Chameleon

Chameleon Svn Source Tree

Root/branches/cparm/i386/libsaio/uthash.h

1/*
2 Copyright (c) 2003-2013, Troy D. Hanson http://uthash.sourceforge.net
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 "libsaio.h"
28
29/* These macros use decltype or the earlier __typeof GNU extension.
30 As decltype is only available in newer compilers (VS2010 or gcc 4.3+
31 when compiling c++ source) this code uses whatever method is needed
32 or, for VS2008 where neither is available, uses casting workarounds. */
33#ifdef _MSC_VER /* MS compiler */
34#if _MSC_VER >= 1600 && defined(__cplusplus) /* VS2010 or newer in C++ mode */
35#define DECLTYPE(x) (decltype(x))
36#else /* VS2008 or older (or VS2010 in C mode) */
37#define NO_DECLTYPE
38#define DECLTYPE(x)
39#endif
40#else /* GNU, Sun and other compilers */
41#define DECLTYPE(x) (__typeof(x))
42#endif
43
44#ifdef NO_DECLTYPE
45#define DECLTYPE_ASSIGN(dst,src) \
46do { \
47char **_da_dst = (char**)(&(dst)); \
48*_da_dst = (char*)(src); \
49} while(0)
50#else
51#define DECLTYPE_ASSIGN(dst,src) \
52do { \
53(dst) = DECLTYPE(dst)(src); \
54} while(0)
55#endif
56
57/* a number of the hash function use uint32_t which isn't defined on win32 */
58#ifdef _MSC_VER
59typedef unsigned int uint32_t;
60typedef unsigned char uint8_t;
61#else
62#include <inttypes.h> /* uint32_t */
63#endif
64
65#define UTHASH_VERSION 1.9.7
66
67#ifndef uthash_fatal
68#define uthash_fatal(msg) Throw(-1) /* fatal error (out of memory,etc) */
69#endif
70#ifndef uthash_malloc
71#define uthash_malloc(sz) malloc(sz) /* malloc fcn */
72#endif
73#ifndef uthash_free
74#define uthash_free(ptr,sz) free(ptr) /* free fcn */
75#endif
76
77#ifndef uthash_noexpand_fyi
78#define uthash_noexpand_fyi(tbl) /* can be defined to log noexpand */
79#endif
80#ifndef uthash_expand_fyi
81#define uthash_expand_fyi(tbl) /* can be defined to log expands */
82#endif
83
84/* initial number of buckets */
85#define HASH_INITIAL_NUM_BUCKETS 32 /* initial number of buckets */
86#define HASH_INITIAL_NUM_BUCKETS_LOG2 5 /* lg2 of initial number of buckets */
87#define HASH_BKT_CAPACITY_THRESH 10 /* expand when bucket count reaches */
88
89/* calculate the element whose hash handle address is hhe */
90#define ELMT_FROM_HH(tbl,hhp) ((void*)(((char*)(hhp)) - ((tbl)->hho)))
91
92#define HASH_FIND(hh,head,keyptr,keylen,out) \
93do { \
94unsigned _hf_bkt,_hf_hashv; \
95out=NULL; \
96if (head) { \
97HASH_FCN(keyptr,keylen, (head)->hh.tbl->num_buckets, _hf_hashv, _hf_bkt); \
98if (HASH_BLOOM_TEST((head)->hh.tbl, _hf_hashv)) { \
99HASH_FIND_IN_BKT((head)->hh.tbl, hh, (head)->hh.tbl->buckets[ _hf_bkt ], \
100keyptr,keylen,out); \
101} \
102} \
103} while (0)
104
105#ifdef HASH_BLOOM
106#define HASH_BLOOM_BITLEN (1ULL << HASH_BLOOM)
107#define HASH_BLOOM_BYTELEN (HASH_BLOOM_BITLEN/8) + ((HASH_BLOOM_BITLEN%8) ? 1:0)
108#define HASH_BLOOM_MAKE(tbl) \
109do { \
110(tbl)->bloom_nbits = HASH_BLOOM; \
111(tbl)->bloom_bv = (uint8_t*)uthash_malloc(HASH_BLOOM_BYTELEN); \
112if (!((tbl)->bloom_bv)) { uthash_fatal( "out of memory"); } \
113memset((tbl)->bloom_bv, 0, HASH_BLOOM_BYTELEN); \
114(tbl)->bloom_sig = HASH_BLOOM_SIGNATURE; \
115} while (0)
116
117#define HASH_BLOOM_FREE(tbl) \
118do { \
119uthash_free((tbl)->bloom_bv, HASH_BLOOM_BYTELEN); \
120} while (0)
121
122#define HASH_BLOOM_BITSET(bv,idx) (bv[(idx)/8] |= (1U << ((idx)%8)))
123#define HASH_BLOOM_BITTEST(bv,idx) (bv[(idx)/8] & (1U << ((idx)%8)))
124
125#define HASH_BLOOM_ADD(tbl,hashv) \
126HASH_BLOOM_BITSET((tbl)->bloom_bv, (hashv & (uint32_t)((1ULL << (tbl)->bloom_nbits) - 1)))
127
128#define HASH_BLOOM_TEST(tbl,hashv) \
129HASH_BLOOM_BITTEST((tbl)->bloom_bv, (hashv & (uint32_t)((1ULL << (tbl)->bloom_nbits) - 1)))
130
131#else
132#define HASH_BLOOM_MAKE(tbl)
133#define HASH_BLOOM_FREE(tbl)
134#define HASH_BLOOM_ADD(tbl,hashv)
135#define HASH_BLOOM_TEST(tbl,hashv) (1)
136#endif
137
138#define HASH_MAKE_TABLE(hh,head) \
139do { \
140(head)->hh.tbl = (UT_hash_table*)uthash_malloc( \
141sizeof(UT_hash_table)); \
142if (!((head)->hh.tbl)) { uthash_fatal( "out of memory"); } \
143memset((head)->hh.tbl, 0, sizeof(UT_hash_table)); \
144(head)->hh.tbl->tail = &((head)->hh); \
145(head)->hh.tbl->num_buckets = HASH_INITIAL_NUM_BUCKETS; \
146(head)->hh.tbl->log2_num_buckets = HASH_INITIAL_NUM_BUCKETS_LOG2; \
147(head)->hh.tbl->hho = (char*)(&(head)->hh) - (char*)(head); \
148(head)->hh.tbl->buckets = (UT_hash_bucket*)uthash_malloc( \
149HASH_INITIAL_NUM_BUCKETS*sizeof(struct UT_hash_bucket)); \
150if (! (head)->hh.tbl->buckets) { uthash_fatal( "out of memory"); } \
151memset((head)->hh.tbl->buckets, 0, \
152HASH_INITIAL_NUM_BUCKETS*sizeof(struct UT_hash_bucket)); \
153HASH_BLOOM_MAKE((head)->hh.tbl); \
154(head)->hh.tbl->signature = HASH_SIGNATURE; \
155} while(0)
156
157#define HASH_ADD(hh,head,fieldname,keylen_in,add) \
158HASH_ADD_KEYPTR(hh,head,&((add)->fieldname),keylen_in,add)
159
160#define HASH_ADD_KEYPTR(hh,head,keyptr,keylen_in,add) \
161do { \
162unsigned _ha_bkt; \
163(add)->hh.next = NULL; \
164(add)->hh.key = (char*)keyptr; \
165(add)->hh.keylen = (unsigned)keylen_in; \
166if (!(head)) { \
167head = (add); \
168(head)->hh.prev = NULL; \
169HASH_MAKE_TABLE(hh,head); \
170} else { \
171(head)->hh.tbl->tail->next = (add); \
172(add)->hh.prev = ELMT_FROM_HH((head)->hh.tbl, (head)->hh.tbl->tail); \
173(head)->hh.tbl->tail = &((add)->hh); \
174} \
175(head)->hh.tbl->num_items++; \
176(add)->hh.tbl = (head)->hh.tbl; \
177HASH_FCN(keyptr,keylen_in, (head)->hh.tbl->num_buckets, \
178(add)->hh.hashv, _ha_bkt); \
179HASH_ADD_TO_BKT((head)->hh.tbl->buckets[_ha_bkt],&(add)->hh); \
180HASH_BLOOM_ADD((head)->hh.tbl,(add)->hh.hashv); \
181HASH_EMIT_KEY(hh,head,keyptr,keylen_in); \
182HASH_FSCK(hh,head); \
183} while(0)
184
185#define HASH_TO_BKT( hashv, num_bkts, bkt ) \
186do { \
187bkt = ((hashv) & ((num_bkts) - 1)); \
188} while(0)
189
190/* delete "delptr" from the hash table.
191 * "the usual" patch-up process for the app-order doubly-linked-list.
192 * The use of _hd_hh_del below deserves special explanation.
193 * These used to be expressed using (delptr) but that led to a bug
194 * if someone used the same symbol for the head and deletee, like
195 * HASH_DELETE(hh,users,users);
196 * We want that to work, but by changing the head (users) below
197 * we were forfeiting our ability to further refer to the deletee (users)
198 * in the patch-up process. Solution: use scratch space to
199 * copy the deletee pointer, then the latter references are via that
200 * scratch pointer rather than through the repointed (users) symbol.
201 */
202#define HASH_DELETE(hh,head,delptr) \
203do { \
204unsigned _hd_bkt; \
205struct UT_hash_handle *_hd_hh_del; \
206if ( ((delptr)->hh.prev == NULL) && ((delptr)->hh.next == NULL) ) { \
207uthash_free((head)->hh.tbl->buckets, \
208(head)->hh.tbl->num_buckets*sizeof(struct UT_hash_bucket) ); \
209HASH_BLOOM_FREE((head)->hh.tbl); \
210uthash_free((head)->hh.tbl, sizeof(UT_hash_table)); \
211head = NULL; \
212} else { \
213_hd_hh_del = &((delptr)->hh); \
214if ((delptr) == ELMT_FROM_HH((head)->hh.tbl,(head)->hh.tbl->tail)) { \
215(head)->hh.tbl->tail = \
216(UT_hash_handle*)((ptrdiff_t)((delptr)->hh.prev) + \
217(head)->hh.tbl->hho); \
218} \
219if ((delptr)->hh.prev) { \
220((UT_hash_handle*)((ptrdiff_t)((delptr)->hh.prev) + \
221(head)->hh.tbl->hho))->next = (delptr)->hh.next; \
222} else { \
223DECLTYPE_ASSIGN(head,(delptr)->hh.next); \
224} \
225if (_hd_hh_del->next) { \
226((UT_hash_handle*)((ptrdiff_t)_hd_hh_del->next + \
227(head)->hh.tbl->hho))->prev = \
228_hd_hh_del->prev; \
229} \
230HASH_TO_BKT( _hd_hh_del->hashv, (head)->hh.tbl->num_buckets, _hd_bkt); \
231HASH_DEL_IN_BKT(hh,(head)->hh.tbl->buckets[_hd_bkt], _hd_hh_del); \
232(head)->hh.tbl->num_items--; \
233} \
234HASH_FSCK(hh,head); \
235} while (0)
236
237
238/* convenience forms of HASH_FIND/HASH_ADD/HASH_DEL */
239#define HASH_FIND_STR(head,findstr,out) \
240HASH_FIND(hh,head,findstr,strlen(findstr),out)
241#define HASH_ADD_STR(head,strfield,add) \
242HASH_ADD(hh,head,strfield,strlen(add->strfield),add)
243#define HASH_FIND_INT(head,findint,out) \
244HASH_FIND(hh,head,findint,sizeof(int),out)
245#define HASH_ADD_INT(head,intfield,add) \
246HASH_ADD(hh,head,intfield,sizeof(int),add)
247#define HASH_FIND_PTR(head,findptr,out) \
248HASH_FIND(hh,head,findptr,sizeof(void *),out)
249#define HASH_ADD_PTR(head,ptrfield,add) \
250HASH_ADD(hh,head,ptrfield,sizeof(void *),add)
251#define HASH_DEL(head,delptr) \
252HASH_DELETE(hh,head,delptr)
253
254/* HASH_FSCK checks hash integrity on every add/delete when HASH_DEBUG is defined.
255 * This is for uthash developer only; it compiles away if HASH_DEBUG isn't defined.
256 */
257#ifdef HASH_DEBUG
258#define HASH_OOPS(...) do { printf(__VA_ARGS__); exit(-1); } while (0)
259#define HASH_FSCK(hh,head) \
260do { \
261unsigned _bkt_i; \
262unsigned _count, _bkt_count; \
263char *_prev; \
264struct UT_hash_handle *_thh; \
265if (head) { \
266_count = 0; \
267for( _bkt_i = 0; _bkt_i < (head)->hh.tbl->num_buckets; _bkt_i++) { \
268_bkt_count = 0; \
269_thh = (head)->hh.tbl->buckets[_bkt_i].hh_head; \
270_prev = NULL; \
271while (_thh) { \
272if (_prev != (char*)(_thh->hh_prev)) { \
273HASH_OOPS("invalid hh_prev %p, actual %p\n", \
274_thh->hh_prev, _prev ); \
275} \
276_bkt_count++; \
277_prev = (char*)(_thh); \
278_thh = _thh->hh_next; \
279} \
280_count += _bkt_count; \
281if ((head)->hh.tbl->buckets[_bkt_i].count != _bkt_count) { \
282HASH_OOPS("invalid bucket count %d, actual %d\n", \
283(head)->hh.tbl->buckets[_bkt_i].count, _bkt_count); \
284} \
285} \
286if (_count != (head)->hh.tbl->num_items) { \
287HASH_OOPS("invalid hh item count %d, actual %d\n", \
288(head)->hh.tbl->num_items, _count ); \
289} \
290/* traverse hh in app order; check next/prev integrity, count */ \
291_count = 0; \
292_prev = NULL; \
293_thh = &(head)->hh; \
294while (_thh) { \
295_count++; \
296if (_prev !=(char*)(_thh->prev)) { \
297HASH_OOPS("invalid prev %p, actual %p\n", \
298_thh->prev, _prev ); \
299} \
300_prev = (char*)ELMT_FROM_HH((head)->hh.tbl, _thh); \
301_thh = ( _thh->next ? (UT_hash_handle*)((char*)(_thh->next) + \
302(head)->hh.tbl->hho) : NULL ); \
303} \
304if (_count != (head)->hh.tbl->num_items) { \
305HASH_OOPS("invalid app item count %d, actual %d\n", \
306(head)->hh.tbl->num_items, _count ); \
307} \
308} \
309} while (0)
310#else
311#define HASH_FSCK(hh,head)
312#endif
313
314/* When compiled with -DHASH_EMIT_KEYS, length-prefixed keys are emitted to
315 * the descriptor to which this macro is defined for tuning the hash function.
316 * The app can #include <unistd.h> to get the prototype for write(2). */
317#ifdef HASH_EMIT_KEYS
318#define HASH_EMIT_KEY(hh,head,keyptr,fieldlen) \
319do { \
320unsigned _klen = fieldlen; \
321write(HASH_EMIT_KEYS, &_klen, sizeof(_klen)); \
322write(HASH_EMIT_KEYS, keyptr, fieldlen); \
323} while (0)
324#else
325#define HASH_EMIT_KEY(hh,head,keyptr,fieldlen)
326#endif
327
328/* default to Jenkin's hash unless overridden e.g. DHASH_FUNCTION=HASH_SAX */
329#ifdef HASH_FUNCTION
330#define HASH_FCN HASH_FUNCTION
331#else
332#define HASH_FCN HASH_JEN
333#endif
334
335/* The Bernstein hash function, used in Perl prior to v5.6 */
336#define HASH_BER(key,keylen,num_bkts,hashv,bkt) \
337do { \
338unsigned _hb_keylen=keylen; \
339char *_hb_key=(char*)(key); \
340(hashv) = 0; \
341while (_hb_keylen--) { (hashv) = ((hashv) * 33) + *_hb_key++; } \
342bkt = (hashv) & (num_bkts-1); \
343} while (0)
344
345
346/* SAX/FNV/OAT/JEN hash functions are macro variants of those listed at
347 * http://eternallyconfuzzled.com/tuts/algorithms/jsw_tut_hashing.aspx */
348#define HASH_SAX(key,keylen,num_bkts,hashv,bkt) \
349do { \
350unsigned _sx_i; \
351char *_hs_key=(char*)(key); \
352hashv = 0; \
353for(_sx_i=0; _sx_i < keylen; _sx_i++) \
354hashv ^= (hashv << 5) + (hashv >> 2) + _hs_key[_sx_i]; \
355bkt = hashv & (num_bkts-1); \
356} while (0)
357
358#define HASH_FNV(key,keylen,num_bkts,hashv,bkt) \
359do { \
360unsigned _fn_i; \
361char *_hf_key=(char*)(key); \
362hashv = 2166136261UL; \
363for(_fn_i=0; _fn_i < keylen; _fn_i++) \
364hashv = (hashv * 16777619) ^ _hf_key[_fn_i]; \
365bkt = hashv & (num_bkts-1); \
366} while(0)
367
368#define HASH_OAT(key,keylen,num_bkts,hashv,bkt) \
369do { \
370unsigned _ho_i; \
371char *_ho_key=(char*)(key); \
372hashv = 0; \
373for(_ho_i=0; _ho_i < keylen; _ho_i++) { \
374hashv += _ho_key[_ho_i]; \
375hashv += (hashv << 10); \
376hashv ^= (hashv >> 6); \
377} \
378hashv += (hashv << 3); \
379hashv ^= (hashv >> 11); \
380hashv += (hashv << 15); \
381bkt = hashv & (num_bkts-1); \
382} while(0)
383
384#define HASH_JEN_MIX(a,b,c) \
385do { \
386a -= b; a -= c; a ^= ( c >> 13 ); \
387b -= c; b -= a; b ^= ( a << 8 ); \
388c -= a; c -= b; c ^= ( b >> 13 ); \
389a -= b; a -= c; a ^= ( c >> 12 ); \
390b -= c; b -= a; b ^= ( a << 16 ); \
391c -= a; c -= b; c ^= ( b >> 5 ); \
392a -= b; a -= c; a ^= ( c >> 3 ); \
393b -= c; b -= a; b ^= ( a << 10 ); \
394c -= a; c -= b; c ^= ( b >> 15 ); \
395} while (0)
396
397#define HASH_JEN(key,keylen,num_bkts,hashv,bkt) \
398do { \
399unsigned _hj_i,_hj_j,_hj_k; \
400char *_hj_key=(char*)(key); \
401hashv = 0xfeedbeef; \
402_hj_i = _hj_j = 0x9e3779b9; \
403_hj_k = (unsigned)keylen; \
404while (_hj_k >= 12) { \
405_hj_i += (_hj_key[0] + ( (unsigned)_hj_key[1] << 8 ) \
406+ ( (unsigned)_hj_key[2] << 16 ) \
407+ ( (unsigned)_hj_key[3] << 24 ) ); \
408_hj_j += (_hj_key[4] + ( (unsigned)_hj_key[5] << 8 ) \
409+ ( (unsigned)_hj_key[6] << 16 ) \
410+ ( (unsigned)_hj_key[7] << 24 ) ); \
411hashv += (_hj_key[8] + ( (unsigned)_hj_key[9] << 8 ) \
412+ ( (unsigned)_hj_key[10] << 16 ) \
413+ ( (unsigned)_hj_key[11] << 24 ) ); \
414\
415HASH_JEN_MIX(_hj_i, _hj_j, hashv); \
416\
417_hj_key += 12; \
418_hj_k -= 12; \
419} \
420hashv += keylen; \
421switch ( _hj_k ) { \
422case 11: hashv += ( (unsigned)_hj_key[10] << 24 ); \
423case 10: hashv += ( (unsigned)_hj_key[9] << 16 ); \
424case 9: hashv += ( (unsigned)_hj_key[8] << 8 ); \
425case 8: _hj_j += ( (unsigned)_hj_key[7] << 24 ); \
426case 7: _hj_j += ( (unsigned)_hj_key[6] << 16 ); \
427case 6: _hj_j += ( (unsigned)_hj_key[5] << 8 ); \
428case 5: _hj_j += _hj_key[4]; \
429case 4: _hj_i += ( (unsigned)_hj_key[3] << 24 ); \
430case 3: _hj_i += ( (unsigned)_hj_key[2] << 16 ); \
431case 2: _hj_i += ( (unsigned)_hj_key[1] << 8 ); \
432case 1: _hj_i += _hj_key[0]; \
433} \
434HASH_JEN_MIX(_hj_i, _hj_j, hashv); \
435bkt = hashv & (num_bkts-1); \
436} while(0)
437
438/* The Paul Hsieh hash function */
439#undef get16bits
440#if (defined(__GNUC__) && defined(__i386__)) || defined(__WATCOMC__) \
441|| defined(_MSC_VER) || defined (__BORLANDC__) || defined (__TURBOC__)
442#define get16bits(d) (*((const uint16_t *) (d)))
443#endif
444
445#if !defined (get16bits)
446#define get16bits(d) ((((uint32_t)(((const uint8_t *)(d))[1])) << 8) \
447+(uint32_t)(((const uint8_t *)(d))[0]) )
448#endif
449#define HASH_SFH(key,keylen,num_bkts,hashv,bkt) \
450do { \
451char *_sfh_key=(char*)(key); \
452uint32_t _sfh_tmp, _sfh_len = keylen; \
453\
454int _sfh_rem = _sfh_len & 3; \
455_sfh_len >>= 2; \
456hashv = 0xcafebabe; \
457\
458/* Main loop */ \
459for (;_sfh_len > 0; _sfh_len--) { \
460hashv += get16bits (_sfh_key); \
461_sfh_tmp = (get16bits (_sfh_key+2) << 11) ^ hashv; \
462hashv = (hashv << 16) ^ _sfh_tmp; \
463_sfh_key += 2*sizeof (uint16_t); \
464hashv += hashv >> 11; \
465} \
466\
467/* Handle end cases */ \
468switch (_sfh_rem) { \
469case 3: hashv += get16bits (_sfh_key); \
470hashv ^= hashv << 16; \
471hashv ^= _sfh_key[sizeof (uint16_t)] << 18; \
472hashv += hashv >> 11; \
473break; \
474case 2: hashv += get16bits (_sfh_key); \
475hashv ^= hashv << 11; \
476hashv += hashv >> 17; \
477break; \
478case 1: hashv += *_sfh_key; \
479hashv ^= hashv << 10; \
480hashv += hashv >> 1; \
481} \
482\
483/* Force "avalanching" of final 127 bits */ \
484hashv ^= hashv << 3; \
485hashv += hashv >> 5; \
486hashv ^= hashv << 4; \
487hashv += hashv >> 17; \
488hashv ^= hashv << 25; \
489hashv += hashv >> 6; \
490bkt = hashv & (num_bkts-1); \
491} while(0)
492
493#ifdef HASH_USING_NO_STRICT_ALIASING
494/* The MurmurHash exploits some CPU's (x86,x86_64) tolerance for unaligned reads.
495 * For other types of CPU's (e.g. Sparc) an unaligned read causes a bus error.
496 * MurmurHash uses the faster approach only on CPU's where we know it's safe.
497 *
498 * Note the preprocessor built-in defines can be emitted using:
499 *
500 * gcc -m64 -dM -E - < /dev/null (on gcc)
501 * cc -## a.c (where a.c is a simple test file) (Sun Studio)
502 */
503#if (defined(__i386__) || defined(__x86_64__) || defined(_M_IX86))
504#define MUR_GETBLOCK(p,i) p[i]
505#else /* non intel */
506#define MUR_PLUS0_ALIGNED(p) (((unsigned long)p & 0x3) == 0)
507#define MUR_PLUS1_ALIGNED(p) (((unsigned long)p & 0x3) == 1)
508#define MUR_PLUS2_ALIGNED(p) (((unsigned long)p & 0x3) == 2)
509#define MUR_PLUS3_ALIGNED(p) (((unsigned long)p & 0x3) == 3)
510#define WP(p) ((uint32_t*)((unsigned long)(p) & ~3UL))
511#if (defined(__BIG_ENDIAN__) || defined(SPARC) || defined(__ppc__) || defined(__ppc64__))
512#define MUR_THREE_ONE(p) ((((*WP(p))&0x00ffffff) << 8) | (((*(WP(p)+1))&0xff000000) >> 24))
513#define MUR_TWO_TWO(p) ((((*WP(p))&0x0000ffff) <<16) | (((*(WP(p)+1))&0xffff0000) >> 16))
514#define MUR_ONE_THREE(p) ((((*WP(p))&0x000000ff) <<24) | (((*(WP(p)+1))&0xffffff00) >> 8))
515#else /* assume little endian non-intel */
516#define MUR_THREE_ONE(p) ((((*WP(p))&0xffffff00) >> 8) | (((*(WP(p)+1))&0x000000ff) << 24))
517#define MUR_TWO_TWO(p) ((((*WP(p))&0xffff0000) >>16) | (((*(WP(p)+1))&0x0000ffff) << 16))
518#define MUR_ONE_THREE(p) ((((*WP(p))&0xff000000) >>24) | (((*(WP(p)+1))&0x00ffffff) << 8))
519#endif
520#define MUR_GETBLOCK(p,i) (MUR_PLUS0_ALIGNED(p) ? ((p)[i]) : \
521(MUR_PLUS1_ALIGNED(p) ? MUR_THREE_ONE(p) : \
522(MUR_PLUS2_ALIGNED(p) ? MUR_TWO_TWO(p) : \
523MUR_ONE_THREE(p))))
524#endif
525#define MUR_ROTL32(x,r) (((x) << (r)) | ((x) >> (32 - (r))))
526#define MUR_FMIX(_h) \
527do { \
528_h ^= _h >> 16; \
529_h *= 0x85ebca6b; \
530_h ^= _h >> 13; \
531_h *= 0xc2b2ae35l; \
532_h ^= _h >> 16; \
533} while(0)
534
535#define HASH_MUR(key,keylen,num_bkts,hashv,bkt) \
536do { \
537const uint8_t *_mur_data = (const uint8_t*)(key); \
538const int _mur_nblocks = (keylen) / 4; \
539uint32_t _mur_h1 = 0xf88D5353; \
540uint32_t _mur_c1 = 0xcc9e2d51; \
541uint32_t _mur_c2 = 0x1b873593; \
542uint32_t _mur_k1 = 0; \
543const uint8_t *_mur_tail; \
544const uint32_t *_mur_blocks = (const uint32_t*)(_mur_data+_mur_nblocks*4); \
545int _mur_i; \
546for(_mur_i = -_mur_nblocks; _mur_i; _mur_i++) { \
547_mur_k1 = MUR_GETBLOCK(_mur_blocks,_mur_i); \
548_mur_k1 *= _mur_c1; \
549_mur_k1 = MUR_ROTL32(_mur_k1,15); \
550_mur_k1 *= _mur_c2; \
551\
552_mur_h1 ^= _mur_k1; \
553_mur_h1 = MUR_ROTL32(_mur_h1,13); \
554_mur_h1 = _mur_h1*5+0xe6546b64; \
555} \
556_mur_tail = (const uint8_t*)(_mur_data + _mur_nblocks*4); \
557_mur_k1=0; \
558switch((keylen) & 3) { \
559case 3: _mur_k1 ^= _mur_tail[2] << 16; \
560case 2: _mur_k1 ^= _mur_tail[1] << 8; \
561case 1: _mur_k1 ^= _mur_tail[0]; \
562_mur_k1 *= _mur_c1; \
563_mur_k1 = MUR_ROTL32(_mur_k1,15); \
564_mur_k1 *= _mur_c2; \
565_mur_h1 ^= _mur_k1; \
566} \
567_mur_h1 ^= (keylen); \
568MUR_FMIX(_mur_h1); \
569hashv = _mur_h1; \
570bkt = hashv & (num_bkts-1); \
571} while(0)
572#endif /* HASH_USING_NO_STRICT_ALIASING */
573
574/* key comparison function; return 0 if keys equal */
575#define HASH_KEYCMP(a,b,len) memcmp(a,b,len)
576
577/* iterate over items in a known bucket to find desired item */
578#define HASH_FIND_IN_BKT(tbl,hh,head,keyptr,keylen_in,out) \
579do { \
580if (head.hh_head) DECLTYPE_ASSIGN(out,ELMT_FROM_HH(tbl,head.hh_head)); \
581else out=NULL; \
582while (out) { \
583if ((out)->hh.keylen == keylen_in) { \
584if ((HASH_KEYCMP((out)->hh.key,keyptr,keylen_in)) == 0) break; \
585} \
586if ((out)->hh.hh_next) DECLTYPE_ASSIGN(out,ELMT_FROM_HH(tbl,(out)->hh.hh_next)); \
587else out = NULL; \
588} \
589} while(0)
590
591/* add an item to a bucket */
592#define HASH_ADD_TO_BKT(head,addhh) \
593do { \
594head.count++; \
595(addhh)->hh_next = head.hh_head; \
596(addhh)->hh_prev = NULL; \
597if (head.hh_head) { (head).hh_head->hh_prev = (addhh); } \
598(head).hh_head=addhh; \
599if (head.count >= ((head.expand_mult+1) * HASH_BKT_CAPACITY_THRESH) \
600&& (addhh)->tbl->noexpand != 1) { \
601HASH_EXPAND_BUCKETS((addhh)->tbl); \
602} \
603} while(0)
604
605/* remove an item from a given bucket */
606#define HASH_DEL_IN_BKT(hh,head,hh_del) \
607(head).count--; \
608if ((head).hh_head == hh_del) { \
609(head).hh_head = hh_del->hh_next; \
610} \
611if (hh_del->hh_prev) { \
612hh_del->hh_prev->hh_next = hh_del->hh_next; \
613} \
614if (hh_del->hh_next) { \
615hh_del->hh_next->hh_prev = hh_del->hh_prev; \
616}
617
618/* Bucket expansion has the effect of doubling the number of buckets
619 * and redistributing the items into the new buckets. Ideally the
620 * items will distribute more or less evenly into the new buckets
621 * (the extent to which this is true is a measure of the quality of
622 * the hash function as it applies to the key domain).
623 *
624 * With the items distributed into more buckets, the chain length
625 * (item count) in each bucket is reduced. Thus by expanding buckets
626 * the hash keeps a bound on the chain length. This bounded chain
627 * length is the essence of how a hash provides constant time lookup.
628 *
629 * The calculation of tbl->ideal_chain_maxlen below deserves some
630 * explanation. First, keep in mind that we're calculating the ideal
631 * maximum chain length based on the *new* (doubled) bucket count.
632 * In fractions this is just n/b (n=number of items,b=new num buckets).
633 * Since the ideal chain length is an integer, we want to calculate
634 * ceil(n/b). We don't depend on floating point arithmetic in this
635 * hash, so to calculate ceil(n/b) with integers we could write
636 *
637 * ceil(n/b) = (n/b) + ((n%b)?1:0)
638 *
639 * and in fact a previous version of this hash did just that.
640 * But now we have improved things a bit by recognizing that b is
641 * always a power of two. We keep its base 2 log handy (call it lb),
642 * so now we can write this with a bit shift and logical AND:
643 *
644 * ceil(n/b) = (n>>lb) + ( (n & (b-1)) ? 1:0)
645 *
646 */
647#define HASH_EXPAND_BUCKETS(tbl) \
648do { \
649unsigned _he_bkt; \
650unsigned _he_bkt_i; \
651struct UT_hash_handle *_he_thh, *_he_hh_nxt; \
652UT_hash_bucket *_he_new_buckets, *_he_newbkt; \
653unsigned long _he_bkt_size = 2 * tbl->num_buckets \
654* sizeof(struct UT_hash_bucket); \
655if (!(_he_bkt_size > 0)) { uthash_fatal( "unknown error"); } \
656_he_new_buckets = (UT_hash_bucket*)uthash_malloc(_he_bkt_size); \
657if (!_he_new_buckets) { uthash_fatal( "out of memory"); } \
658memset(_he_new_buckets, 0, \
6592 * tbl->num_buckets * sizeof(struct UT_hash_bucket)); \
660tbl->ideal_chain_maxlen = \
661(tbl->num_items >> (tbl->log2_num_buckets+1)) + \
662((tbl->num_items & ((tbl->num_buckets*2)-1)) ? 1 : 0); \
663tbl->nonideal_items = 0; \
664for(_he_bkt_i = 0; _he_bkt_i < tbl->num_buckets; _he_bkt_i++) \
665{ \
666_he_thh = tbl->buckets[ _he_bkt_i ].hh_head; \
667while (_he_thh) { \
668_he_hh_nxt = _he_thh->hh_next; \
669HASH_TO_BKT( _he_thh->hashv, tbl->num_buckets*2, _he_bkt); \
670_he_newbkt = &(_he_new_buckets[ _he_bkt ]); \
671if (_he_newbkt) { \
672if (++(_he_newbkt->count) > tbl->ideal_chain_maxlen) { \
673tbl->nonideal_items++; \
674_he_newbkt->expand_mult = _he_newbkt->count / \
675tbl->ideal_chain_maxlen; \
676} \
677_he_thh->hh_prev = NULL; \
678_he_thh->hh_next = _he_newbkt->hh_head; \
679if (_he_newbkt->hh_head) _he_newbkt->hh_head->hh_prev = \
680_he_thh; \
681_he_newbkt->hh_head = _he_thh; \
682_he_thh = _he_hh_nxt; \
683} \
684else { uthash_fatal( "out of memory"); } \
685} \
686} \
687uthash_free( tbl->buckets, tbl->num_buckets*sizeof(struct UT_hash_bucket) ); \
688tbl->num_buckets *= 2; \
689tbl->log2_num_buckets++; \
690tbl->buckets = _he_new_buckets; \
691tbl->ineff_expands = (tbl->nonideal_items > (tbl->num_items >> 1)) ? \
692(tbl->ineff_expands+1) : 0; \
693if (tbl->ineff_expands > 1) { \
694tbl->noexpand=1; \
695uthash_noexpand_fyi(tbl); \
696} \
697uthash_expand_fyi(tbl); \
698} while(0)
699
700
701/* This is an adaptation of Simon Tatham's O(n log(n)) mergesort */
702/* Note that HASH_SORT assumes the hash handle name to be hh.
703 * HASH_SRT was added to allow the hash handle name to be passed in. */
704#define HASH_SORT(head,cmpfcn) HASH_SRT(hh,head,cmpfcn)
705#define HASH_SRT(hh,head,cmpfcn) \
706do { \
707unsigned _hs_i; \
708unsigned _hs_looping,_hs_nmerges,_hs_insize,_hs_psize,_hs_qsize; \
709struct UT_hash_handle *_hs_p, *_hs_q, *_hs_e, *_hs_list, *_hs_tail; \
710if (head) { \
711_hs_insize = 1; \
712_hs_looping = 1; \
713_hs_list = &((head)->hh); \
714while (_hs_looping) { \
715_hs_p = _hs_list; \
716_hs_list = NULL; \
717_hs_tail = NULL; \
718_hs_nmerges = 0; \
719while (_hs_p) { \
720_hs_nmerges++; \
721_hs_q = _hs_p; \
722_hs_psize = 0; \
723for ( _hs_i = 0; _hs_i < _hs_insize; _hs_i++ ) { \
724_hs_psize++; \
725_hs_q = (UT_hash_handle*)((_hs_q->next) ? \
726((void*)((char*)(_hs_q->next) + \
727(head)->hh.tbl->hho)) : NULL); \
728if (! (_hs_q) ) break; \
729} \
730_hs_qsize = _hs_insize; \
731while ((_hs_psize > 0) || ((_hs_qsize > 0) && _hs_q )) { \
732if (_hs_psize == 0) { \
733_hs_e = _hs_q; \
734_hs_q = (UT_hash_handle*)((_hs_q->next) ? \
735((void*)((char*)(_hs_q->next) + \
736(head)->hh.tbl->hho)) : NULL); \
737_hs_qsize--; \
738} else if ( (_hs_qsize == 0) || !(_hs_q) ) { \
739_hs_e = _hs_p; \
740_hs_p = (UT_hash_handle*)((_hs_p->next) ? \
741((void*)((char*)(_hs_p->next) + \
742(head)->hh.tbl->hho)) : NULL); \
743_hs_psize--; \
744} else if (( \
745cmpfcn(DECLTYPE(head)(ELMT_FROM_HH((head)->hh.tbl,_hs_p)), \
746DECLTYPE(head)(ELMT_FROM_HH((head)->hh.tbl,_hs_q))) \
747) <= 0) { \
748_hs_e = _hs_p; \
749_hs_p = (UT_hash_handle*)((_hs_p->next) ? \
750((void*)((char*)(_hs_p->next) + \
751(head)->hh.tbl->hho)) : NULL); \
752_hs_psize--; \
753} else { \
754_hs_e = _hs_q; \
755_hs_q = (UT_hash_handle*)((_hs_q->next) ? \
756((void*)((char*)(_hs_q->next) + \
757(head)->hh.tbl->hho)) : NULL); \
758_hs_qsize--; \
759} \
760if ( _hs_tail ) { \
761_hs_tail->next = ((_hs_e) ? \
762ELMT_FROM_HH((head)->hh.tbl,_hs_e) : NULL); \
763} else { \
764_hs_list = _hs_e; \
765} \
766_hs_e->prev = ((_hs_tail) ? \
767ELMT_FROM_HH((head)->hh.tbl,_hs_tail) : NULL); \
768_hs_tail = _hs_e; \
769} \
770_hs_p = _hs_q; \
771} \
772_hs_tail->next = NULL; \
773if ( _hs_nmerges <= 1 ) { \
774_hs_looping=0; \
775(head)->hh.tbl->tail = _hs_tail; \
776DECLTYPE_ASSIGN(head,ELMT_FROM_HH((head)->hh.tbl, _hs_list)); \
777} \
778_hs_insize *= 2; \
779} \
780HASH_FSCK(hh,head); \
781} \
782} while (0)
783
784/* This function selects items from one hash into another hash.
785 * The end result is that the selected items have dual presence
786 * in both hashes. There is no copy of the items made; rather
787 * they are added into the new hash through a secondary hash
788 * hash handle that must be present in the structure. */
789#define HASH_SELECT(hh_dst, dst, hh_src, src, cond) \
790do { \
791unsigned _src_bkt, _dst_bkt; \
792void *_last_elt=NULL, *_elt; \
793UT_hash_handle *_src_hh, *_dst_hh, *_last_elt_hh=NULL; \
794ptrdiff_t _dst_hho = ((char*)(&(dst)->hh_dst) - (char*)(dst)); \
795if (src) { \
796for(_src_bkt=0; _src_bkt < (src)->hh_src.tbl->num_buckets; _src_bkt++) { \
797for(_src_hh = (src)->hh_src.tbl->buckets[_src_bkt].hh_head; \
798_src_hh; \
799_src_hh = _src_hh->hh_next) { \
800_elt = ELMT_FROM_HH((src)->hh_src.tbl, _src_hh); \
801if (cond(_elt)) { \
802_dst_hh = (UT_hash_handle*)(((char*)_elt) + _dst_hho); \
803_dst_hh->key = _src_hh->key; \
804_dst_hh->keylen = _src_hh->keylen; \
805_dst_hh->hashv = _src_hh->hashv; \
806_dst_hh->prev = _last_elt; \
807_dst_hh->next = NULL; \
808if (_last_elt_hh) { _last_elt_hh->next = _elt; } \
809if (!dst) { \
810DECLTYPE_ASSIGN(dst,_elt); \
811HASH_MAKE_TABLE(hh_dst,dst); \
812} else { \
813_dst_hh->tbl = (dst)->hh_dst.tbl; \
814} \
815HASH_TO_BKT(_dst_hh->hashv, _dst_hh->tbl->num_buckets, _dst_bkt); \
816HASH_ADD_TO_BKT(_dst_hh->tbl->buckets[_dst_bkt],_dst_hh); \
817(dst)->hh_dst.tbl->num_items++; \
818_last_elt = _elt; \
819_last_elt_hh = _dst_hh; \
820} \
821} \
822} \
823} \
824HASH_FSCK(hh_dst,dst); \
825} while (0)
826
827#define HASH_CLEAR(hh,head) \
828do { \
829if (head) { \
830uthash_free((head)->hh.tbl->buckets, \
831(head)->hh.tbl->num_buckets*sizeof(struct UT_hash_bucket)); \
832HASH_BLOOM_FREE((head)->hh.tbl); \
833uthash_free((head)->hh.tbl, sizeof(UT_hash_table)); \
834(head)=NULL; \
835} \
836} while(0)
837
838#ifdef NO_DECLTYPE
839#define HASH_ITER(hh,head,el,tmp) \
840for((el)=(head), (*(char**)(&(tmp)))=(char*)((head)?(head)->hh.next:NULL); \
841el; (el)=(tmp),(*(char**)(&(tmp)))=(char*)((tmp)?(tmp)->hh.next:NULL))
842#else
843#define HASH_ITER(hh,head,el,tmp) \
844for((el)=(head),(tmp)=DECLTYPE(el)((head)?(head)->hh.next:NULL); \
845el; (el)=(tmp),(tmp)=DECLTYPE(el)((tmp)?(tmp)->hh.next:NULL))
846#endif
847
848/* obtain a count of items in the hash */
849#define HASH_COUNT(head) HASH_CNT(hh,head)
850#define HASH_CNT(hh,head) ((head)?((head)->hh.tbl->num_items):0)
851
852typedef struct UT_hash_bucket {
853struct UT_hash_handle *hh_head;
854unsigned count;
855
856/* expand_mult is normally set to 0. In this situation, the max chain length
857 * threshold is enforced at its default value, HASH_BKT_CAPACITY_THRESH. (If
858 * the bucket's chain exceeds this length, bucket expansion is triggered).
859 * However, setting expand_mult to a non-zero value delays bucket expansion
860 * (that would be triggered by additions to this particular bucket)
861 * until its chain length reaches a *multiple* of HASH_BKT_CAPACITY_THRESH.
862 * (The multiplier is simply expand_mult+1). The whole idea of this
863 * multiplier is to reduce bucket expansions, since they are expensive, in
864 * situations where we know that a particular bucket tends to be overused.
865 * It is better to let its chain length grow to a longer yet-still-bounded
866 * value, than to do an O(n) bucket expansion too often.
867 */
868unsigned expand_mult;
869
870} UT_hash_bucket;
871
872/* random signature used only to find hash tables in external analysis */
873#define HASH_SIGNATURE 0xa0111fe1
874#define HASH_BLOOM_SIGNATURE 0xb12220f2
875
876typedef struct UT_hash_table {
877UT_hash_bucket *buckets;
878unsigned num_buckets, log2_num_buckets;
879unsigned num_items;
880struct UT_hash_handle *tail; /* tail hh in app order, for fast append */
881ptrdiff_t hho; /* hash handle offset (byte pos of hash handle in element */
882
883/* in an ideal situation (all buckets used equally), no bucket would have
884 * more than ceil(#items/#buckets) items. that's the ideal chain length. */
885unsigned ideal_chain_maxlen;
886
887/* nonideal_items is the number of items in the hash whose chain position
888 * exceeds the ideal chain maxlen. these items pay the penalty for an uneven
889 * hash distribution; reaching them in a chain traversal takes >ideal steps */
890unsigned nonideal_items;
891
892/* ineffective expands occur when a bucket doubling was performed, but
893 * afterward, more than half the items in the hash had nonideal chain
894 * positions. If this happens on two consecutive expansions we inhibit any
895 * further expansion, as it's not helping; this happens when the hash
896 * function isn't a good fit for the key domain. When expansion is inhibited
897 * the hash will still work, albeit no longer in constant time. */
898unsigned ineff_expands, noexpand;
899
900uint32_t signature; /* used only to find hash tables in external analysis */
901#ifdef HASH_BLOOM
902uint32_t bloom_sig; /* used only to test bloom exists in external analysis */
903uint8_t *bloom_bv;
904char bloom_nbits;
905#endif
906
907} UT_hash_table;
908
909typedef struct UT_hash_handle {
910struct UT_hash_table *tbl;
911void *prev; /* prev element in app order */
912void *next; /* next element in app order */
913struct UT_hash_handle *hh_prev; /* previous hh in bucket order */
914struct UT_hash_handle *hh_next; /* next hh in bucket order */
915void *key; /* ptr to enclosing struct's key */
916unsigned keylen; /* enclosing struct's key len */
917unsigned hashv; /* result of hash-fcn(key) */
918} UT_hash_handle;
919
920#endif /* UTHASH_H */
921

Archive Download this file

Revision: HEAD