head)->hh.tbl->num_items, _count ); \
335 #define HASH_FSCK(hh,head)
338 /* When compiled with -DHASH_EMIT_KEYS, length-prefixed keys are emitted to
339 * the descriptor to which this macro is defined for tuning the hash function.
340 * The app can #include <unistd.h> to get the prototype for write(2). */
341 #ifdef HASH_EMIT_KEYS
342 #define HASH_EMIT_KEY(hh,head,keyptr,fieldlen) \
344 unsigned _klen = fieldlen; \
345 write(HASH_EMIT_KEYS, &_klen, sizeof(_klen)); \
346 write(HASH_EMIT_KEYS, keyptr, fieldlen); \
349 #define HASH_EMIT_KEY(hh,head,keyptr,fieldlen)
352 /* default to Jenkin's hash unless overridden e.g. DHASH_FUNCTION=HASH_SAX */
354 #define HASH_FCN HASH_FUNCTION
356 #define HASH_FCN HASH_JEN
359 /* The Bernstein hash function, used in Perl prior to v5.6 */
360 #define HASH_BER(key,keylen,num_bkts,hashv,bkt) \
362 unsigned _hb_keylen=keylen; \
363 char *_hb_key=(char*)(key); \
365 while (_hb_keylen--) { (hashv) = ((hashv) * 33) + *_hb_key++; } \
366 bkt = (hashv) & (num_bkts-1); \
370 /* SAX/FNV/OAT/JEN hash functions are macro variants of those listed at
371 * http://eternallyconfuzzled.com/tuts/algorithms/jsw_tut_hashing.aspx */
372 #define HASH_SAX(key,keylen,num_bkts,hashv,bkt) \
375 char *_hs_key=(char*)(key); \
377 for(_sx_i=0; _sx_i < keylen; _sx_i++) \
378 hashv ^= (hashv << 5) + (hashv >> 2) + _hs_key[_sx_i]; \
379 bkt = hashv & (num_bkts-1); \
382 #define HASH_FNV(key,keylen,num_bkts,hashv,bkt) \
385 char *_hf_key=(char*)(key); \
386 hashv = 2166136261UL; \
387 for(_fn_i=0; _fn_i < keylen; _fn_i++) \
388 hashv = (hashv * 16777619) ^ _hf_key[_fn_i]; \
389 bkt = hashv & (num_bkts-1); \
392 #define HASH_OAT(key,keylen,num_bkts,hashv,bkt) \
395 char *_ho_key=(char*)(key); \
397 for(_ho_i=0; _ho_i < keylen; _ho_i++) { \
398 hashv += _ho_key[_ho_i]; \
399 hashv += (hashv << 10); \
400 hashv ^= (hashv >> 6); \
402 hashv += (hashv << 3); \
403 hashv ^= (hashv >> 11); \
404 hashv += (hashv << 15); \
405 bkt = hashv & (num_bkts-1); \
408 #define HASH_JEN_MIX(a,b,c) \
410 a -= b; a -= c; a ^= ( c >> 13 ); \
411 b -= c; b -= a; b ^= ( a << 8 ); \
412 c -= a; c -= b; c ^= ( b >> 13 ); \
413 a -= b; a -= c; a ^= ( c >> 12 ); \
414 b -= c; b -= a; b ^= ( a << 16 ); \
415 c -= a; c -= b; c ^= ( b >> 5 ); \
416 a -= b; a -= c; a ^= ( c >> 3 ); \
417 b -= c; b -= a; b ^= ( a << 10 ); \
418 c -= a; c -= b; c ^= ( b >> 15 ); \
421 #define HASH_JEN(key,keylen,num_bkts,hashv,bkt) \
423 unsigned _hj_i,_hj_j,_hj_k; \
424 unsigned char *_hj_key=(unsigned char*)(key); \
425 hashv = 0xfeedbeef; \
426 _hj_i = _hj_j = 0x9e3779b9; \
427 _hj_k = (unsigned)keylen; \
428 while (_hj_k >= 12) { \
429 _hj_i += (_hj_key[0] + ( (unsigned)_hj_key[1] << 8 ) \
430 + ( (unsigned)_hj_key[2] << 16 ) \
431 + ( (unsigned)_hj_key[3] << 24 ) ); \
432 _hj_j += (_hj_key[4] + ( (unsigned)_hj_key[5] << 8 ) \
433 + ( (unsigned)_hj_key[6] << 16 ) \
434 + ( (unsigned)_hj_key[7] << 24 ) ); \
435 hashv += (_hj_key[8] + ( (unsigned)_hj_key[9] << 8 ) \
436 + ( (unsigned)_hj_key[10] << 16 ) \
437 + ( (unsigned)_hj_key[11] << 24 ) ); \
439 HASH_JEN_MIX(_hj_i, _hj_j, hashv); \
446 case 11: hashv += ( (unsigned)_hj_key[10] << 24 ); \
447 case 10: hashv += ( (unsigned)_hj_key[9] << 16 ); \
448 case 9: hashv += ( (unsigned)_hj_key[8] << 8 ); \
449 case 8: _hj_j += ( (unsigned)_hj_key[7] << 24 ); \
450 case 7: _hj_j += ( (unsigned)_hj_key[6] << 16 ); \
451 case 6: _hj_j += ( (unsigned)_hj_key[5] << 8 ); \
452 case 5: _hj_j += _hj_key[4]; \
453 case 4: _hj_i += ( (unsigned)_hj_key[3] << 24 ); \
454 case 3: _hj_i += ( (unsigned)_hj_key[2] << 16 ); \
455 case 2: _hj_i += ( (unsigned)_hj_key[1] << 8 ); \
456 case 1: _hj_i += _hj_key[0]; \
458 HASH_JEN_MIX(_hj_i, _hj_j, hashv); \
459 bkt = hashv & (num_bkts-1); \
462 /* The Paul Hsieh hash function */
464 #if (defined(__GNUC__) && defined(__i386__)) || defined(__WATCOMC__) \
465 || defined(_MSC_VER) || defined (__BORLANDC__) || defined (__TURBOC__)
466 #define get16bits(d) (*((const uint16_t *) (d)))
469 #if !defined (get16bits)
470 #define get16bits(d) ((((uint32_t)(((const uint8_t *)(d))[1])) << 8) \
471 +(uint32_t)(((const uint8_t *)(d))[0]) )
473 #define HASH_SFH(key,keylen,num_bkts,hashv,bkt) \
475 unsigned char *_sfh_key=(unsigned char*)(key); \
476 uint32_t _sfh_tmp, _sfh_len = keylen; \
478 int _sfh_rem = _sfh_len & 3; \
480 hashv = 0xcafebabe; \
483 for (;_sfh_len > 0; _sfh_len--) { \
484 hashv += get16bits (_sfh_key); \
485 _sfh_tmp = (uint32_t)(get16bits (_sfh_key+2)) << 11 ^ hashv; \
486 hashv = (hashv << 16) ^ _sfh_tmp; \
487 _sfh_key += 2*sizeof (uint16_t); \
488 hashv += hashv >> 11; \
491 /* Handle end cases */ \
492 switch (_sfh_rem) { \
493 case 3: hashv += get16bits (_sfh_key); \
494 hashv ^= hashv << 16; \
495 hashv ^= (uint32_t)(_sfh_key[sizeof (uint16_t)] << 18); \
496 hashv += hashv >> 11; \
498 case 2: hashv += get16bits (_sfh_key); \
499 hashv ^= hashv << 11; \
500 hashv += hashv >> 17; \
502 case 1: hashv += *_sfh_key; \
503 hashv ^= hashv << 10; \
504 hashv += hashv >> 1; \
507 /* Force "avalanching" of final 127 bits */ \
508 hashv ^= hashv << 3; \
509 hashv += hashv >> 5; \
510 hashv ^= hashv << 4; \
511 hashv += hashv >> 17; \
512 hashv ^= hashv << 25; \
513 hashv += hashv >> 6; \
514 bkt = hashv & (num_bkts-1); \
517 #ifdef HASH_USING_NO_STRICT_ALIASING
518 /* The MurmurHash exploits some CPU's (x86,x86_64) tolerance for unaligned reads.
519 * For other types of CPU's (e.g. Sparc) an unaligned read causes a bus error.
520 * MurmurHash uses the faster approach only on CPU's where we know it's safe.
522 * Note the preprocessor built-in defines can be emitted using:
524 * gcc -m64 -dM -E - < /dev/null (on gcc)
525 * cc -## a.c (where a.c is a simple test file) (Sun Studio)
527 #if (defined(__i386__) || defined(__x86_64__) || defined(_M_IX86))
528 #define MUR_GETBLOCK(p,i) p[i]
529 #else /* non intel */
530 #define MUR_PLUS0_ALIGNED(p) (((unsigned long)p & 0x3) == 0)
531 #define MUR_PLUS1_ALIGNED(p) (((unsigned long)p & 0x3) == 1)
532 #define MUR_PLUS2_ALIGNED(p) (((unsigned long)p & 0x3) == 2)
533 #define MUR_PLUS3_ALIGNED(p) (((unsigned long)p & 0x3) == 3)
534 #define WP(p) ((uint32_t*)((unsigned long)(p) & ~3UL))
535 #if (defined(__BIG_ENDIAN__) || defined(SPARC) || defined(__ppc__) || defined(__ppc64__))
536 #define MUR_THREE_ONE(p) ((((*WP(p))&0x00ffffff) << 8) | (((*(WP(p)+1))&0xff000000) >> 24))
537 #define MUR_TWO_TWO(p) ((((*WP(p))&0x0000ffff) <<16) | (((*(WP(p)+1))&0xffff0000) >> 16))
538 #define MUR_ONE_THREE(p) ((((*WP(p))&0x000000ff) <<24) | (((*(WP(p)+1))&0xffffff00) >> 8))
539 #else /* assume little endian non-intel */
540 #define MUR_THREE_ONE(p) ((((*WP(p))&0xffffff00) >> 8) | (((*(WP(p)+1))&0x000000ff) << 24))
541 #define MUR_TWO_TWO(p) ((((*WP(p))&0xffff0000) >>16) | (((*(WP(p)+1))&0x0000ffff) << 16))
542 #define MUR_ONE_THREE(p) ((((*WP(p))&0xff000000) >>24) | (((*(WP(p)+1))&0x00ffffff) << 8))
544 #define MUR_GETBLOCK(p,i) (MUR_PLUS0_ALIGNED(p) ? ((p)[i]) : \
545 (MUR_PLUS1_ALIGNED(p) ? MUR_THREE_ONE(p) : \
546 (MUR_PLUS2_ALIGNED(p) ? MUR_TWO_TWO(p) : \
549 #define MUR_ROTL32(x,r) (((x) << (r)) | ((x) >> (32 - (r))))
550 #define MUR_FMIX(_h) \
559 #define HASH_MUR(key,keylen,num_bkts,hashv,bkt) \
561 const uint8_t *_mur_data = (const uint8_t*)(key); \
562 const int _mur_nblocks = (keylen) / 4; \
563 uint32_t _mur_h1 = 0xf88D5353; \
564 uint32_t _mur_c1 = 0xcc9e2d51; \
565 uint32_t _mur_c2 = 0x1b873593; \
566 uint32_t _mur_k1 = 0; \
567 const uint8_t *_mur_tail; \
568 const uint32_t *_mur_blocks = (const uint32_t*)(_mur_data+_mur_nblocks*4); \
570 for(_mur_i = -_mur_nblocks; _mur_i; _mur_i++) { \
571 _mur_k1 = MUR_GETBLOCK(_mur_blocks,_mur_i); \
572 _mur_k1 *= _mur_c1; \
573 _mur_k1 = MUR_ROTL32(_mur_k1,15); \
574 _mur_k1 *= _mur_c2; \
576 _mur_h1 ^= _mur_k1; \
577 _mur_h1 = MUR_ROTL32(_mur_h1,13); \
578 _mur_h1 = _mur_h1*5+0xe6546b64; \
580 _mur_tail = (const uint8_t*)(_mur_data + _mur_nblocks*4); \
582 switch((keylen) & 3) { \
583 case 3: _mur_k1 ^= _mur_tail[2] << 16; \
584 case 2: _mur_k1 ^= _mur_tail[1] << 8; \
585 case 1: _mur_k1 ^= _mur_tail[0]; \
586 _mur_k1 *= _mur_c1; \
587 _mur_k1 = MUR_ROTL32(_mur_k1,15); \
588 _mur_k1 *= _mur_c2; \
589 _mur_h1 ^= _mur_k1; \
591 _mur_h1 ^= (keylen); \
594 bkt = hashv & (num_bkts-1); \
596 #endif /* HASH_USING_NO_STRICT_ALIASING */
598 /* key comparison function; return 0 if keys equal */
599 #define HASH_KEYCMP(a,b,len) memcmp(a,b,len)
601 /* iterate over items in a known bucket to find desired item */
602 #define HASH_FIND_IN_BKT(tbl,hh,head,keyptr,keylen_in,out) \
604 if (head.hh_head) DECLTYPE_ASSIGN(out,ELMT_FROM_HH(tbl,head.hh_head)); \
607 if ((out)->hh.keylen == keylen_in) { \
608 if ((HASH_KEYCMP((out)->hh.key,keyptr,keylen_in)) == 0) break; \
610 if ((out)->hh.hh_next) DECLTYPE_ASSIGN(out,ELMT_FROM_HH(tbl,(out)->hh.hh_next)); \
615 /* add an item to a bucket */
616 #define HASH_ADD_TO_BKT(head,addhh) \
619 (addhh)->hh_next = head.hh_head; \
620 (addhh)->hh_prev = NULL; \
621 if (head.hh_head) { (head).hh_head->hh_prev = (addhh); } \
622 (head).hh_head=addhh; \
623 if (head.count >= ((head.expand_mult+1) * HASH_BKT_CAPACITY_THRESH) \
624 && (addhh)->tbl->noexpand != 1) { \
625 HASH_EXPAND_BUCKETS((addhh)->tbl); \
629 /* remove an item from a given bucket */
630 #define HASH_DEL_IN_BKT(hh,head,hh_del) \
632 if ((head).hh_head == hh_del) { \
633 (head).hh_head = hh_del->hh_next; \
635 if (hh_del->hh_prev) { \
636 hh_del->hh_prev->hh_next = hh_del->hh_next; \
638 if (hh_del->hh_next) { \
639 hh_del->hh_next->hh_prev = hh_del->hh_prev; \
642 /* Bucket expansion has the effect of doubling the number of buckets
643 * and redistributing the items into the new buckets. Ideally the
644 * items will distribute more or less evenly into the new buckets
645 * (the extent to which this is true is a measure of the quality of
646 * the hash function as it applies to the key domain).
648 * With the items distributed into more buckets, the chain length
649 * (item count) in each bucket is reduced. Thus by expanding buckets
650 * the hash keeps a bound on the chain length. This bounded chain
651 * length is the essence of how a hash provides constant time lookup.
653 * The calculation of tbl->ideal_chain_maxlen below deserves some
654 * explanation. First, keep in mind that we're calculating the ideal
655 * maximum chain length based on the *new* (doubled) bucket count.
656 * In fractions this is just n/b (n=number of items,b=new num buckets).
657 * Since the ideal chain length is an integer, we want to calculate
658 * ceil(n/b). We don't depend on floating point arithmetic in this
659 * hash, so to calculate ceil(n/b) with integers we could write
661 * ceil(n/b) = (n/b) + ((n%b)?1:0)
663 * and in fact a previous version of this hash did just that.
664 * But now we have improved things a bit by recognizing that b is
665 * always a power of two. We keep its base 2 log handy (call it lb),
666 * so now we can write this with a bit shift and logical AND:
668 * ceil(n/b) = (n>>lb) + ( (n & (b-1)) ? 1:0)
671 #define HASH_EXPAND_BUCKETS(tbl) \
674 unsigned _he_bkt_i; \
675 struct UT_hash_handle *_he_thh, *_he_hh_nxt; \
676 UT_hash_bucket *_he_new_buckets, *_he_newbkt; \
677 _he_new_buckets = (UT_hash_bucket*)uthash_malloc( \
678 2 * tbl->num_buckets * sizeof(struct UT_hash_bucket)); \
679 if (!_he_new_buckets) { uthash_fatal( "out of memory"); } \
680 memset(_he_new_buckets, 0, \
681 2 * tbl->num_buckets * sizeof(struct UT_hash_bucket)); \
682 tbl->ideal_chain_maxlen = \
683 (tbl->num_items >> (tbl->log2_num_buckets+1)) + \
684 ((tbl->num_items & ((tbl->num_buckets*2)-1)) ? 1 : 0); \
685 tbl->nonideal_items = 0; \
686 for(_he_bkt_i = 0; _he_bkt_i < tbl->num_buckets; _he_bkt_i++) \
688 _he_thh = tbl->buckets[ _he_bkt_i ].hh_head; \
690 _he_hh_nxt = _he_thh->hh_next; \
691 HASH_TO_BKT( _he_thh->hashv, tbl->num_buckets*2, _he_bkt); \
692 _he_newbkt = &(_he_new_buckets[ _he_bkt ]); \
693 if (++(_he_newbkt->count) > tbl->ideal_chain_maxlen) { \
694 tbl->nonideal_items++; \
695 _he_newbkt->expand_mult = _he_newbkt->count / \
696 tbl->ideal_chain_maxlen; \
698 _he_thh->hh_prev = NULL; \
699 _he_thh->hh_next = _he_newbkt->hh_head; \
700 if (_he_newbkt->hh_head) _he_newbkt->hh_head->hh_prev = \
702 _he_newbkt->hh_head = _he_thh; \
703 _he_thh = _he_hh_nxt; \
706 uthash_free( tbl->buckets, tbl->num_buckets*sizeof(struct UT_hash_bucket) ); \
707 tbl->num_buckets *= 2; \
708 tbl->log2_num_buckets++; \
709 tbl->buckets = _he_new_buckets; \
710 tbl->ineff_expands = (tbl->nonideal_items > (tbl->num_items >> 1)) ? \
711 (tbl->ineff_expands+1) : 0; \
712 if (tbl->ineff_expands > 1) { \
714 uthash_noexpand_fyi(tbl); \
716 uthash_expand_fyi(tbl); \
720 /* This is an adaptation of Simon Tatham's O(n log(n)) mergesort */
721 /* Note that HASH_SORT assumes the hash handle name to be hh.
722 * HASH_SRT was added to allow the hash handle name to be passed in. */
723 #define HASH_SORT(head,cmpfcn) HASH_SRT(hh,head,cmpfcn)
724 #define HASH_SRT(hh,head,cmpfcn) \
727 unsigned _hs_looping,_hs_nmerges,_hs_insize,_hs_psize,_hs_qsize; \
728 struct UT_hash_handle *_hs_p, *_hs_q, *_hs_e, *_hs_list, *_hs_tail; \
732 _hs_list = &((head)->hh); \
733 while (_hs_looping) { \
742 for ( _hs_i = 0; _hs_i < _hs_insize; _hs_i++ ) { \
744 _hs_q = (UT_hash_handle*)((_hs_q->next) ? \
745 ((void*)((char*)(_hs_q->next) + \
746 (head)->hh.tbl->hho)) : NULL); \
747 if (! (_hs_q) ) break; \
749 _hs_qsize = _hs_insize; \
750 while ((_hs_psize > 0) || ((_hs_qsize > 0) && _hs_q )) { \
751 if (_hs_psize == 0) { \
753 _hs_q = (UT_hash_handle*)((_hs_q->next) ? \
754 ((void*)((char*)(_hs_q->next) + \
755 (head)->hh.tbl->hho)) : NULL); \
757 } else if ( (_hs_qsize == 0) || !(_hs_q) ) { \
759 _hs_p = (UT_hash_handle*)((_hs_p->next) ? \
760 ((void*)((char*)(_hs_p->next) + \
761 (head)->hh.tbl->hho)) : NULL); \
764 cmpfcn(DECLTYPE(head)(ELMT_FROM_HH((head)->hh.tbl,_hs_p)), \
765 DECLTYPE(head)(ELMT_FROM_HH((head)->hh.tbl,_hs_q))) \
768 _hs_p = (UT_hash_handle*)((_hs_p->next) ? \
769 ((void*)((char*)(_hs_p->next) + \
770 (head)->hh.tbl->hho)) : NULL); \
774 _hs_q = (UT_hash_handle*)((_hs_q->next) ? \
775 ((void*)((char*)(_hs_q->next) + \
776 (head)->hh.tbl->hho)) : NULL); \
780 _hs_tail->next = ((_hs_e) ? \
781 ELMT_FROM_HH((head)->hh.tbl,_hs_e) : NULL); \
785 _hs_e->prev = ((_hs_tail) ? \
786 ELMT_FROM_HH((head)->hh.tbl,_hs_tail) : NULL); \
791 _hs_tail->next = NULL; \
792 if ( _hs_nmerges <= 1 ) { \
794 (head)->hh.tbl->tail = _hs_tail; \
795 DECLTYPE_ASSIGN(head,ELMT_FROM_HH((head)->hh.tbl, _hs_list)); \
799 HASH_FSCK(hh,head); \
803 /* This function selects items from one hash into another hash.
804 * The end result is that the selected items have dual presence
805 * in both hashes. There is no copy of the items made; rather
806 * they are added into the new hash through a secondary hash
807 * hash handle that must be present in the structure. */
808 #define HASH_SELECT(hh_dst, dst, hh_src, src, cond) \
810 unsigned _src_bkt, _dst_bkt; \
811 void *_last_elt=NULL, *_elt; \
812 UT_hash_handle *_src_hh, *_dst_hh, *_last_elt_hh=NULL; \
813 ptrdiff_t _dst_hho = ((char*)(&(dst)->hh_dst) - (char*)(dst)); \
815 for(_src_bkt=0; _src_bkt < (src)->hh_src.tbl->num_buckets; _src_bkt++) { \
816 for(_src_hh = (src)->hh_src.tbl->buckets[_src_bkt].hh_head; \
818 _src_hh = _src_hh->hh_next) { \
819 _elt = ELMT_FROM_HH((src)->hh_src.tbl, _src_hh); \
821 _dst_hh = (UT_hash_handle*)(((char*)_elt) + _dst_hho); \
822 _dst_hh->key = _src_hh->key; \
823 _dst_hh->keylen = _src_hh->keylen; \
824 _dst_hh->hashv = _src_hh->hashv; \
825 _dst_hh->prev = _last_elt; \
826 _dst_hh->next = NULL; \
827 if (_last_elt_hh) { _last_elt_hh->next = _elt; } \
829 DECLTYPE_ASSIGN(dst,_elt); \
830 HASH_MAKE_TABLE(hh_dst,dst); \
832 _dst_hh->tbl = (dst)->hh_dst.tbl; \
834 HASH_TO_BKT(_dst_hh->hashv, _dst_hh->tbl->num_buckets, _dst_bkt); \
835 HASH_ADD_TO_BKT(_dst_hh->tbl->buckets[_dst_bkt],_dst_hh); \
836 (dst)->hh_dst.tbl->num_items++; \
838 _last_elt_hh = _dst_hh; \
843 HASH_FSCK(hh_dst,dst); \
846 #define HASH_CLEAR(hh,head) \
849 uthash_free((head)->hh.tbl->buckets, \
850 (head)->hh.tbl->num_buckets*sizeof(struct UT_hash_bucket)); \
851 HASH_BLOOM_FREE((head)->hh.tbl); \
852 uthash_free((head)->hh.tbl, sizeof(UT_hash_table)); \
857 #define HASH_OVERHEAD(hh,head) \
858 (size_t)((((head)->hh.tbl->num_items * sizeof(UT_hash_handle)) + \
859 ((head)->hh.tbl->num_buckets * sizeof(UT_hash_bucket)) + \
860 (sizeof(UT_hash_table)) + \
861 (HASH_BLOOM_BYTELEN)))
865 #define HASH_ITER(hh,head,el,tmp) \
866 for((el)=(head), (*(char**)(&(tmp)))=(char*)((head)?(head)->hh.next:NULL); \
867 el; (el)=(tmp),(*(char**)(&(tmp)))=(char*)((tmp)?(tmp)->hh.next:NULL))
869 #define HASH_ITER_START(hh,head,el,tmp) \
870 do { (el)=(head); (*(char**)(&(tmp)))=(char*)((head)?(head)->hh.next:NULL); } while(0)
872 #define HASH_ITER_NEXT(hh,el,tmp) \
873 do { (el)=(tmp); (*(char**)(&(tmp)))=(char*)((tmp)?(tmp)->hh.next:NULL); } while(0)
877 #define HASH_ITER(hh,head,el,tmp) \
878 for((el)=(head),(tmp)=DECLTYPE(el)((head)?(head)->hh.next:NULL); \
879 el; (el)=(tmp),(tmp)=DECLTYPE(el)((tmp)?(tmp)->hh.next:NULL))
881 #define HASH_ITER_START(hh,head,el,tmp) \
882 do { (el)=(head); (tmp)=DECLTYPE(el)((head)?(head)->hh.next:NULL); } while(0)
884 #define HASH_ITER_NEXT(hh,el,tmp) \
885 do { (el)=(tmp); (tmp)=DECLTYPE(el)((tmp)?(tmp)->hh.next:NULL); } while(0)
889 /* obtain a count of items in the hash */
890 #define HASH_COUNT(head) HASH_CNT(hh,head)
891 #define HASH_CNT(hh,head) ((head)?((head)->hh.tbl->num_items):0)
893 typedef struct UT_hash_bucket
{
894 struct UT_hash_handle
*hh_head
;
897 /* expand_mult is normally set to 0. In this situation, the max chain length
898 * threshold is enforced at its default value, HASH_BKT_CAPACITY_THRESH. (If
899 * the bucket's chain exceeds this length, bucket expansion is triggered).
900 * However, setting expand_mult to a non-zero value delays bucket expansion
901 * (that would be triggered by additions to this particular bucket)
902 * until its chain length reaches a *multiple* of HASH_BKT_CAPACITY_THRESH.
903 * (The multiplier is simply expand_mult+1). The whole idea of this
904 * multiplier is to reduce bucket expansions, since they are expensive, in
905 * situations where we know that a particular bucket tends to be overused.
906 * It is better to let its chain length grow to a longer yet-still-bounded
907 * value, than to do an O(n) bucket expansion too often.
909 unsigned expand_mult
;
913 /* random signature used only to find hash tables in external analysis */
914 #define HASH_SIGNATURE 0xa0111fe1
915 #define HASH_BLOOM_SIGNATURE 0xb12220f2
917 typedef struct UT_hash_table
{
918 UT_hash_bucket
*buckets
;
919 unsigned num_buckets
, log2_num_buckets
;
921 struct UT_hash_handle
*tail
; /* tail hh in app order, for fast append */
922 ptrdiff_t hho
; /* hash handle offset (byte pos of hash handle in element */
924 /* in an ideal situation (all buckets used equally), no bucket would have
925 * more than ceil(#items/#buckets) items. that's the ideal chain length. */
926 unsigned ideal_chain_maxlen
;
928 /* nonideal_items is the number of items in the hash whose chain position
929 * exceeds the ideal chain maxlen. these items pay the penalty for an uneven
930 * hash distribution; reaching them in a chain traversal takes >ideal steps */
931 unsigned nonideal_items
;
933 /* ineffective expands occur when a bucket doubling was performed, but
934 * afterward, more than half the items in the hash had nonideal chain
935 * positions. If this happens on two consecutive expansions we inhibit any
936 * further expansion, as it's not helping; this happens when the hash
937 * function isn't a good fit for the key domain. When expansion is inhibited
938 * the hash will still work, albeit no longer in constant time. */
939 unsigned ineff_expands
, noexpand
;
941 uint32_t signature
; /* used only to find hash tables in external analysis */
943 uint32_t bloom_sig
; /* used only to test bloom exists in external analysis */
950 typedef struct UT_hash_handle
{
951 struct UT_hash_table
*tbl
;
952 void *prev
; /* prev element in app order */
953 void *next
; /* next element in app order */
954 struct UT_hash_handle
*hh_prev
; /* previous hh in bucket order */
955 struct UT_hash_handle
*hh_next
; /* next hh in bucket order */
956 void *key
; /* ptr to enclosing struct's key */
957 unsigned keylen
; /* enclosing struct's key len */
958 unsigned hashv
; /* result of hash-fcn(key) */
961 #endif /* UTHASH_H */