]>
Commit | Line | Data |
---|---|---|
d2471545 | 1 | |
2 | /* | |
3 | Copyright (c) 2003-2014, Troy D. Hanson http://troydhanson.github.com/uthash/ | |
4 | All rights reserved. | |
5 | Redistribution and use in source and binary forms, with or without | |
6 | modification, are permitted provided that the following conditions are met: | |
7 | * Redistributions of source code must retain the above copyright | |
8 | notice, this list of conditions and the following disclaimer. | |
9 | THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS | |
10 | IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED | |
11 | TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A | |
12 | PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER | |
13 | OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, | |
14 | EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, | |
15 | PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR | |
16 | PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF | |
17 | LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING | |
18 | NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS | |
19 | SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. | |
20 | */ | |
21 | ||
22 | #ifndef UTHASH_H | |
23 | #define UTHASH_H | |
24 | ||
25 | //#define HASH_BLOOM 16 | |
26 | ||
27 | #include <string.h> /* memcmp,strlen */ | |
28 | #include <stddef.h> /* ptrdiff_t */ | |
29 | #include <stdlib.h> /* exit() */ | |
30 | ||
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) */ | |
39 | #define NO_DECLTYPE | |
40 | #define DECLTYPE(x) | |
41 | #endif | |
42 | #elif defined(__BORLANDC__) || defined(__LCC__) || defined(__WATCOMC__) | |
43 | #define NO_DECLTYPE | |
44 | #define DECLTYPE(x) | |
45 | #else /* GNU, Sun and other compilers */ | |
46 | #define DECLTYPE(x) (__typeof(x)) | |
47 | #endif | |
48 | ||
49 | #ifdef NO_DECLTYPE | |
50 | #define DECLTYPE_ASSIGN(dst,src) \ | |
51 | do { \ | |
52 | char **_da_dst = (char**)(&(dst)); \ | |
53 | *_da_dst = (char*)(src); \ | |
54 | } while(0) | |
55 | #else | |
56 | #define DECLTYPE_ASSIGN(dst,src) \ | |
57 | do { \ | |
58 | (dst) = DECLTYPE(dst)(src); \ | |
59 | } while(0) | |
60 | #endif | |
61 | ||
62 | /* a number of the hash function use uint32_t which isn't defined on Pre VS2010 */ | |
63 | #if defined (_WIN32) | |
64 | #if defined(_MSC_VER) && _MSC_VER >= 1600 | |
65 | #include <stdint.h> | |
66 | #elif defined(__WATCOMC__) | |
67 | #include <stdint.h> | |
68 | #else | |
69 | //typedef unsigned int uint32_t; | |
70 | //typedef unsigned char uint8_t; | |
71 | #endif | |
72 | #else | |
73 | #include <stdint.h> | |
74 | #endif | |
75 | ||
76 | #define UTHASH_VERSION 1.9.9 | |
77 | ||
78 | #ifndef uthash_fatal | |
79 | #define uthash_fatal(msg) exit(-1) /* fatal error (out of memory,etc) */ | |
80 | #endif | |
81 | #ifndef uthash_malloc | |
82 | #define uthash_malloc(sz) malloc(sz) /* malloc fcn */ | |
83 | #endif | |
84 | #ifndef uthash_free | |
85 | #define uthash_free(ptr,sz) free(ptr) /* free fcn */ | |
86 | #endif | |
87 | ||
88 | #ifndef uthash_noexpand_fyi | |
89 | #define uthash_noexpand_fyi(tbl) /* can be defined to log noexpand */ | |
90 | #endif | |
91 | #ifndef uthash_expand_fyi | |
92 | #define uthash_expand_fyi(tbl) /* can be defined to log expands */ | |
93 | #endif | |
94 | ||
95 | /* initial number of buckets */ | |
96 | #ifndef HASH_INITIAL_NUM_BUCKETS_LOG2 | |
97 | #define HASH_INITIAL_NUM_BUCKETS_LOG2 5 /* lg2 of initial number of buckets */ | |
98 | #endif | |
99 | #define HASH_INITIAL_NUM_BUCKETS (1<<HASH_INITIAL_NUM_BUCKETS_LOG2) /* initial number of buckets */ | |
100 | #define HASH_BKT_CAPACITY_THRESH 10 /* expand when bucket count reaches */ | |
101 | ||
102 | /* calculate the element whose hash handle address is hhe */ | |
103 | #define ELMT_FROM_HH(tbl,hhp) ((void*)(((char*)(hhp)) - ((tbl)->hho))) | |
104 | ||
105 | #define HASH_FIND(hh,head,keyptr,keylen,out) \ | |
106 | do { \ | |
107 | out=NULL; \ | |
108 | if (head) { \ | |
109 | uint32_t _hf_bkt,_hf_hashv; \ | |
110 | HASH_FCN(keyptr,keylen, (head)->hh.tbl->num_buckets, _hf_hashv, _hf_bkt); \ | |
111 | if (HASH_BLOOM_TEST((head)->hh.tbl, _hf_hashv)) { \ | |
112 | HASH_FIND_IN_BKT((head)->hh.tbl, hh, (head)->hh.tbl->buckets[ _hf_bkt ], \ | |
113 | keyptr,keylen,out); \ | |
114 | } \ | |
115 | } \ | |
116 | } while (0) | |
117 | ||
118 | #ifdef HASH_BLOOM | |
119 | #define HASH_BLOOM_BITLEN (1ULL << HASH_BLOOM) | |
120 | #define HASH_BLOOM_BYTELEN (HASH_BLOOM_BITLEN/8) + ((HASH_BLOOM_BITLEN%8) ? 1:0) | |
121 | #define HASH_BLOOM_MAKE(tbl) \ | |
122 | do { \ | |
123 | (tbl)->bloom_nbits = HASH_BLOOM; \ | |
124 | (tbl)->bloom_bv = (uint8_t*)uthash_malloc(HASH_BLOOM_BYTELEN); \ | |
125 | if (!((tbl)->bloom_bv)) { uthash_fatal( "out of memory"); } \ | |
126 | memset((tbl)->bloom_bv, 0, HASH_BLOOM_BYTELEN); \ | |
127 | (tbl)->bloom_sig = HASH_BLOOM_SIGNATURE; \ | |
128 | } while (0) | |
129 | ||
130 | #define HASH_BLOOM_FREE(tbl) \ | |
131 | do { \ | |
132 | uthash_free((tbl)->bloom_bv, HASH_BLOOM_BYTELEN); \ | |
133 | } while (0) | |
134 | ||
135 | #define HASH_BLOOM_BITSET(bv,idx) (bv[(idx)/8] |= (1U << ((idx)%8))) | |
136 | #define HASH_BLOOM_BITTEST(bv,idx) (bv[(idx)/8] & (1U << ((idx)%8))) | |
137 | ||
138 | #define HASH_BLOOM_ADD(tbl,hashv) \ | |
139 | HASH_BLOOM_BITSET((tbl)->bloom_bv, (hashv & (uint32_t)((1ULL << (tbl)->bloom_nbits) - 1))) | |
140 | ||
141 | #define HASH_BLOOM_TEST(tbl,hashv) \ | |
142 | HASH_BLOOM_BITTEST((tbl)->bloom_bv, (hashv & (uint32_t)((1ULL << (tbl)->bloom_nbits) - 1))) | |
143 | ||
144 | #else | |
145 | #define HASH_BLOOM_MAKE(tbl) | |
146 | #define HASH_BLOOM_FREE(tbl) | |
147 | #define HASH_BLOOM_ADD(tbl,hashv) | |
148 | #define HASH_BLOOM_TEST(tbl,hashv) (1) | |
149 | #define HASH_BLOOM_BYTELEN 0 | |
150 | #endif | |
151 | ||
152 | #define HASH_MAKE_TABLE(hh,head) \ | |
153 | do { \ | |
154 | (head)->hh.tbl = (UT_hash_table*)uthash_malloc( \ | |
155 | sizeof(UT_hash_table)); \ | |
156 | if (!((head)->hh.tbl)) { uthash_fatal( "out of memory"); } \ | |
157 | memset((head)->hh.tbl, 0, sizeof(UT_hash_table)); \ | |
158 | (head)->hh.tbl->tail = &((head)->hh); \ | |
159 | (head)->hh.tbl->num_buckets = HASH_INITIAL_NUM_BUCKETS; \ | |
160 | (head)->hh.tbl->log2_num_buckets = HASH_INITIAL_NUM_BUCKETS_LOG2; \ | |
161 | (head)->hh.tbl->hho = (char*)(&(head)->hh) - (char*)(head); \ | |
162 | (head)->hh.tbl->buckets = (UT_hash_bucket*)uthash_malloc( \ | |
163 | HASH_INITIAL_NUM_BUCKETS*sizeof(struct UT_hash_bucket)); \ | |
164 | if (! (head)->hh.tbl->buckets) { uthash_fatal( "out of memory"); } \ | |
165 | memset((head)->hh.tbl->buckets, 0, \ | |
166 | HASH_INITIAL_NUM_BUCKETS*sizeof(struct UT_hash_bucket)); \ | |
167 | HASH_BLOOM_MAKE((head)->hh.tbl); \ | |
168 | (head)->hh.tbl->signature = HASH_SIGNATURE; \ | |
169 | } while(0) | |
170 | ||
171 | #define HASH_ADD(hh,head,fieldname,keylen_in,add) \ | |
172 | HASH_ADD_KEYPTR(hh,head,&((add)->fieldname),keylen_in,add) | |
173 | ||
174 | #define HASH_REPLACE(hh,head,fieldname,keylen_in,add,replaced) \ | |
175 | do { \ | |
176 | replaced=NULL; \ | |
177 | HASH_FIND(hh,head,&((add)->fieldname),keylen_in,replaced); \ | |
178 | if (replaced!=NULL) { \ | |
179 | HASH_DELETE(hh,head,replaced); \ | |
180 | }; \ | |
181 | HASH_ADD(hh,head,fieldname,keylen_in,add); \ | |
182 | } while(0) | |
183 | ||
184 | #define HASH_ADD_KEYPTR(hh,head,keyptr,keylen_in,add) \ | |
185 | do { \ | |
186 | uint32_t _ha_bkt; \ | |
187 | (add)->hh.next = NULL; \ | |
188 | (add)->hh.key = (char*)(keyptr); \ | |
189 | (add)->hh.keylen = (uint32_t)(keylen_in); \ | |
190 | if (!(head)) { \ | |
191 | head = (add); \ | |
192 | (head)->hh.prev = NULL; \ | |
193 | HASH_MAKE_TABLE(hh,head); \ | |
194 | } else { \ | |
195 | (head)->hh.tbl->tail->next = (add); \ | |
196 | (add)->hh.prev = ELMT_FROM_HH((head)->hh.tbl, (head)->hh.tbl->tail); \ | |
197 | (head)->hh.tbl->tail = &((add)->hh); \ | |
198 | } \ | |
199 | (head)->hh.tbl->num_items++; \ | |
200 | (add)->hh.tbl = (head)->hh.tbl; \ | |
201 | HASH_FCN(keyptr,keylen_in, (head)->hh.tbl->num_buckets, \ | |
202 | (add)->hh.hashv, _ha_bkt); \ | |
203 | HASH_ADD_TO_BKT((head)->hh.tbl->buckets[_ha_bkt],&(add)->hh); \ | |
204 | HASH_BLOOM_ADD((head)->hh.tbl,(add)->hh.hashv); \ | |
205 | HASH_EMIT_KEY(hh,head,keyptr,keylen_in); \ | |
206 | HASH_FSCK(hh,head); \ | |
207 | } while(0) | |
208 | ||
209 | #define HASH_TO_BKT( hashv, num_bkts, bkt ) \ | |
210 | do { \ | |
211 | bkt = ((hashv) & ((num_bkts) - 1)); \ | |
212 | } while(0) | |
213 | ||
214 | /* delete "delptr" from the hash table. | |
215 | * "the usual" patch-up process for the app-order doubly-linked-list. | |
216 | * The use of _hd_hh_del below deserves special explanation. | |
217 | * These used to be expressed using (delptr) but that led to a bug | |
218 | * if someone used the same symbol for the head and deletee, like | |
219 | * HASH_DELETE(hh,users,users); | |
220 | * We want that to work, but by changing the head (users) below | |
221 | * we were forfeiting our ability to further refer to the deletee (users) | |
222 | * in the patch-up process. Solution: use scratch space to | |
223 | * copy the deletee pointer, then the latter references are via that | |
224 | * scratch pointer rather than through the repointed (users) symbol. | |
225 | */ | |
226 | #define HASH_DELETE(hh,head,delptr) \ | |
227 | do { \ | |
228 | struct UT_hash_handle *_hd_hh_del; \ | |
229 | if ( ((delptr)->hh.prev == NULL) && ((delptr)->hh.next == NULL) ) { \ | |
230 | uthash_free((head)->hh.tbl->buckets, \ | |
231 | (head)->hh.tbl->num_buckets*sizeof(struct UT_hash_bucket) ); \ | |
232 | HASH_BLOOM_FREE((head)->hh.tbl); \ | |
233 | uthash_free((head)->hh.tbl, sizeof(UT_hash_table)); \ | |
234 | head = NULL; \ | |
235 | } else { \ | |
236 | uint32_t _hd_bkt; \ | |
237 | _hd_hh_del = &((delptr)->hh); \ | |
238 | if ((delptr) == ELMT_FROM_HH((head)->hh.tbl,(head)->hh.tbl->tail)) { \ | |
239 | (head)->hh.tbl->tail = \ | |
240 | (UT_hash_handle*)((ptrdiff_t)((delptr)->hh.prev) + \ | |
241 | (head)->hh.tbl->hho); \ | |
242 | } \ | |
243 | if ((delptr)->hh.prev) { \ | |
244 | ((UT_hash_handle*)((ptrdiff_t)((delptr)->hh.prev) + \ | |
245 | (head)->hh.tbl->hho))->next = (delptr)->hh.next; \ | |
246 | } else { \ | |
247 | DECLTYPE_ASSIGN(head,(delptr)->hh.next); \ | |
248 | } \ | |
249 | if (_hd_hh_del->next) { \ | |
250 | ((UT_hash_handle*)((ptrdiff_t)_hd_hh_del->next + \ | |
251 | (head)->hh.tbl->hho))->prev = \ | |
252 | _hd_hh_del->prev; \ | |
253 | } \ | |
254 | HASH_TO_BKT( _hd_hh_del->hashv, (head)->hh.tbl->num_buckets, _hd_bkt); \ | |
255 | HASH_DEL_IN_BKT(hh,(head)->hh.tbl->buckets[_hd_bkt], _hd_hh_del); \ | |
256 | (head)->hh.tbl->num_items--; \ | |
257 | } \ | |
258 | HASH_FSCK(hh,head); \ | |
259 | } while (0) | |
260 | ||
261 | ||
262 | /* convenience forms of HASH_FIND/HASH_ADD/HASH_DEL */ | |
263 | #define HASH_FIND_STR(head,findstr,out) \ | |
264 | HASH_FIND(hh,head,findstr,(uint32_t)strlen(findstr),out) | |
265 | #define HASH_ADD_STR(head,strfield,add) \ | |
266 | HASH_ADD(hh,head,strfield[0],strlen(add->strfield),add) | |
267 | #define HASH_REPLACE_STR(head,strfield,add,replaced) \ | |
268 | HASH_REPLACE(hh,head,strfield[0],(uint32_t)strlen(add->strfield),add,replaced) | |
269 | #define HASH_FIND_INT(head,findint,out) \ | |
270 | HASH_FIND(hh,head,findint,sizeof(int),out) | |
271 | #define HASH_ADD_INT(head,intfield,add) \ | |
272 | HASH_ADD(hh,head,intfield,sizeof(int),add) | |
273 | #define HASH_REPLACE_INT(head,intfield,add,replaced) \ | |
274 | HASH_REPLACE(hh,head,intfield,sizeof(int),add,replaced) | |
275 | #define HASH_FIND_PTR(head,findptr,out) \ | |
276 | HASH_FIND(hh,head,findptr,sizeof(void *),out) | |
277 | #define HASH_ADD_PTR(head,ptrfield,add) \ | |
278 | HASH_ADD(hh,head,ptrfield,sizeof(void *),add) | |
279 | #define HASH_REPLACE_PTR(head,ptrfield,add,replaced) \ | |
280 | HASH_REPLACE(hh,head,ptrfield,sizeof(void *),add,replaced) | |
281 | #define HASH_DEL(head,delptr) \ | |
282 | HASH_DELETE(hh,head,delptr) | |
283 | ||
284 | /* HASH_FSCK checks hash integrity on every add/delete when HASH_DEBUG is defined. | |
285 | * This is for uthash developer only; it compiles away if HASH_DEBUG isn't defined. | |
286 | */ | |
287 | #ifdef HASH_DEBUG | |
288 | #define HASH_OOPS(...) do { fprintf(stderr,__VA_ARGS__); exit(-1); } while (0) | |
289 | #define HASH_FSCK(hh,head) \ | |
290 | do { \ | |
291 | struct UT_hash_handle *_thh; \ | |
292 | if (head) { \ | |
293 | uint32_t _bkt_i; \ | |
294 | uint32_t _count; \ | |
295 | char *_prev; \ | |
296 | _count = 0; \ | |
297 | for( _bkt_i = 0; _bkt_i < (head)->hh.tbl->num_buckets; _bkt_i++) { \ | |
298 | uint32_t _bkt_count = 0; \ | |
299 | _thh = (head)->hh.tbl->buckets[_bkt_i].hh_head; \ | |
300 | _prev = NULL; \ | |
301 | while (_thh) { \ | |
302 | if (_prev != (char*)(_thh->hh_prev)) { \ | |
303 | HASH_OOPS("invalid hh_prev %p, actual %p\n", \ | |
304 | _thh->hh_prev, _prev ); \ | |
305 | } \ | |
306 | _bkt_count++; \ | |
307 | _prev = (char*)(_thh); \ | |
308 | _thh = _thh->hh_next; \ | |
309 | } \ | |
310 | _count += _bkt_count; \ | |
311 | if ((head)->hh.tbl->buckets[_bkt_i].count != _bkt_count) { \ | |
312 | HASH_OOPS("invalid bucket count %u, actual %u\n", \ | |
313 | (head)->hh.tbl->buckets[_bkt_i].count, _bkt_count); \ | |
314 | } \ | |
315 | } \ | |
316 | if (_count != (head)->hh.tbl->num_items) { \ | |
317 | HASH_OOPS("invalid hh item count %u, actual %u\n", \ | |
318 | (head)->hh.tbl->num_items, _count ); \ | |
319 | } \ | |
320 | /* traverse hh in app order; check next/prev integrity, count */ \ | |
321 | _count = 0; \ | |
322 | _prev = NULL; \ | |
323 | _thh = &(head)->hh; \ | |
324 | while (_thh) { \ | |
325 | _count++; \ | |
326 | if (_prev !=(char*)(_thh->prev)) { \ | |
327 | HASH_OOPS("invalid prev %p, actual %p\n", \ | |
328 | _thh->prev, _prev ); \ | |
329 | } \ | |
330 | _prev = (char*)ELMT_FROM_HH((head)->hh.tbl, _thh); \ | |
331 | _thh = ( _thh->next ? (UT_hash_handle*)((char*)(_thh->next) + \ | |
332 | (head)->hh.tbl->hho) : NULL ); \ | |
333 | } \ | |
334 | if (_count != (head)->hh.tbl->num_items) { \ | |
335 | HASH_OOPS("invalid app item count %u, actual %u\n", \ | |
336 | (head)->hh.tbl->num_items, _count ); \ | |
337 | } \ | |
338 | } \ | |
339 | } while (0) | |
340 | #else | |
341 | #define HASH_FSCK(hh,head) | |
342 | #endif | |
343 | ||
344 | /* When compiled with -DHASH_EMIT_KEYS, length-prefixed keys are emitted to | |
345 | * the descriptor to which this macro is defined for tuning the hash function. | |
346 | * The app can #include <unistd.h> to get the prototype for write(2). */ | |
347 | #ifdef HASH_EMIT_KEYS | |
348 | #define HASH_EMIT_KEY(hh,head,keyptr,fieldlen) \ | |
349 | do { \ | |
350 | uint32_t _klen = fieldlen; \ | |
351 | write(HASH_EMIT_KEYS, &_klen, sizeof(_klen)); \ | |
352 | write(HASH_EMIT_KEYS, keyptr, fieldlen); \ | |
353 | } while (0) | |
354 | #else | |
355 | #define HASH_EMIT_KEY(hh,head,keyptr,fieldlen) | |
356 | #endif | |
357 | ||
358 | /* default to Jenkin's hash unless overridden e.g. DHASH_FUNCTION=HASH_SAX */ | |
359 | #ifdef HASH_FUNCTION | |
360 | #define HASH_FCN HASH_FUNCTION | |
361 | #else | |
362 | #define HASH_FCN HASH_JEN | |
363 | #endif | |
364 | ||
365 | /* The Bernstein hash function, used in Perl prior to v5.6. Note (x<<5+x)=x*33. */ | |
366 | #define HASH_BER(key,keylen,num_bkts,hashv,bkt) \ | |
367 | do { \ | |
368 | uint32_t _hb_keylen=keylen; \ | |
369 | char *_hb_key=(char*)(key); \ | |
370 | (hashv) = 0; \ | |
371 | while (_hb_keylen--) { (hashv) = (((hashv) << 5) + (hashv)) + *_hb_key++; } \ | |
372 | bkt = (hashv) & (num_bkts-1); \ | |
373 | } while (0) | |
374 | ||
375 | ||
376 | /* SAX/FNV/OAT/JEN hash functions are macro variants of those listed at | |
377 | * http://eternallyconfuzzled.com/tuts/algorithms/jsw_tut_hashing.aspx */ | |
378 | #define HASH_SAX(key,keylen,num_bkts,hashv,bkt) \ | |
379 | do { \ | |
380 | uint32_t _sx_i; \ | |
381 | char *_hs_key=(char*)(key); \ | |
382 | hashv = 0; \ | |
383 | for(_sx_i=0; _sx_i < keylen; _sx_i++) \ | |
384 | hashv ^= (hashv << 5) + (hashv >> 2) + _hs_key[_sx_i]; \ | |
385 | bkt = hashv & (num_bkts-1); \ | |
386 | } while (0) | |
387 | /* FNV-1a variation */ | |
388 | #define HASH_FNV(key,keylen,num_bkts,hashv,bkt) \ | |
389 | do { \ | |
390 | uint32_t _fn_i; \ | |
391 | char *_hf_key=(char*)(key); \ | |
392 | hashv = 2166136261UL; \ | |
393 | for(_fn_i=0; _fn_i < keylen; _fn_i++) { \ | |
394 | hashv = hashv ^ _hf_key[_fn_i]; \ | |
395 | hashv = hashv * 16777619; \ | |
396 | } \ | |
397 | bkt = hashv & (num_bkts-1); \ | |
398 | } while(0) | |
399 | ||
400 | #define HASH_OAT(key,keylen,num_bkts,hashv,bkt) \ | |
401 | do { \ | |
402 | uint32_t _ho_i; \ | |
403 | char *_ho_key=(char*)(key); \ | |
404 | hashv = 0; \ | |
405 | for(_ho_i=0; _ho_i < keylen; _ho_i++) { \ | |
406 | hashv += _ho_key[_ho_i]; \ | |
407 | hashv += (hashv << 10); \ | |
408 | hashv ^= (hashv >> 6); \ | |
409 | } \ | |
410 | hashv += (hashv << 3); \ | |
411 | hashv ^= (hashv >> 11); \ | |
412 | hashv += (hashv << 15); \ | |
413 | bkt = hashv & (num_bkts-1); \ | |
414 | } while(0) | |
415 | ||
416 | #define HASH_JEN_MIX(a,b,c) \ | |
417 | do { \ | |
418 | a -= b; a -= c; a ^= ( c >> 13 ); \ | |
419 | b -= c; b -= a; b ^= ( a << 8 ); \ | |
420 | c -= a; c -= b; c ^= ( b >> 13 ); \ | |
421 | a -= b; a -= c; a ^= ( c >> 12 ); \ | |
422 | b -= c; b -= a; b ^= ( a << 16 ); \ | |
423 | c -= a; c -= b; c ^= ( b >> 5 ); \ | |
424 | a -= b; a -= c; a ^= ( c >> 3 ); \ | |
425 | b -= c; b -= a; b ^= ( a << 10 ); \ | |
426 | c -= a; c -= b; c ^= ( b >> 15 ); \ | |
427 | } while (0) | |
428 | ||
429 | #define HASH_JEN(key,keylen,num_bkts,hashv,bkt) \ | |
430 | do { \ | |
431 | uint32_t _hj_i,_hj_j,_hj_k; \ | |
432 | unsigned char *_hj_key=(unsigned char*)(key); \ | |
433 | hashv = 0xfeedbeef; \ | |
434 | _hj_i = _hj_j = 0x9e3779b9; \ | |
435 | _hj_k = (uint32_t)(keylen); \ | |
436 | while (_hj_k >= 12) { \ | |
437 | _hj_i += (_hj_key[0] + ( (uint32_t)_hj_key[1] << 8 ) \ | |
438 | + ( (uint32_t)_hj_key[2] << 16 ) \ | |
439 | + ( (uint32_t)_hj_key[3] << 24 ) ); \ | |
440 | _hj_j += (_hj_key[4] + ( (uint32_t)_hj_key[5] << 8 ) \ | |
441 | + ( (uint32_t)_hj_key[6] << 16 ) \ | |
442 | + ( (uint32_t)_hj_key[7] << 24 ) ); \ | |
443 | hashv += (_hj_key[8] + ( (uint32_t)_hj_key[9] << 8 ) \ | |
444 | + ( (uint32_t)_hj_key[10] << 16 ) \ | |
445 | + ( (uint32_t)_hj_key[11] << 24 ) ); \ | |
446 | \ | |
447 | HASH_JEN_MIX(_hj_i, _hj_j, hashv); \ | |
448 | \ | |
449 | _hj_key += 12; \ | |
450 | _hj_k -= 12; \ | |
451 | } \ | |
452 | hashv += keylen; \ | |
453 | switch ( _hj_k ) { \ | |
454 | case 11: hashv += ( (uint32_t)_hj_key[10] << 24 ); \ | |
455 | case 10: hashv += ( (uint32_t)_hj_key[9] << 16 ); \ | |
456 | case 9: hashv += ( (uint32_t)_hj_key[8] << 8 ); \ | |
457 | case 8: _hj_j += ( (uint32_t)_hj_key[7] << 24 ); \ | |
458 | case 7: _hj_j += ( (uint32_t)_hj_key[6] << 16 ); \ | |
459 | case 6: _hj_j += ( (uint32_t)_hj_key[5] << 8 ); \ | |
460 | case 5: _hj_j += _hj_key[4]; \ | |
461 | case 4: _hj_i += ( (uint32_t)_hj_key[3] << 24 ); \ | |
462 | case 3: _hj_i += ( (uint32_t)_hj_key[2] << 16 ); \ | |
463 | case 2: _hj_i += ( (uint32_t)_hj_key[1] << 8 ); \ | |
464 | case 1: _hj_i += _hj_key[0]; \ | |
465 | } \ | |
466 | HASH_JEN_MIX(_hj_i, _hj_j, hashv); \ | |
467 | bkt = hashv & (num_bkts-1); \ | |
468 | } while(0) | |
469 | ||
470 | /* The Paul Hsieh hash function */ | |
471 | #undef get16bits | |
472 | #if (defined(__GNUC__) && defined(__i386__)) || defined(__WATCOMC__) \ | |
473 | || defined(_MSC_VER) || defined (__BORLANDC__) || defined (__TURBOC__) | |
474 | #define get16bits(d) (*((const uint16_t *) (d))) | |
475 | #endif | |
476 | ||
477 | #if !defined (get16bits) | |
478 | #define get16bits(d) ((((uint32_t)(((const uint8_t *)(d))[1])) << 8) \ | |
479 | +(uint32_t)(((const uint8_t *)(d))[0]) ) | |
480 | #endif | |
481 | #define HASH_SFH(key,keylen,num_bkts,hashv,bkt) \ | |
482 | do { \ | |
483 | unsigned char *_sfh_key=(unsigned char*)(key); \ | |
484 | uint32_t _sfh_tmp, _sfh_len = keylen; \ | |
485 | \ | |
486 | int _sfh_rem = _sfh_len & 3; \ | |
487 | _sfh_len >>= 2; \ | |
488 | hashv = 0xcafebabe; \ | |
489 | \ | |
490 | /* Main loop */ \ | |
491 | for (;_sfh_len > 0; _sfh_len--) { \ | |
492 | hashv += get16bits (_sfh_key); \ | |
493 | _sfh_tmp = (uint32_t)(get16bits (_sfh_key+2)) << 11 ^ hashv; \ | |
494 | hashv = (hashv << 16) ^ _sfh_tmp; \ | |
495 | _sfh_key += 2*sizeof (uint16_t); \ | |
496 | hashv += hashv >> 11; \ | |
497 | } \ | |
498 | \ | |
499 | /* Handle end cases */ \ | |
500 | switch (_sfh_rem) { \ | |
501 | case 3: hashv += get16bits (_sfh_key); \ | |
502 | hashv ^= hashv << 16; \ | |
503 | hashv ^= (uint32_t)(_sfh_key[sizeof (uint16_t)] << 18); \ | |
504 | hashv += hashv >> 11; \ | |
505 | break; \ | |
506 | case 2: hashv += get16bits (_sfh_key); \ | |
507 | hashv ^= hashv << 11; \ | |
508 | hashv += hashv >> 17; \ | |
509 | break; \ | |
510 | case 1: hashv += *_sfh_key; \ | |
511 | hashv ^= hashv << 10; \ | |
512 | hashv += hashv >> 1; \ | |
513 | } \ | |
514 | \ | |
515 | /* Force "avalanching" of final 127 bits */ \ | |
516 | hashv ^= hashv << 3; \ | |
517 | hashv += hashv >> 5; \ | |
518 | hashv ^= hashv << 4; \ | |
519 | hashv += hashv >> 17; \ | |
520 | hashv ^= hashv << 25; \ | |
521 | hashv += hashv >> 6; \ | |
522 | bkt = hashv & (num_bkts-1); \ | |
523 | } while(0) | |
524 | ||
525 | #ifdef HASH_USING_NO_STRICT_ALIASING | |
526 | /* The MurmurHash exploits some CPU's (x86,x86_64) tolerance for unaligned reads. | |
527 | * For other types of CPU's (e.g. Sparc) an unaligned read causes a bus error. | |
528 | * MurmurHash uses the faster approach only on CPU's where we know it's safe. | |
529 | * | |
530 | * Note the preprocessor built-in defines can be emitted using: | |
531 | * | |
532 | * gcc -m64 -dM -E - < /dev/null (on gcc) | |
533 | * cc -## a.c (where a.c is a simple test file) (Sun Studio) | |
534 | */ | |
535 | #if (defined(__i386__) || defined(__x86_64__) || defined(_M_IX86)) | |
536 | #define MUR_GETBLOCK(p,i) p[i] | |
537 | #else /* non intel */ | |
538 | #define MUR_PLUS0_ALIGNED(p) (((uint64_t)p & 0x3) == 0) | |
539 | #define MUR_PLUS1_ALIGNED(p) (((uint64_t)p & 0x3) == 1) | |
540 | #define MUR_PLUS2_ALIGNED(p) (((uint64_t)p & 0x3) == 2) | |
541 | #define MUR_PLUS3_ALIGNED(p) (((uint64_t)p & 0x3) == 3) | |
542 | #define WP(p) ((uint32_t*)((uint64_t)(p) & ~3UL)) | |
543 | #if (defined(__BIG_ENDIAN__) || defined(SPARC) || defined(__ppc__) || defined(__ppc64__)) | |
544 | #define MUR_THREE_ONE(p) ((((*WP(p))&0x00ffffff) << 8) | (((*(WP(p)+1))&0xff000000) >> 24)) | |
545 | #define MUR_TWO_TWO(p) ((((*WP(p))&0x0000ffff) <<16) | (((*(WP(p)+1))&0xffff0000) >> 16)) | |
546 | #define MUR_ONE_THREE(p) ((((*WP(p))&0x000000ff) <<24) | (((*(WP(p)+1))&0xffffff00) >> 8)) | |
547 | #else /* assume little endian non-intel */ | |
548 | #define MUR_THREE_ONE(p) ((((*WP(p))&0xffffff00) >> 8) | (((*(WP(p)+1))&0x000000ff) << 24)) | |
549 | #define MUR_TWO_TWO(p) ((((*WP(p))&0xffff0000) >>16) | (((*(WP(p)+1))&0x0000ffff) << 16)) | |
550 | #define MUR_ONE_THREE(p) ((((*WP(p))&0xff000000) >>24) | (((*(WP(p)+1))&0x00ffffff) << 8)) | |
551 | #endif | |
552 | #define MUR_GETBLOCK(p,i) (MUR_PLUS0_ALIGNED(p) ? ((p)[i]) : \ | |
553 | (MUR_PLUS1_ALIGNED(p) ? MUR_THREE_ONE(p) : \ | |
554 | (MUR_PLUS2_ALIGNED(p) ? MUR_TWO_TWO(p) : \ | |
555 | MUR_ONE_THREE(p)))) | |
556 | #endif | |
557 | #define MUR_ROTL32(x,r) (((x) << (r)) | ((x) >> (32 - (r)))) | |
558 | #define MUR_FMIX(_h) \ | |
559 | do { \ | |
560 | _h ^= _h >> 16; \ | |
561 | _h *= 0x85ebca6b; \ | |
562 | _h ^= _h >> 13; \ | |
563 | _h *= 0xc2b2ae35l; \ | |
564 | _h ^= _h >> 16; \ | |
565 | } while(0) | |
566 | ||
567 | #define HASH_MUR(key,keylen,num_bkts,hashv,bkt) \ | |
568 | do { \ | |
569 | const uint8_t *_mur_data = (const uint8_t*)(key); \ | |
570 | const int _mur_nblocks = (keylen) / 4; \ | |
571 | uint32_t _mur_h1 = 0xf88D5353; \ | |
572 | uint32_t _mur_c1 = 0xcc9e2d51; \ | |
573 | uint32_t _mur_c2 = 0x1b873593; \ | |
574 | uint32_t _mur_k1 = 0; \ | |
575 | const uint8_t *_mur_tail; \ | |
576 | const uint32_t *_mur_blocks = (const uint32_t*)(_mur_data+_mur_nblocks*4); \ | |
577 | int _mur_i; \ | |
578 | for(_mur_i = -_mur_nblocks; _mur_i; _mur_i++) { \ | |
579 | _mur_k1 = MUR_GETBLOCK(_mur_blocks,_mur_i); \ | |
580 | _mur_k1 *= _mur_c1; \ | |
581 | _mur_k1 = MUR_ROTL32(_mur_k1,15); \ | |
582 | _mur_k1 *= _mur_c2; \ | |
583 | \ | |
584 | _mur_h1 ^= _mur_k1; \ | |
585 | _mur_h1 = MUR_ROTL32(_mur_h1,13); \ | |
586 | _mur_h1 = _mur_h1*5+0xe6546b64; \ | |
587 | } \ | |
588 | _mur_tail = (const uint8_t*)(_mur_data + _mur_nblocks*4); \ | |
589 | _mur_k1=0; \ | |
590 | switch((keylen) & 3) { \ | |
591 | case 3: _mur_k1 ^= _mur_tail[2] << 16; \ | |
592 | case 2: _mur_k1 ^= _mur_tail[1] << 8; \ | |
593 | case 1: _mur_k1 ^= _mur_tail[0]; \ | |
594 | _mur_k1 *= _mur_c1; \ | |
595 | _mur_k1 = MUR_ROTL32(_mur_k1,15); \ | |
596 | _mur_k1 *= _mur_c2; \ | |
597 | _mur_h1 ^= _mur_k1; \ | |
598 | } \ | |
599 | _mur_h1 ^= (keylen); \ | |
600 | MUR_FMIX(_mur_h1); \ | |
601 | hashv = _mur_h1; \ | |
602 | bkt = hashv & (num_bkts-1); \ | |
603 | } while(0) | |
604 | #endif /* HASH_USING_NO_STRICT_ALIASING */ | |
605 | ||
606 | /* key comparison function; return 0 if keys equal */ | |
607 | #define HASH_KEYCMP(a,b,len) memcmp(a,b,len) | |
608 | ||
609 | /* iterate over items in a known bucket to find desired item */ | |
610 | #define HASH_FIND_IN_BKT(tbl,hh,head,keyptr,keylen_in,out) \ | |
611 | do { \ | |
612 | if (head.hh_head) DECLTYPE_ASSIGN(out,ELMT_FROM_HH(tbl,head.hh_head)); \ | |
613 | else out=NULL; \ | |
614 | while (out) { \ | |
615 | if ((out)->hh.keylen == keylen_in) { \ | |
616 | if ((HASH_KEYCMP((out)->hh.key,keyptr,keylen_in)) == 0) break; \ | |
617 | } \ | |
618 | if ((out)->hh.hh_next) DECLTYPE_ASSIGN(out,ELMT_FROM_HH(tbl,(out)->hh.hh_next)); \ | |
619 | else out = NULL; \ | |
620 | } \ | |
621 | } while(0) | |
622 | ||
623 | /* add an item to a bucket */ | |
624 | #define HASH_ADD_TO_BKT(head,addhh) \ | |
625 | do { \ | |
626 | head.count++; \ | |
627 | (addhh)->hh_next = head.hh_head; \ | |
628 | (addhh)->hh_prev = NULL; \ | |
629 | if (head.hh_head) { (head).hh_head->hh_prev = (addhh); } \ | |
630 | (head).hh_head=addhh; \ | |
631 | if (head.count >= ((head.expand_mult+1) * HASH_BKT_CAPACITY_THRESH) \ | |
632 | && (addhh)->tbl->noexpand != 1) { \ | |
633 | HASH_EXPAND_BUCKETS((addhh)->tbl); \ | |
634 | } \ | |
635 | } while(0) | |
636 | ||
637 | /* remove an item from a given bucket */ | |
638 | #define HASH_DEL_IN_BKT(hh,head,hh_del) \ | |
639 | (head).count--; \ | |
640 | if ((head).hh_head == hh_del) { \ | |
641 | (head).hh_head = hh_del->hh_next; \ | |
642 | } \ | |
643 | if (hh_del->hh_prev) { \ | |
644 | hh_del->hh_prev->hh_next = hh_del->hh_next; \ | |
645 | } \ | |
646 | if (hh_del->hh_next) { \ | |
647 | hh_del->hh_next->hh_prev = hh_del->hh_prev; \ | |
648 | } | |
649 | ||
650 | /* Bucket expansion has the effect of doubling the number of buckets | |
651 | * and redistributing the items into the new buckets. Ideally the | |
652 | * items will distribute more or less evenly into the new buckets | |
653 | * (the extent to which this is true is a measure of the quality of | |
654 | * the hash function as it applies to the key domain). | |
655 | * | |
656 | * With the items distributed into more buckets, the chain length | |
657 | * (item count) in each bucket is reduced. Thus by expanding buckets | |
658 | * the hash keeps a bound on the chain length. This bounded chain | |
659 | * length is the essence of how a hash provides constant time lookup. | |
660 | * | |
661 | * The calculation of tbl->ideal_chain_maxlen below deserves some | |
662 | * explanation. First, keep in mind that we're calculating the ideal | |
663 | * maximum chain length based on the *new* (doubled) bucket count. | |
664 | * In fractions this is just n/b (n=number of items,b=new num buckets). | |
665 | * Since the ideal chain length is an integer, we want to calculate | |
666 | * ceil(n/b). We don't depend on floating point arithmetic in this | |
667 | * hash, so to calculate ceil(n/b) with integers we could write | |
668 | * | |
669 | * ceil(n/b) = (n/b) + ((n%b)?1:0) | |
670 | * | |
671 | * and in fact a previous version of this hash did just that. | |
672 | * But now we have improved things a bit by recognizing that b is | |
673 | * always a power of two. We keep its base 2 log handy (call it lb), | |
674 | * so now we can write this with a bit shift and logical AND: | |
675 | * | |
676 | * ceil(n/b) = (n>>lb) + ( (n & (b-1)) ? 1:0) | |
677 | * | |
678 | */ | |
679 | #define HASH_EXPAND_BUCKETS(tbl) \ | |
680 | do { \ | |
681 | uint32_t _he_bkt; \ | |
682 | uint32_t _he_bkt_i; \ | |
683 | struct UT_hash_handle *_he_thh, *_he_hh_nxt; \ | |
684 | UT_hash_bucket *_he_new_buckets, *_he_newbkt; \ | |
685 | _he_new_buckets = (UT_hash_bucket*)uthash_malloc( \ | |
686 | 2 * tbl->num_buckets * sizeof(struct UT_hash_bucket)); \ | |
687 | if (!_he_new_buckets) { uthash_fatal( "out of memory"); } \ | |
688 | memset(_he_new_buckets, 0, \ | |
689 | 2 * tbl->num_buckets * sizeof(struct UT_hash_bucket)); \ | |
690 | tbl->ideal_chain_maxlen = \ | |
691 | (tbl->num_items >> (tbl->log2_num_buckets+1)) + \ | |
692 | ((tbl->num_items & ((tbl->num_buckets*2)-1)) ? 1 : 0); \ | |
693 | tbl->nonideal_items = 0; \ | |
694 | for(_he_bkt_i = 0; _he_bkt_i < tbl->num_buckets; _he_bkt_i++) \ | |
695 | { \ | |
696 | _he_thh = tbl->buckets[ _he_bkt_i ].hh_head; \ | |
697 | while (_he_thh) { \ | |
698 | _he_hh_nxt = _he_thh->hh_next; \ | |
699 | HASH_TO_BKT( _he_thh->hashv, tbl->num_buckets*2, _he_bkt); \ | |
700 | _he_newbkt = &(_he_new_buckets[ _he_bkt ]); \ | |
701 | if (++(_he_newbkt->count) > tbl->ideal_chain_maxlen) { \ | |
702 | tbl->nonideal_items++; \ | |
703 | _he_newbkt->expand_mult = _he_newbkt->count / \ | |
704 | tbl->ideal_chain_maxlen; \ | |
705 | } \ | |
706 | _he_thh->hh_prev = NULL; \ | |
707 | _he_thh->hh_next = _he_newbkt->hh_head; \ | |
708 | if (_he_newbkt->hh_head) _he_newbkt->hh_head->hh_prev = \ | |
709 | _he_thh; \ | |
710 | _he_newbkt->hh_head = _he_thh; \ | |
711 | _he_thh = _he_hh_nxt; \ | |
712 | } \ | |
713 | } \ | |
714 | uthash_free( tbl->buckets, tbl->num_buckets*sizeof(struct UT_hash_bucket) ); \ | |
715 | tbl->num_buckets *= 2; \ | |
716 | tbl->log2_num_buckets++; \ | |
717 | tbl->buckets = _he_new_buckets; \ | |
718 | tbl->ineff_expands = (tbl->nonideal_items > (tbl->num_items >> 1)) ? \ | |
719 | (tbl->ineff_expands+1) : 0; \ | |
720 | if (tbl->ineff_expands > 1) { \ | |
721 | tbl->noexpand=1; \ | |
722 | uthash_noexpand_fyi(tbl); \ | |
723 | } \ | |
724 | uthash_expand_fyi(tbl); \ | |
725 | } while(0) | |
726 | ||
727 | ||
728 | /* This is an adaptation of Simon Tatham's O(n log(n)) mergesort */ | |
729 | /* Note that HASH_SORT assumes the hash handle name to be hh. | |
730 | * HASH_SRT was added to allow the hash handle name to be passed in. */ | |
731 | #define HASH_SORT(head,cmpfcn) HASH_SRT(hh,head,cmpfcn) | |
732 | #define HASH_SRT(hh,head,cmpfcn) \ | |
733 | do { \ | |
734 | uint32_t _hs_i; \ | |
735 | uint32_t _hs_looping,_hs_nmerges,_hs_insize,_hs_psize,_hs_qsize; \ | |
736 | struct UT_hash_handle *_hs_p, *_hs_q, *_hs_e, *_hs_list, *_hs_tail; \ | |
737 | if (head) { \ | |
738 | _hs_insize = 1; \ | |
739 | _hs_looping = 1; \ | |
740 | _hs_list = &((head)->hh); \ | |
741 | while (_hs_looping) { \ | |
742 | _hs_p = _hs_list; \ | |
743 | _hs_list = NULL; \ | |
744 | _hs_tail = NULL; \ | |
745 | _hs_nmerges = 0; \ | |
746 | while (_hs_p) { \ | |
747 | _hs_nmerges++; \ | |
748 | _hs_q = _hs_p; \ | |
749 | _hs_psize = 0; \ | |
750 | for ( _hs_i = 0; _hs_i < _hs_insize; _hs_i++ ) { \ | |
751 | _hs_psize++; \ | |
752 | _hs_q = (UT_hash_handle*)((_hs_q->next) ? \ | |
753 | ((void*)((char*)(_hs_q->next) + \ | |
754 | (head)->hh.tbl->hho)) : NULL); \ | |
755 | if (! (_hs_q) ) break; \ | |
756 | } \ | |
757 | _hs_qsize = _hs_insize; \ | |
758 | while ((_hs_psize > 0) || ((_hs_qsize > 0) && _hs_q )) { \ | |
759 | if (_hs_psize == 0) { \ | |
760 | _hs_e = _hs_q; \ | |
761 | _hs_q = (UT_hash_handle*)((_hs_q->next) ? \ | |
762 | ((void*)((char*)(_hs_q->next) + \ | |
763 | (head)->hh.tbl->hho)) : NULL); \ | |
764 | _hs_qsize--; \ | |
765 | } else if ( (_hs_qsize == 0) || !(_hs_q) ) { \ | |
766 | _hs_e = _hs_p; \ | |
767 | if (_hs_p){ \ | |
768 | _hs_p = (UT_hash_handle*)((_hs_p->next) ? \ | |
769 | ((void*)((char*)(_hs_p->next) + \ | |
770 | (head)->hh.tbl->hho)) : NULL); \ | |
771 | } \ | |
772 | _hs_psize--; \ | |
773 | } else if (( \ | |
774 | cmpfcn(DECLTYPE(head)(ELMT_FROM_HH((head)->hh.tbl,_hs_p)), \ | |
775 | DECLTYPE(head)(ELMT_FROM_HH((head)->hh.tbl,_hs_q))) \ | |
776 | ) <= 0) { \ | |
777 | _hs_e = _hs_p; \ | |
778 | if (_hs_p){ \ | |
779 | _hs_p = (UT_hash_handle*)((_hs_p->next) ? \ | |
780 | ((void*)((char*)(_hs_p->next) + \ | |
781 | (head)->hh.tbl->hho)) : NULL); \ | |
782 | } \ | |
783 | _hs_psize--; \ | |
784 | } else { \ | |
785 | _hs_e = _hs_q; \ | |
786 | _hs_q = (UT_hash_handle*)((_hs_q->next) ? \ | |
787 | ((void*)((char*)(_hs_q->next) + \ | |
788 | (head)->hh.tbl->hho)) : NULL); \ | |
789 | _hs_qsize--; \ | |
790 | } \ | |
791 | if ( _hs_tail ) { \ | |
792 | _hs_tail->next = ((_hs_e) ? \ | |
793 | ELMT_FROM_HH((head)->hh.tbl,_hs_e) : NULL); \ | |
794 | } else { \ | |
795 | _hs_list = _hs_e; \ | |
796 | } \ | |
797 | if (_hs_e) { \ | |
798 | _hs_e->prev = ((_hs_tail) ? \ | |
799 | ELMT_FROM_HH((head)->hh.tbl,_hs_tail) : NULL); \ | |
800 | } \ | |
801 | _hs_tail = _hs_e; \ | |
802 | } \ | |
803 | _hs_p = _hs_q; \ | |
804 | } \ | |
805 | if (_hs_tail){ \ | |
806 | _hs_tail->next = NULL; \ | |
807 | } \ | |
808 | if ( _hs_nmerges <= 1 ) { \ | |
809 | _hs_looping=0; \ | |
810 | (head)->hh.tbl->tail = _hs_tail; \ | |
811 | DECLTYPE_ASSIGN(head,ELMT_FROM_HH((head)->hh.tbl, _hs_list)); \ | |
812 | } \ | |
813 | _hs_insize *= 2; \ | |
814 | } \ | |
815 | HASH_FSCK(hh,head); \ | |
816 | } \ | |
817 | } while (0) | |
818 | ||
819 | /* This function selects items from one hash into another hash. | |
820 | * The end result is that the selected items have dual presence | |
821 | * in both hashes. There is no copy of the items made; rather | |
822 | * they are added into the new hash through a secondary hash | |
823 | * hash handle that must be present in the structure. */ | |
824 | #define HASH_SELECT(hh_dst, dst, hh_src, src, cond) \ | |
825 | do { \ | |
826 | uint32_t _src_bkt, _dst_bkt; \ | |
827 | void *_last_elt=NULL, *_elt; \ | |
828 | UT_hash_handle *_src_hh, *_dst_hh, *_last_elt_hh=NULL; \ | |
829 | ptrdiff_t _dst_hho = ((char*)(&(dst)->hh_dst) - (char*)(dst)); \ | |
830 | if (src) { \ | |
831 | for(_src_bkt=0; _src_bkt < (src)->hh_src.tbl->num_buckets; _src_bkt++) { \ | |
832 | for(_src_hh = (src)->hh_src.tbl->buckets[_src_bkt].hh_head; \ | |
833 | _src_hh; \ | |
834 | _src_hh = _src_hh->hh_next) { \ | |
835 | _elt = ELMT_FROM_HH((src)->hh_src.tbl, _src_hh); \ | |
836 | if (cond(_elt)) { \ | |
837 | _dst_hh = (UT_hash_handle*)(((char*)_elt) + _dst_hho); \ | |
838 | _dst_hh->key = _src_hh->key; \ | |
839 | _dst_hh->keylen = _src_hh->keylen; \ | |
840 | _dst_hh->hashv = _src_hh->hashv; \ | |
841 | _dst_hh->prev = _last_elt; \ | |
842 | _dst_hh->next = NULL; \ | |
843 | if (_last_elt_hh) { _last_elt_hh->next = _elt; } \ | |
844 | if (!dst) { \ | |
845 | DECLTYPE_ASSIGN(dst,_elt); \ | |
846 | HASH_MAKE_TABLE(hh_dst,dst); \ | |
847 | } else { \ | |
848 | _dst_hh->tbl = (dst)->hh_dst.tbl; \ | |
849 | } \ | |
850 | HASH_TO_BKT(_dst_hh->hashv, _dst_hh->tbl->num_buckets, _dst_bkt); \ | |
851 | HASH_ADD_TO_BKT(_dst_hh->tbl->buckets[_dst_bkt],_dst_hh); \ | |
852 | (dst)->hh_dst.tbl->num_items++; \ | |
853 | _last_elt = _elt; \ | |
854 | _last_elt_hh = _dst_hh; \ | |
855 | } \ | |
856 | } \ | |
857 | } \ | |
858 | } \ | |
859 | HASH_FSCK(hh_dst,dst); \ | |
860 | } while (0) | |
861 | ||
862 | #define HASH_CLEAR(hh,head) \ | |
863 | do { \ | |
864 | if (head) { \ | |
865 | uthash_free((head)->hh.tbl->buckets, \ | |
866 | (head)->hh.tbl->num_buckets*sizeof(struct UT_hash_bucket)); \ | |
867 | HASH_BLOOM_FREE((head)->hh.tbl); \ | |
868 | uthash_free((head)->hh.tbl, sizeof(UT_hash_table)); \ | |
869 | (head)=NULL; \ | |
870 | } \ | |
871 | } while(0) | |
872 | ||
873 | #define HASH_OVERHEAD(hh,head) \ | |
874 | (size_t)((((head)->hh.tbl->num_items * sizeof(UT_hash_handle)) + \ | |
875 | ((head)->hh.tbl->num_buckets * sizeof(UT_hash_bucket)) + \ | |
876 | (sizeof(UT_hash_table)) + \ | |
877 | (HASH_BLOOM_BYTELEN))) | |
878 | ||
879 | #ifdef NO_DECLTYPE | |
880 | #define HASH_ITER(hh,head,el,tmp) \ | |
881 | for((el)=(head), (*(char**)(&(tmp)))=(char*)((head)?(head)->hh.next:NULL); \ | |
882 | el; (el)=(tmp),(*(char**)(&(tmp)))=(char*)((tmp)?(tmp)->hh.next:NULL)) | |
883 | #else | |
884 | #define HASH_ITER(hh,head,el,tmp) \ | |
885 | for((el)=(head),(tmp)=DECLTYPE(el)((head)?(head)->hh.next:NULL); \ | |
886 | el; (el)=(tmp),(tmp)=DECLTYPE(el)((tmp)?(tmp)->hh.next:NULL)) | |
887 | #endif | |
888 | ||
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) | |
892 | ||
893 | typedef struct UT_hash_bucket { | |
894 | struct UT_hash_handle *hh_head; | |
895 | uint32_t count; | |
896 | ||
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. | |
908 | */ | |
909 | uint32_t expand_mult; | |
910 | ||
911 | } UT_hash_bucket; | |
912 | ||
913 | /* random signature used only to find hash tables in external analysis */ | |
914 | #define HASH_SIGNATURE 0xa0111fe1 | |
915 | #define HASH_BLOOM_SIGNATURE 0xb12220f2 | |
916 | ||
917 | typedef struct UT_hash_table { | |
918 | UT_hash_bucket *buckets; | |
919 | uint32_t num_buckets, log2_num_buckets; | |
920 | uint32_t num_items; | |
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 */ | |
923 | ||
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 | uint32_t ideal_chain_maxlen; | |
927 | ||
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 | uint32_t nonideal_items; | |
932 | ||
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 | uint32_t ineff_expands, noexpand; | |
940 | ||
941 | uint32_t signature; /* used only to find hash tables in external analysis */ | |
942 | #ifdef HASH_BLOOM | |
943 | uint32_t bloom_sig; /* used only to test bloom exists in external analysis */ | |
944 | uint8_t *bloom_bv; | |
945 | char bloom_nbits; | |
946 | #endif | |
947 | ||
948 | } UT_hash_table; | |
949 | ||
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 | uint32_t hashv; /* result of hash-fcn(key) */ | |
958 | uint8_t keylen; /* enclosing struct's key len */ | |
959 | uint32_t itemind; | |
960 | } UT_hash_handle; | |
961 | ||
962 | #endif /* UTHASH_H */ | |
963 |