Mercurial > urweb
comparison include/urweb/uthash.h @ 2234:2f7ed04332a0
Progress on LRU cache but still more known bugs to fix.
author | Ziv Scully <ziv@mit.edu> |
---|---|
date | Sun, 28 Jun 2015 12:46:51 -0700 |
parents | |
children |
comparison
equal
deleted
inserted
replaced
2233:af1585e7d645 | 2234:2f7ed04332a0 |
---|---|
1 /* | |
2 Copyright (c) 2003-2014, Troy D. Hanson http://troydhanson.github.com/uthash/ | |
3 All rights reserved. | |
4 | |
5 Redistribution and use in source and binary forms, with or without | |
6 modification, are permitted provided that the following conditions are met: | |
7 | |
8 * Redistributions of source code must retain the above copyright | |
9 notice, this list of conditions and the following disclaimer. | |
10 | |
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. | |
22 */ | |
23 | |
24 #ifndef UTHASH_H | |
25 #define UTHASH_H | |
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 #define HASH_INITIAL_NUM_BUCKETS 32U /* initial number of buckets */ | |
97 #define HASH_INITIAL_NUM_BUCKETS_LOG2 5U /* lg2 of initial number of buckets */ | |
98 #define HASH_BKT_CAPACITY_THRESH 10U /* expand when bucket count reaches */ | |
99 | |
100 /* calculate the element whose hash handle address is hhe */ | |
101 #define ELMT_FROM_HH(tbl,hhp) ((void*)(((char*)(hhp)) - ((tbl)->hho))) | |
102 | |
103 #define HASH_FIND(hh,head,keyptr,keylen,out) \ | |
104 do { \ | |
105 out=NULL; \ | |
106 if (head != NULL) { \ | |
107 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) != 0) { \ | |
110 HASH_FIND_IN_BKT((head)->hh.tbl, hh, (head)->hh.tbl->buckets[ _hf_bkt ], \ | |
111 keyptr,keylen,out); \ | |
112 } \ | |
113 } \ | |
114 } while (0) | |
115 | |
116 #ifdef HASH_BLOOM | |
117 #define HASH_BLOOM_BITLEN (1UL << HASH_BLOOM) | |
118 #define HASH_BLOOM_BYTELEN (HASH_BLOOM_BITLEN/8UL) + (((HASH_BLOOM_BITLEN%8UL)!=0UL) ? 1UL : 0UL) | |
119 #define HASH_BLOOM_MAKE(tbl) \ | |
120 do { \ | |
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; \ | |
126 } while (0) | |
127 | |
128 #define HASH_BLOOM_FREE(tbl) \ | |
129 do { \ | |
130 uthash_free((tbl)->bloom_bv, HASH_BLOOM_BYTELEN); \ | |
131 } while (0) | |
132 | |
133 #define HASH_BLOOM_BITSET(bv,idx) (bv[(idx)/8U] |= (1U << ((idx)%8U))) | |
134 #define HASH_BLOOM_BITTEST(bv,idx) (bv[(idx)/8U] & (1U << ((idx)%8U))) | |
135 | |
136 #define HASH_BLOOM_ADD(tbl,hashv) \ | |
137 HASH_BLOOM_BITSET((tbl)->bloom_bv, (hashv & (uint32_t)((1ULL << (tbl)->bloom_nbits) - 1U))) | |
138 | |
139 #define HASH_BLOOM_TEST(tbl,hashv) \ | |
140 HASH_BLOOM_BITTEST((tbl)->bloom_bv, (hashv & (uint32_t)((1ULL << (tbl)->bloom_nbits) - 1U))) | |
141 | |
142 #else | |
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 0U | |
148 #endif | |
149 | |
150 #define HASH_MAKE_TABLE(hh,head) \ | |
151 do { \ | |
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; \ | |
167 } while(0) | |
168 | |
169 #define HASH_ADD(hh,head,fieldname,keylen_in,add) \ | |
170 HASH_ADD_KEYPTR(hh,head,&((add)->fieldname),keylen_in,add) | |
171 | |
172 #define HASH_REPLACE(hh,head,fieldname,keylen_in,add,replaced) \ | |
173 do { \ | |
174 replaced=NULL; \ | |
175 HASH_FIND(hh,head,&((add)->fieldname),keylen_in,replaced); \ | |
176 if (replaced!=NULL) { \ | |
177 HASH_DELETE(hh,head,replaced); \ | |
178 } \ | |
179 HASH_ADD(hh,head,fieldname,keylen_in,add); \ | |
180 } while(0) | |
181 | |
182 #define HASH_ADD_KEYPTR(hh,head,keyptr,keylen_in,add) \ | |
183 do { \ | |
184 unsigned _ha_bkt; \ | |
185 (add)->hh.next = NULL; \ | |
186 (add)->hh.key = (char*)(keyptr); \ | |
187 (add)->hh.keylen = (unsigned)(keylen_in); \ | |
188 if (!(head)) { \ | |
189 head = (add); \ | |
190 (head)->hh.prev = NULL; \ | |
191 HASH_MAKE_TABLE(hh,head); \ | |
192 } else { \ | |
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); \ | |
196 } \ | |
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); \ | |
205 } while(0) | |
206 | |
207 #define HASH_TO_BKT( hashv, num_bkts, bkt ) \ | |
208 do { \ | |
209 bkt = ((hashv) & ((num_bkts) - 1U)); \ | |
210 } while(0) | |
211 | |
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. | |
223 */ | |
224 #define HASH_DELETE(hh,head,delptr) \ | |
225 do { \ | |
226 struct UT_hash_handle *_hd_hh_del; \ | |
227 if ( ((delptr)->hh.prev == NULL) && ((delptr)->hh.next == NULL) ) { \ | |
228 uthash_free((head)->hh.tbl->buckets, \ | |
229 (head)->hh.tbl->num_buckets*sizeof(struct UT_hash_bucket) ); \ | |
230 HASH_BLOOM_FREE((head)->hh.tbl); \ | |
231 uthash_free((head)->hh.tbl, sizeof(UT_hash_table)); \ | |
232 head = NULL; \ | |
233 } else { \ | |
234 unsigned _hd_bkt; \ | |
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); \ | |
240 } \ | |
241 if ((delptr)->hh.prev != NULL) { \ | |
242 ((UT_hash_handle*)((ptrdiff_t)((delptr)->hh.prev) + \ | |
243 (head)->hh.tbl->hho))->next = (delptr)->hh.next; \ | |
244 } else { \ | |
245 DECLTYPE_ASSIGN(head,(delptr)->hh.next); \ | |
246 } \ | |
247 if (_hd_hh_del->next != NULL) { \ | |
248 ((UT_hash_handle*)((ptrdiff_t)_hd_hh_del->next + \ | |
249 (head)->hh.tbl->hho))->prev = \ | |
250 _hd_hh_del->prev; \ | |
251 } \ | |
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--; \ | |
255 } \ | |
256 HASH_FSCK(hh,head); \ | |
257 } while (0) | |
258 | |
259 | |
260 /* convenience forms of HASH_FIND/HASH_ADD/HASH_DEL */ | |
261 #define HASH_FIND_STR(head,findstr,out) \ | |
262 HASH_FIND(hh,head,findstr,(unsigned)strlen(findstr),out) | |
263 #define HASH_ADD_STR(head,strfield,add) \ | |
264 HASH_ADD(hh,head,strfield[0],(unsigned int)strlen(add->strfield),add) | |
265 #define HASH_REPLACE_STR(head,strfield,add,replaced) \ | |
266 HASH_REPLACE(hh,head,strfield[0],(unsigned)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) | |
281 | |
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. | |
284 */ | |
285 #ifdef HASH_DEBUG | |
286 #define HASH_OOPS(...) do { fprintf(stderr,__VA_ARGS__); exit(-1); } while (0) | |
287 #define HASH_FSCK(hh,head) \ | |
288 do { \ | |
289 struct UT_hash_handle *_thh; \ | |
290 if (head) { \ | |
291 unsigned _bkt_i; \ | |
292 unsigned _count; \ | |
293 char *_prev; \ | |
294 _count = 0; \ | |
295 for( _bkt_i = 0; _bkt_i < (head)->hh.tbl->num_buckets; _bkt_i++) { \ | |
296 unsigned _bkt_count = 0; \ | |
297 _thh = (head)->hh.tbl->buckets[_bkt_i].hh_head; \ | |
298 _prev = NULL; \ | |
299 while (_thh) { \ | |
300 if (_prev != (char*)(_thh->hh_prev)) { \ | |
301 HASH_OOPS("invalid hh_prev %p, actual %p\n", \ | |
302 _thh->hh_prev, _prev ); \ | |
303 } \ | |
304 _bkt_count++; \ | |
305 _prev = (char*)(_thh); \ | |
306 _thh = _thh->hh_next; \ | |
307 } \ | |
308 _count += _bkt_count; \ | |
309 if ((head)->hh.tbl->buckets[_bkt_i].count != _bkt_count) { \ | |
310 HASH_OOPS("invalid bucket count %u, actual %u\n", \ | |
311 (head)->hh.tbl->buckets[_bkt_i].count, _bkt_count); \ | |
312 } \ | |
313 } \ | |
314 if (_count != (head)->hh.tbl->num_items) { \ | |
315 HASH_OOPS("invalid hh item count %u, actual %u\n", \ | |
316 (head)->hh.tbl->num_items, _count ); \ | |
317 } \ | |
318 /* traverse hh in app order; check next/prev integrity, count */ \ | |
319 _count = 0; \ | |
320 _prev = NULL; \ | |
321 _thh = &(head)->hh; \ | |
322 while (_thh) { \ | |
323 _count++; \ | |
324 if (_prev !=(char*)(_thh->prev)) { \ | |
325 HASH_OOPS("invalid prev %p, actual %p\n", \ | |
326 _thh->prev, _prev ); \ | |
327 } \ | |
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 ); \ | |
331 } \ | |
332 if (_count != (head)->hh.tbl->num_items) { \ | |
333 HASH_OOPS("invalid app item count %u, actual %u\n", \ | |
334 (head)->hh.tbl->num_items, _count ); \ | |
335 } \ | |
336 } \ | |
337 } while (0) | |
338 #else | |
339 #define HASH_FSCK(hh,head) | |
340 #endif | |
341 | |
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) \ | |
347 do { \ | |
348 unsigned _klen = fieldlen; \ | |
349 write(HASH_EMIT_KEYS, &_klen, sizeof(_klen)); \ | |
350 write(HASH_EMIT_KEYS, keyptr, (unsigned long)fieldlen); \ | |
351 } while (0) | |
352 #else | |
353 #define HASH_EMIT_KEY(hh,head,keyptr,fieldlen) | |
354 #endif | |
355 | |
356 /* default to Jenkin's hash unless overridden e.g. DHASH_FUNCTION=HASH_SAX */ | |
357 #ifdef HASH_FUNCTION | |
358 #define HASH_FCN HASH_FUNCTION | |
359 #else | |
360 #define HASH_FCN HASH_JEN | |
361 #endif | |
362 | |
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) \ | |
365 do { \ | |
366 unsigned _hb_keylen=(unsigned)keylen; \ | |
367 const unsigned char *_hb_key=(const unsigned char*)(key); \ | |
368 (hashv) = 0; \ | |
369 while (_hb_keylen-- != 0U) { \ | |
370 (hashv) = (((hashv) << 5) + (hashv)) + *_hb_key++; \ | |
371 } \ | |
372 bkt = (hashv) & (num_bkts-1U); \ | |
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 unsigned _sx_i; \ | |
381 const unsigned char *_hs_key=(const unsigned 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 } \ | |
386 bkt = hashv & (num_bkts-1U); \ | |
387 } while (0) | |
388 /* FNV-1a variation */ | |
389 #define HASH_FNV(key,keylen,num_bkts,hashv,bkt) \ | |
390 do { \ | |
391 unsigned _fn_i; \ | |
392 const unsigned char *_hf_key=(const unsigned char*)(key); \ | |
393 hashv = 2166136261U; \ | |
394 for(_fn_i=0; _fn_i < keylen; _fn_i++) { \ | |
395 hashv = hashv ^ _hf_key[_fn_i]; \ | |
396 hashv = hashv * 16777619U; \ | |
397 } \ | |
398 bkt = hashv & (num_bkts-1U); \ | |
399 } while(0) | |
400 | |
401 #define HASH_OAT(key,keylen,num_bkts,hashv,bkt) \ | |
402 do { \ | |
403 unsigned _ho_i; \ | |
404 const unsigned char *_ho_key=(const unsigned char*)(key); \ | |
405 hashv = 0; \ | |
406 for(_ho_i=0; _ho_i < keylen; _ho_i++) { \ | |
407 hashv += _ho_key[_ho_i]; \ | |
408 hashv += (hashv << 10); \ | |
409 hashv ^= (hashv >> 6); \ | |
410 } \ | |
411 hashv += (hashv << 3); \ | |
412 hashv ^= (hashv >> 11); \ | |
413 hashv += (hashv << 15); \ | |
414 bkt = hashv & (num_bkts-1U); \ | |
415 } while(0) | |
416 | |
417 #define HASH_JEN_MIX(a,b,c) \ | |
418 do { \ | |
419 a -= b; a -= c; a ^= ( c >> 13 ); \ | |
420 b -= c; b -= a; b ^= ( a << 8 ); \ | |
421 c -= a; c -= b; c ^= ( b >> 13 ); \ | |
422 a -= b; a -= c; a ^= ( c >> 12 ); \ | |
423 b -= c; b -= a; b ^= ( a << 16 ); \ | |
424 c -= a; c -= b; c ^= ( b >> 5 ); \ | |
425 a -= b; a -= c; a ^= ( c >> 3 ); \ | |
426 b -= c; b -= a; b ^= ( a << 10 ); \ | |
427 c -= a; c -= b; c ^= ( b >> 15 ); \ | |
428 } while (0) | |
429 | |
430 #define HASH_JEN(key,keylen,num_bkts,hashv,bkt) \ | |
431 do { \ | |
432 unsigned _hj_i,_hj_j,_hj_k; \ | |
433 unsigned const char *_hj_key=(unsigned const char*)(key); \ | |
434 hashv = 0xfeedbeefu; \ | |
435 _hj_i = _hj_j = 0x9e3779b9u; \ | |
436 _hj_k = (unsigned)(keylen); \ | |
437 while (_hj_k >= 12U) { \ | |
438 _hj_i += (_hj_key[0] + ( (unsigned)_hj_key[1] << 8 ) \ | |
439 + ( (unsigned)_hj_key[2] << 16 ) \ | |
440 + ( (unsigned)_hj_key[3] << 24 ) ); \ | |
441 _hj_j += (_hj_key[4] + ( (unsigned)_hj_key[5] << 8 ) \ | |
442 + ( (unsigned)_hj_key[6] << 16 ) \ | |
443 + ( (unsigned)_hj_key[7] << 24 ) ); \ | |
444 hashv += (_hj_key[8] + ( (unsigned)_hj_key[9] << 8 ) \ | |
445 + ( (unsigned)_hj_key[10] << 16 ) \ | |
446 + ( (unsigned)_hj_key[11] << 24 ) ); \ | |
447 \ | |
448 HASH_JEN_MIX(_hj_i, _hj_j, hashv); \ | |
449 \ | |
450 _hj_key += 12; \ | |
451 _hj_k -= 12U; \ | |
452 } \ | |
453 hashv += (unsigned)(keylen); \ | |
454 switch ( _hj_k ) { \ | |
455 case 11: hashv += ( (unsigned)_hj_key[10] << 24 ); /* FALLTHROUGH */ \ | |
456 case 10: hashv += ( (unsigned)_hj_key[9] << 16 ); /* FALLTHROUGH */ \ | |
457 case 9: hashv += ( (unsigned)_hj_key[8] << 8 ); /* FALLTHROUGH */ \ | |
458 case 8: _hj_j += ( (unsigned)_hj_key[7] << 24 ); /* FALLTHROUGH */ \ | |
459 case 7: _hj_j += ( (unsigned)_hj_key[6] << 16 ); /* FALLTHROUGH */ \ | |
460 case 6: _hj_j += ( (unsigned)_hj_key[5] << 8 ); /* FALLTHROUGH */ \ | |
461 case 5: _hj_j += _hj_key[4]; /* FALLTHROUGH */ \ | |
462 case 4: _hj_i += ( (unsigned)_hj_key[3] << 24 ); /* FALLTHROUGH */ \ | |
463 case 3: _hj_i += ( (unsigned)_hj_key[2] << 16 ); /* FALLTHROUGH */ \ | |
464 case 2: _hj_i += ( (unsigned)_hj_key[1] << 8 ); /* FALLTHROUGH */ \ | |
465 case 1: _hj_i += _hj_key[0]; \ | |
466 } \ | |
467 HASH_JEN_MIX(_hj_i, _hj_j, hashv); \ | |
468 bkt = hashv & (num_bkts-1U); \ | |
469 } while(0) | |
470 | |
471 /* The Paul Hsieh hash function */ | |
472 #undef get16bits | |
473 #if (defined(__GNUC__) && defined(__i386__)) || defined(__WATCOMC__) \ | |
474 || defined(_MSC_VER) || defined (__BORLANDC__) || defined (__TURBOC__) | |
475 #define get16bits(d) (*((const uint16_t *) (d))) | |
476 #endif | |
477 | |
478 #if !defined (get16bits) | |
479 #define get16bits(d) ((((uint32_t)(((const uint8_t *)(d))[1])) << 8) \ | |
480 +(uint32_t)(((const uint8_t *)(d))[0]) ) | |
481 #endif | |
482 #define HASH_SFH(key,keylen,num_bkts,hashv,bkt) \ | |
483 do { \ | |
484 unsigned const char *_sfh_key=(unsigned const char*)(key); \ | |
485 uint32_t _sfh_tmp, _sfh_len = (uint32_t)keylen; \ | |
486 \ | |
487 unsigned _sfh_rem = _sfh_len & 3U; \ | |
488 _sfh_len >>= 2; \ | |
489 hashv = 0xcafebabeu; \ | |
490 \ | |
491 /* Main loop */ \ | |
492 for (;_sfh_len > 0U; _sfh_len--) { \ | |
493 hashv += get16bits (_sfh_key); \ | |
494 _sfh_tmp = ((uint32_t)(get16bits (_sfh_key+2)) << 11) ^ hashv; \ | |
495 hashv = (hashv << 16) ^ _sfh_tmp; \ | |
496 _sfh_key += 2U*sizeof (uint16_t); \ | |
497 hashv += hashv >> 11; \ | |
498 } \ | |
499 \ | |
500 /* Handle end cases */ \ | |
501 switch (_sfh_rem) { \ | |
502 case 3: hashv += get16bits (_sfh_key); \ | |
503 hashv ^= hashv << 16; \ | |
504 hashv ^= (uint32_t)(_sfh_key[sizeof (uint16_t)]) << 18; \ | |
505 hashv += hashv >> 11; \ | |
506 break; \ | |
507 case 2: hashv += get16bits (_sfh_key); \ | |
508 hashv ^= hashv << 11; \ | |
509 hashv += hashv >> 17; \ | |
510 break; \ | |
511 case 1: hashv += *_sfh_key; \ | |
512 hashv ^= hashv << 10; \ | |
513 hashv += hashv >> 1; \ | |
514 } \ | |
515 \ | |
516 /* Force "avalanching" of final 127 bits */ \ | |
517 hashv ^= hashv << 3; \ | |
518 hashv += hashv >> 5; \ | |
519 hashv ^= hashv << 4; \ | |
520 hashv += hashv >> 17; \ | |
521 hashv ^= hashv << 25; \ | |
522 hashv += hashv >> 6; \ | |
523 bkt = hashv & (num_bkts-1U); \ | |
524 } while(0) | |
525 | |
526 #ifdef HASH_USING_NO_STRICT_ALIASING | |
527 /* The MurmurHash exploits some CPU's (x86,x86_64) tolerance for unaligned reads. | |
528 * For other types of CPU's (e.g. Sparc) an unaligned read causes a bus error. | |
529 * MurmurHash uses the faster approach only on CPU's where we know it's safe. | |
530 * | |
531 * Note the preprocessor built-in defines can be emitted using: | |
532 * | |
533 * gcc -m64 -dM -E - < /dev/null (on gcc) | |
534 * cc -## a.c (where a.c is a simple test file) (Sun Studio) | |
535 */ | |
536 #if (defined(__i386__) || defined(__x86_64__) || defined(_M_IX86)) | |
537 #define MUR_GETBLOCK(p,i) p[i] | |
538 #else /* non intel */ | |
539 #define MUR_PLUS0_ALIGNED(p) (((unsigned long)p & 3UL) == 0UL) | |
540 #define MUR_PLUS1_ALIGNED(p) (((unsigned long)p & 3UL) == 1UL) | |
541 #define MUR_PLUS2_ALIGNED(p) (((unsigned long)p & 3UL) == 2UL) | |
542 #define MUR_PLUS3_ALIGNED(p) (((unsigned long)p & 3UL) == 3UL) | |
543 #define WP(p) ((uint32_t*)((unsigned long)(p) & ~3UL)) | |
544 #if (defined(__BIG_ENDIAN__) || defined(SPARC) || defined(__ppc__) || defined(__ppc64__)) | |
545 #define MUR_THREE_ONE(p) ((((*WP(p))&0x00ffffff) << 8) | (((*(WP(p)+1))&0xff000000) >> 24)) | |
546 #define MUR_TWO_TWO(p) ((((*WP(p))&0x0000ffff) <<16) | (((*(WP(p)+1))&0xffff0000) >> 16)) | |
547 #define MUR_ONE_THREE(p) ((((*WP(p))&0x000000ff) <<24) | (((*(WP(p)+1))&0xffffff00) >> 8)) | |
548 #else /* assume little endian non-intel */ | |
549 #define MUR_THREE_ONE(p) ((((*WP(p))&0xffffff00) >> 8) | (((*(WP(p)+1))&0x000000ff) << 24)) | |
550 #define MUR_TWO_TWO(p) ((((*WP(p))&0xffff0000) >>16) | (((*(WP(p)+1))&0x0000ffff) << 16)) | |
551 #define MUR_ONE_THREE(p) ((((*WP(p))&0xff000000) >>24) | (((*(WP(p)+1))&0x00ffffff) << 8)) | |
552 #endif | |
553 #define MUR_GETBLOCK(p,i) (MUR_PLUS0_ALIGNED(p) ? ((p)[i]) : \ | |
554 (MUR_PLUS1_ALIGNED(p) ? MUR_THREE_ONE(p) : \ | |
555 (MUR_PLUS2_ALIGNED(p) ? MUR_TWO_TWO(p) : \ | |
556 MUR_ONE_THREE(p)))) | |
557 #endif | |
558 #define MUR_ROTL32(x,r) (((x) << (r)) | ((x) >> (32 - (r)))) | |
559 #define MUR_FMIX(_h) \ | |
560 do { \ | |
561 _h ^= _h >> 16; \ | |
562 _h *= 0x85ebca6bu; \ | |
563 _h ^= _h >> 13; \ | |
564 _h *= 0xc2b2ae35u; \ | |
565 _h ^= _h >> 16; \ | |
566 } while(0) | |
567 | |
568 #define HASH_MUR(key,keylen,num_bkts,hashv,bkt) \ | |
569 do { \ | |
570 const uint8_t *_mur_data = (const uint8_t*)(key); \ | |
571 const int _mur_nblocks = (int)(keylen) / 4; \ | |
572 uint32_t _mur_h1 = 0xf88D5353u; \ | |
573 uint32_t _mur_c1 = 0xcc9e2d51u; \ | |
574 uint32_t _mur_c2 = 0x1b873593u; \ | |
575 uint32_t _mur_k1 = 0; \ | |
576 const uint8_t *_mur_tail; \ | |
577 const uint32_t *_mur_blocks = (const uint32_t*)(_mur_data+(_mur_nblocks*4)); \ | |
578 int _mur_i; \ | |
579 for(_mur_i = -_mur_nblocks; _mur_i!=0; _mur_i++) { \ | |
580 _mur_k1 = MUR_GETBLOCK(_mur_blocks,_mur_i); \ | |
581 _mur_k1 *= _mur_c1; \ | |
582 _mur_k1 = MUR_ROTL32(_mur_k1,15); \ | |
583 _mur_k1 *= _mur_c2; \ | |
584 \ | |
585 _mur_h1 ^= _mur_k1; \ | |
586 _mur_h1 = MUR_ROTL32(_mur_h1,13); \ | |
587 _mur_h1 = (_mur_h1*5U) + 0xe6546b64u; \ | |
588 } \ | |
589 _mur_tail = (const uint8_t*)(_mur_data + (_mur_nblocks*4)); \ | |
590 _mur_k1=0; \ | |
591 switch((keylen) & 3U) { \ | |
592 case 3: _mur_k1 ^= (uint32_t)_mur_tail[2] << 16; /* FALLTHROUGH */ \ | |
593 case 2: _mur_k1 ^= (uint32_t)_mur_tail[1] << 8; /* FALLTHROUGH */ \ | |
594 case 1: _mur_k1 ^= (uint32_t)_mur_tail[0]; \ | |
595 _mur_k1 *= _mur_c1; \ | |
596 _mur_k1 = MUR_ROTL32(_mur_k1,15); \ | |
597 _mur_k1 *= _mur_c2; \ | |
598 _mur_h1 ^= _mur_k1; \ | |
599 } \ | |
600 _mur_h1 ^= (uint32_t)(keylen); \ | |
601 MUR_FMIX(_mur_h1); \ | |
602 hashv = _mur_h1; \ | |
603 bkt = hashv & (num_bkts-1U); \ | |
604 } while(0) | |
605 #endif /* HASH_USING_NO_STRICT_ALIASING */ | |
606 | |
607 /* key comparison function; return 0 if keys equal */ | |
608 #define HASH_KEYCMP(a,b,len) memcmp(a,b,(unsigned long)(len)) | |
609 | |
610 /* iterate over items in a known bucket to find desired item */ | |
611 #define HASH_FIND_IN_BKT(tbl,hh,head,keyptr,keylen_in,out) \ | |
612 do { \ | |
613 if (head.hh_head != NULL) { DECLTYPE_ASSIGN(out,ELMT_FROM_HH(tbl,head.hh_head)); } \ | |
614 else { out=NULL; } \ | |
615 while (out != NULL) { \ | |
616 if ((out)->hh.keylen == (keylen_in)) { \ | |
617 if ((HASH_KEYCMP((out)->hh.key,keyptr,keylen_in)) == 0) { break; } \ | |
618 } \ | |
619 if ((out)->hh.hh_next != NULL) { DECLTYPE_ASSIGN(out,ELMT_FROM_HH(tbl,(out)->hh.hh_next)); } \ | |
620 else { out = NULL; } \ | |
621 } \ | |
622 } while(0) | |
623 | |
624 /* add an item to a bucket */ | |
625 #define HASH_ADD_TO_BKT(head,addhh) \ | |
626 do { \ | |
627 head.count++; \ | |
628 (addhh)->hh_next = head.hh_head; \ | |
629 (addhh)->hh_prev = NULL; \ | |
630 if (head.hh_head != NULL) { (head).hh_head->hh_prev = (addhh); } \ | |
631 (head).hh_head=addhh; \ | |
632 if ((head.count >= ((head.expand_mult+1U) * HASH_BKT_CAPACITY_THRESH)) \ | |
633 && ((addhh)->tbl->noexpand != 1U)) { \ | |
634 HASH_EXPAND_BUCKETS((addhh)->tbl); \ | |
635 } \ | |
636 } while(0) | |
637 | |
638 /* remove an item from a given bucket */ | |
639 #define HASH_DEL_IN_BKT(hh,head,hh_del) \ | |
640 (head).count--; \ | |
641 if ((head).hh_head == hh_del) { \ | |
642 (head).hh_head = hh_del->hh_next; \ | |
643 } \ | |
644 if (hh_del->hh_prev) { \ | |
645 hh_del->hh_prev->hh_next = hh_del->hh_next; \ | |
646 } \ | |
647 if (hh_del->hh_next) { \ | |
648 hh_del->hh_next->hh_prev = hh_del->hh_prev; \ | |
649 } | |
650 | |
651 /* Bucket expansion has the effect of doubling the number of buckets | |
652 * and redistributing the items into the new buckets. Ideally the | |
653 * items will distribute more or less evenly into the new buckets | |
654 * (the extent to which this is true is a measure of the quality of | |
655 * the hash function as it applies to the key domain). | |
656 * | |
657 * With the items distributed into more buckets, the chain length | |
658 * (item count) in each bucket is reduced. Thus by expanding buckets | |
659 * the hash keeps a bound on the chain length. This bounded chain | |
660 * length is the essence of how a hash provides constant time lookup. | |
661 * | |
662 * The calculation of tbl->ideal_chain_maxlen below deserves some | |
663 * explanation. First, keep in mind that we're calculating the ideal | |
664 * maximum chain length based on the *new* (doubled) bucket count. | |
665 * In fractions this is just n/b (n=number of items,b=new num buckets). | |
666 * Since the ideal chain length is an integer, we want to calculate | |
667 * ceil(n/b). We don't depend on floating point arithmetic in this | |
668 * hash, so to calculate ceil(n/b) with integers we could write | |
669 * | |
670 * ceil(n/b) = (n/b) + ((n%b)?1:0) | |
671 * | |
672 * and in fact a previous version of this hash did just that. | |
673 * But now we have improved things a bit by recognizing that b is | |
674 * always a power of two. We keep its base 2 log handy (call it lb), | |
675 * so now we can write this with a bit shift and logical AND: | |
676 * | |
677 * ceil(n/b) = (n>>lb) + ( (n & (b-1)) ? 1:0) | |
678 * | |
679 */ | |
680 #define HASH_EXPAND_BUCKETS(tbl) \ | |
681 do { \ | |
682 unsigned _he_bkt; \ | |
683 unsigned _he_bkt_i; \ | |
684 struct UT_hash_handle *_he_thh, *_he_hh_nxt; \ | |
685 UT_hash_bucket *_he_new_buckets, *_he_newbkt; \ | |
686 _he_new_buckets = (UT_hash_bucket*)uthash_malloc( \ | |
687 2UL * tbl->num_buckets * sizeof(struct UT_hash_bucket)); \ | |
688 if (!_he_new_buckets) { uthash_fatal( "out of memory"); } \ | |
689 memset(_he_new_buckets, 0, \ | |
690 2UL * tbl->num_buckets * sizeof(struct UT_hash_bucket)); \ | |
691 tbl->ideal_chain_maxlen = \ | |
692 (tbl->num_items >> (tbl->log2_num_buckets+1U)) + \ | |
693 (((tbl->num_items & ((tbl->num_buckets*2U)-1U)) != 0U) ? 1U : 0U); \ | |
694 tbl->nonideal_items = 0; \ | |
695 for(_he_bkt_i = 0; _he_bkt_i < tbl->num_buckets; _he_bkt_i++) \ | |
696 { \ | |
697 _he_thh = tbl->buckets[ _he_bkt_i ].hh_head; \ | |
698 while (_he_thh != NULL) { \ | |
699 _he_hh_nxt = _he_thh->hh_next; \ | |
700 HASH_TO_BKT( _he_thh->hashv, tbl->num_buckets*2U, _he_bkt); \ | |
701 _he_newbkt = &(_he_new_buckets[ _he_bkt ]); \ | |
702 if (++(_he_newbkt->count) > tbl->ideal_chain_maxlen) { \ | |
703 tbl->nonideal_items++; \ | |
704 _he_newbkt->expand_mult = _he_newbkt->count / \ | |
705 tbl->ideal_chain_maxlen; \ | |
706 } \ | |
707 _he_thh->hh_prev = NULL; \ | |
708 _he_thh->hh_next = _he_newbkt->hh_head; \ | |
709 if (_he_newbkt->hh_head != NULL) { _he_newbkt->hh_head->hh_prev = \ | |
710 _he_thh; } \ | |
711 _he_newbkt->hh_head = _he_thh; \ | |
712 _he_thh = _he_hh_nxt; \ | |
713 } \ | |
714 } \ | |
715 uthash_free( tbl->buckets, tbl->num_buckets*sizeof(struct UT_hash_bucket) ); \ | |
716 tbl->num_buckets *= 2U; \ | |
717 tbl->log2_num_buckets++; \ | |
718 tbl->buckets = _he_new_buckets; \ | |
719 tbl->ineff_expands = (tbl->nonideal_items > (tbl->num_items >> 1)) ? \ | |
720 (tbl->ineff_expands+1U) : 0U; \ | |
721 if (tbl->ineff_expands > 1U) { \ | |
722 tbl->noexpand=1; \ | |
723 uthash_noexpand_fyi(tbl); \ | |
724 } \ | |
725 uthash_expand_fyi(tbl); \ | |
726 } while(0) | |
727 | |
728 | |
729 /* This is an adaptation of Simon Tatham's O(n log(n)) mergesort */ | |
730 /* Note that HASH_SORT assumes the hash handle name to be hh. | |
731 * HASH_SRT was added to allow the hash handle name to be passed in. */ | |
732 #define HASH_SORT(head,cmpfcn) HASH_SRT(hh,head,cmpfcn) | |
733 #define HASH_SRT(hh,head,cmpfcn) \ | |
734 do { \ | |
735 unsigned _hs_i; \ | |
736 unsigned _hs_looping,_hs_nmerges,_hs_insize,_hs_psize,_hs_qsize; \ | |
737 struct UT_hash_handle *_hs_p, *_hs_q, *_hs_e, *_hs_list, *_hs_tail; \ | |
738 if (head != NULL) { \ | |
739 _hs_insize = 1; \ | |
740 _hs_looping = 1; \ | |
741 _hs_list = &((head)->hh); \ | |
742 while (_hs_looping != 0U) { \ | |
743 _hs_p = _hs_list; \ | |
744 _hs_list = NULL; \ | |
745 _hs_tail = NULL; \ | |
746 _hs_nmerges = 0; \ | |
747 while (_hs_p != NULL) { \ | |
748 _hs_nmerges++; \ | |
749 _hs_q = _hs_p; \ | |
750 _hs_psize = 0; \ | |
751 for ( _hs_i = 0; _hs_i < _hs_insize; _hs_i++ ) { \ | |
752 _hs_psize++; \ | |
753 _hs_q = (UT_hash_handle*)((_hs_q->next != NULL) ? \ | |
754 ((void*)((char*)(_hs_q->next) + \ | |
755 (head)->hh.tbl->hho)) : NULL); \ | |
756 if (! (_hs_q) ) { break; } \ | |
757 } \ | |
758 _hs_qsize = _hs_insize; \ | |
759 while ((_hs_psize > 0U) || ((_hs_qsize > 0U) && (_hs_q != NULL))) {\ | |
760 if (_hs_psize == 0U) { \ | |
761 _hs_e = _hs_q; \ | |
762 _hs_q = (UT_hash_handle*)((_hs_q->next != NULL) ? \ | |
763 ((void*)((char*)(_hs_q->next) + \ | |
764 (head)->hh.tbl->hho)) : NULL); \ | |
765 _hs_qsize--; \ | |
766 } else if ( (_hs_qsize == 0U) || (_hs_q == NULL) ) { \ | |
767 _hs_e = _hs_p; \ | |
768 if (_hs_p != NULL){ \ | |
769 _hs_p = (UT_hash_handle*)((_hs_p->next != NULL) ? \ | |
770 ((void*)((char*)(_hs_p->next) + \ | |
771 (head)->hh.tbl->hho)) : NULL); \ | |
772 } \ | |
773 _hs_psize--; \ | |
774 } else if (( \ | |
775 cmpfcn(DECLTYPE(head)(ELMT_FROM_HH((head)->hh.tbl,_hs_p)), \ | |
776 DECLTYPE(head)(ELMT_FROM_HH((head)->hh.tbl,_hs_q))) \ | |
777 ) <= 0) { \ | |
778 _hs_e = _hs_p; \ | |
779 if (_hs_p != NULL){ \ | |
780 _hs_p = (UT_hash_handle*)((_hs_p->next != NULL) ? \ | |
781 ((void*)((char*)(_hs_p->next) + \ | |
782 (head)->hh.tbl->hho)) : NULL); \ | |
783 } \ | |
784 _hs_psize--; \ | |
785 } else { \ | |
786 _hs_e = _hs_q; \ | |
787 _hs_q = (UT_hash_handle*)((_hs_q->next != NULL) ? \ | |
788 ((void*)((char*)(_hs_q->next) + \ | |
789 (head)->hh.tbl->hho)) : NULL); \ | |
790 _hs_qsize--; \ | |
791 } \ | |
792 if ( _hs_tail != NULL ) { \ | |
793 _hs_tail->next = ((_hs_e != NULL) ? \ | |
794 ELMT_FROM_HH((head)->hh.tbl,_hs_e) : NULL); \ | |
795 } else { \ | |
796 _hs_list = _hs_e; \ | |
797 } \ | |
798 if (_hs_e != NULL) { \ | |
799 _hs_e->prev = ((_hs_tail != NULL) ? \ | |
800 ELMT_FROM_HH((head)->hh.tbl,_hs_tail) : NULL); \ | |
801 } \ | |
802 _hs_tail = _hs_e; \ | |
803 } \ | |
804 _hs_p = _hs_q; \ | |
805 } \ | |
806 if (_hs_tail != NULL){ \ | |
807 _hs_tail->next = NULL; \ | |
808 } \ | |
809 if ( _hs_nmerges <= 1U ) { \ | |
810 _hs_looping=0; \ | |
811 (head)->hh.tbl->tail = _hs_tail; \ | |
812 DECLTYPE_ASSIGN(head,ELMT_FROM_HH((head)->hh.tbl, _hs_list)); \ | |
813 } \ | |
814 _hs_insize *= 2U; \ | |
815 } \ | |
816 HASH_FSCK(hh,head); \ | |
817 } \ | |
818 } while (0) | |
819 | |
820 /* This function selects items from one hash into another hash. | |
821 * The end result is that the selected items have dual presence | |
822 * in both hashes. There is no copy of the items made; rather | |
823 * they are added into the new hash through a secondary hash | |
824 * hash handle that must be present in the structure. */ | |
825 #define HASH_SELECT(hh_dst, dst, hh_src, src, cond) \ | |
826 do { \ | |
827 unsigned _src_bkt, _dst_bkt; \ | |
828 void *_last_elt=NULL, *_elt; \ | |
829 UT_hash_handle *_src_hh, *_dst_hh, *_last_elt_hh=NULL; \ | |
830 ptrdiff_t _dst_hho = ((char*)(&(dst)->hh_dst) - (char*)(dst)); \ | |
831 if (src != NULL) { \ | |
832 for(_src_bkt=0; _src_bkt < (src)->hh_src.tbl->num_buckets; _src_bkt++) { \ | |
833 for(_src_hh = (src)->hh_src.tbl->buckets[_src_bkt].hh_head; \ | |
834 _src_hh != NULL; \ | |
835 _src_hh = _src_hh->hh_next) { \ | |
836 _elt = ELMT_FROM_HH((src)->hh_src.tbl, _src_hh); \ | |
837 if (cond(_elt)) { \ | |
838 _dst_hh = (UT_hash_handle*)(((char*)_elt) + _dst_hho); \ | |
839 _dst_hh->key = _src_hh->key; \ | |
840 _dst_hh->keylen = _src_hh->keylen; \ | |
841 _dst_hh->hashv = _src_hh->hashv; \ | |
842 _dst_hh->prev = _last_elt; \ | |
843 _dst_hh->next = NULL; \ | |
844 if (_last_elt_hh != NULL) { _last_elt_hh->next = _elt; } \ | |
845 if (dst == NULL) { \ | |
846 DECLTYPE_ASSIGN(dst,_elt); \ | |
847 HASH_MAKE_TABLE(hh_dst,dst); \ | |
848 } else { \ | |
849 _dst_hh->tbl = (dst)->hh_dst.tbl; \ | |
850 } \ | |
851 HASH_TO_BKT(_dst_hh->hashv, _dst_hh->tbl->num_buckets, _dst_bkt); \ | |
852 HASH_ADD_TO_BKT(_dst_hh->tbl->buckets[_dst_bkt],_dst_hh); \ | |
853 (dst)->hh_dst.tbl->num_items++; \ | |
854 _last_elt = _elt; \ | |
855 _last_elt_hh = _dst_hh; \ | |
856 } \ | |
857 } \ | |
858 } \ | |
859 } \ | |
860 HASH_FSCK(hh_dst,dst); \ | |
861 } while (0) | |
862 | |
863 #define HASH_CLEAR(hh,head) \ | |
864 do { \ | |
865 if (head != NULL) { \ | |
866 uthash_free((head)->hh.tbl->buckets, \ | |
867 (head)->hh.tbl->num_buckets*sizeof(struct UT_hash_bucket)); \ | |
868 HASH_BLOOM_FREE((head)->hh.tbl); \ | |
869 uthash_free((head)->hh.tbl, sizeof(UT_hash_table)); \ | |
870 (head)=NULL; \ | |
871 } \ | |
872 } while(0) | |
873 | |
874 #define HASH_OVERHEAD(hh,head) \ | |
875 ((head != NULL) ? ( \ | |
876 (size_t)(((head)->hh.tbl->num_items * sizeof(UT_hash_handle)) + \ | |
877 ((head)->hh.tbl->num_buckets * sizeof(UT_hash_bucket)) + \ | |
878 sizeof(UT_hash_table) + \ | |
879 (HASH_BLOOM_BYTELEN))) : 0U) | |
880 | |
881 #ifdef NO_DECLTYPE | |
882 #define HASH_ITER(hh,head,el,tmp) \ | |
883 for(((el)=(head)), ((*(char**)(&(tmp)))=(char*)((head!=NULL)?(head)->hh.next:NULL)); \ | |
884 (el) != NULL; ((el)=(tmp)), ((*(char**)(&(tmp)))=(char*)((tmp!=NULL)?(tmp)->hh.next:NULL))) | |
885 #else | |
886 #define HASH_ITER(hh,head,el,tmp) \ | |
887 for(((el)=(head)), ((tmp)=DECLTYPE(el)((head!=NULL)?(head)->hh.next:NULL)); \ | |
888 (el) != NULL; ((el)=(tmp)), ((tmp)=DECLTYPE(el)((tmp!=NULL)?(tmp)->hh.next:NULL))) | |
889 #endif | |
890 | |
891 /* obtain a count of items in the hash */ | |
892 #define HASH_COUNT(head) HASH_CNT(hh,head) | |
893 #define HASH_CNT(hh,head) ((head != NULL)?((head)->hh.tbl->num_items):0U) | |
894 | |
895 typedef struct UT_hash_bucket { | |
896 struct UT_hash_handle *hh_head; | |
897 unsigned count; | |
898 | |
899 /* expand_mult is normally set to 0. In this situation, the max chain length | |
900 * threshold is enforced at its default value, HASH_BKT_CAPACITY_THRESH. (If | |
901 * the bucket's chain exceeds this length, bucket expansion is triggered). | |
902 * However, setting expand_mult to a non-zero value delays bucket expansion | |
903 * (that would be triggered by additions to this particular bucket) | |
904 * until its chain length reaches a *multiple* of HASH_BKT_CAPACITY_THRESH. | |
905 * (The multiplier is simply expand_mult+1). The whole idea of this | |
906 * multiplier is to reduce bucket expansions, since they are expensive, in | |
907 * situations where we know that a particular bucket tends to be overused. | |
908 * It is better to let its chain length grow to a longer yet-still-bounded | |
909 * value, than to do an O(n) bucket expansion too often. | |
910 */ | |
911 unsigned expand_mult; | |
912 | |
913 } UT_hash_bucket; | |
914 | |
915 /* random signature used only to find hash tables in external analysis */ | |
916 #define HASH_SIGNATURE 0xa0111fe1u | |
917 #define HASH_BLOOM_SIGNATURE 0xb12220f2u | |
918 | |
919 typedef struct UT_hash_table { | |
920 UT_hash_bucket *buckets; | |
921 unsigned num_buckets, log2_num_buckets; | |
922 unsigned num_items; | |
923 struct UT_hash_handle *tail; /* tail hh in app order, for fast append */ | |
924 ptrdiff_t hho; /* hash handle offset (byte pos of hash handle in element */ | |
925 | |
926 /* in an ideal situation (all buckets used equally), no bucket would have | |
927 * more than ceil(#items/#buckets) items. that's the ideal chain length. */ | |
928 unsigned ideal_chain_maxlen; | |
929 | |
930 /* nonideal_items is the number of items in the hash whose chain position | |
931 * exceeds the ideal chain maxlen. these items pay the penalty for an uneven | |
932 * hash distribution; reaching them in a chain traversal takes >ideal steps */ | |
933 unsigned nonideal_items; | |
934 | |
935 /* ineffective expands occur when a bucket doubling was performed, but | |
936 * afterward, more than half the items in the hash had nonideal chain | |
937 * positions. If this happens on two consecutive expansions we inhibit any | |
938 * further expansion, as it's not helping; this happens when the hash | |
939 * function isn't a good fit for the key domain. When expansion is inhibited | |
940 * the hash will still work, albeit no longer in constant time. */ | |
941 unsigned ineff_expands, noexpand; | |
942 | |
943 uint32_t signature; /* used only to find hash tables in external analysis */ | |
944 #ifdef HASH_BLOOM | |
945 uint32_t bloom_sig; /* used only to test bloom exists in external analysis */ | |
946 uint8_t *bloom_bv; | |
947 uint8_t bloom_nbits; | |
948 #endif | |
949 | |
950 } UT_hash_table; | |
951 | |
952 typedef struct UT_hash_handle { | |
953 struct UT_hash_table *tbl; | |
954 void *prev; /* prev element in app order */ | |
955 void *next; /* next element in app order */ | |
956 struct UT_hash_handle *hh_prev; /* previous hh in bucket order */ | |
957 struct UT_hash_handle *hh_next; /* next hh in bucket order */ | |
958 void *key; /* ptr to enclosing struct's key */ | |
959 unsigned keylen; /* enclosing struct's key len */ | |
960 unsigned hashv; /* result of hash-fcn(key) */ | |
961 } UT_hash_handle; | |
962 | |
963 #endif /* UTHASH_H */ |