/* -*- Mode: C; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 4 -*- */ /* Copyright (c) 2003-2010, Troy D. Hanson http://uthash.sourceforge.net All rights reserved.
Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met:
* Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer.
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
/* These macros use decltype or the earlier __typeof GNU extension. As decltype is only available in newer compilers (VS2010 or gcc 4.3+ when compiling c++ source) this code uses whatever method is needed
or, for VS2008 where neither is available, uses casting workarounds. */ #ifdef _MSC_VER /* MS compiler */ #if _MSC_VER >= 1600 && defined(__cplusplus) /* VS2010 or newer in C++ mode */ #define DECLTYPE(x) (decltype(x)) #else/* VS2008 or older (or VS2010 in C mode) */ #define NO_DECLTYPE #define DECLTYPE(x) #endif #else/* GNU, Sun and other compilers */ #define DECLTYPE(x) (__typeof(x)) #endif
/* a number of the hash function use uint32_t which isn't defined on win32 */ #ifdef _MSC_VER typedefunsignedint uint32_t; #else #include <inttypes.h> /* uint32_t */ #endif
#define uthash_noexpand_fyi(tbl) /* can be defined to log noexpand */ #define uthash_expand_fyi(tbl) /* can be defined to log expands */
/* initial number of buckets */ #define HASH_INITIAL_NUM_BUCKETS 32 /* initial number of buckets */ #define HASH_INITIAL_NUM_BUCKETS_LOG2 5 /* lg2 of initial number of buckets */ #define HASH_BKT_CAPACITY_THRESH 10 /* expand when bucket count reaches */
/* calculate the element whose hash handle address is the */ #define ELMT_FROM_HH(tbl,hhp) ((void*)(((char*)(hhp)) - ((tbl)->hho)))
#define HASH_FIND(hh,head,keyptr,keylen,out) \ do { \
out=NULL; \ if (head) { \ unsigned _hf_bkt,_hf_hashv; \
HASH_FCN(keyptr,keylen, (head)->hh.tbl->num_buckets, _hf_hashv, _hf_bkt); \ if (HASH_BLOOM_TEST((head)->hh.tbl, _hf_hashv)) { \
HASH_FIND_IN_BKT((head)->hh.tbl, hh, (head)->hh.tbl->buckets[ _hf_bkt ], \
keyptr,keylen,out); \
} \
} \
} while (0)
/* delete "delptr" from the hash table. * "the usual" patch-up process for the app-order doubly-linked-list. * The use of _hd_hh_del below deserves special explanation. * These used to be expressed using (delptr) but that led to a bug * if someone used the same symbol for the head and deletee, like * HASH_DELETE(hh,users,users); * We want that to work, but by changing the head (users) below * we were forfeiting our ability to further refer to the deletee (users) * in the patch-up process. Solution: use scratch space to * copy the deletee pointer, then the latter references are via that * scratch pointer rather than through the repointed (users) symbol.
*/ #define HASH_DELETE(hh,head,delptr) \ do { \ unsigned _hd_bkt; \ struct UT_hash_handle *_hd_hh_del; \ if ( ((delptr)->hh.prev == NULL) && ((delptr)->hh.next == NULL) ) { \
uthash_free((head)->hh.tbl->buckets, \
(head)->hh.tbl->num_buckets*sizeof(struct UT_hash_bucket) ); \
HASH_BLOOM_FREE((head)->hh.tbl); \
uthash_free((head)->hh.tbl, sizeof(UT_hash_table)); \
head = NULL; \
} else { \
_hd_hh_del = &((delptr)->hh); \ if ((delptr) == ELMT_FROM_HH((head)->hh.tbl,(head)->hh.tbl->tail)) { \
(head)->hh.tbl->tail = \
(UT_hash_handle*)((char*)((delptr)->hh.prev) + \
(head)->hh.tbl->hho); \
} \ if ((delptr)->hh.prev) { \
((UT_hash_handle*)((char*)((delptr)->hh.prev) + \
(head)->hh.tbl->hho))->next = (delptr)->hh.next; \
} else { \
DECLTYPE_ASSIGN(head,(delptr)->hh.next); \
} \ if (_hd_hh_del->next) { \
((UT_hash_handle*)((char*)_hd_hh_del->next + \
(head)->hh.tbl->hho))->prev = \
_hd_hh_del->prev; \
} \
HASH_TO_BKT( _hd_hh_del->hashv, (head)->hh.tbl->num_buckets, _hd_bkt); \
HASH_DEL_IN_BKT(hh,(head)->hh.tbl->buckets[_hd_bkt], _hd_hh_del); \
(head)->hh.tbl->num_items--; \
} \
HASH_FSCK(hh,head); \
} while (0)
/* HASH_FSCK checks hash integrity on every add/delete when HASH_DEBUG is defined. * This is for uthash developer only; it compiles away if HASH_DEBUG isn't defined.
*/ #ifdef HASH_DEBUG #define HASH_OOPS(...) do { fprintf(stderr,__VA_ARGS__); exit(-1); } while (0) #define HASH_FSCK(hh,head) \ do { \ if (head) { \ unsigned _bkt_i; \ unsigned _count, _bkt_count; \ char *_prev; \ struct UT_hash_handle *_thh; \
_count = 0; \ for( _bkt_i = 0; _bkt_i < (head)->hh.tbl->num_buckets; _bkt_i++) { \
_bkt_count = 0; \
_thh = (head)->hh.tbl->buckets[_bkt_i].hh_head; \
_prev = NULL; \ while (_thh) { \ if (_prev != (char*)(_thh->hh_prev)) { \
HASH_OOPS("invalid hh_prev %p, actual %p\n", \
_thh->hh_prev, _prev ); \
} \
_bkt_count++; \
_prev = (char*)(_thh); \
_thh = _thh->hh_next; \
} \
_count += _bkt_count; \ if ((head)->hh.tbl->buckets[_bkt_i].count != _bkt_count) { \
HASH_OOPS("invalid bucket count %d, actual %u\n", \
(head)->hh.tbl->buckets[_bkt_i].count, _bkt_count); \
} \
} \ if (_count != (head)->hh.tbl->num_items) { \
HASH_OOPS("invalid hh item count %d, actual %u\n", \
(head)->hh.tbl->num_items, _count ); \
} \ /* traverse hh in app order; check next/prev integrity, count */ \
_count = 0; \
_prev = NULL; \
_thh = &(head)->hh; \ while (_thh) { \
_count++; \ if (_prev !=(char*)(_thh->prev)) { \
HASH_OOPS("invalid prev %p, actual %p\n", \
_thh->prev, _prev ); \
} \
_prev = (char*)ELMT_FROM_HH((head)->hh.tbl, _thh); \
_thh = ( _thh->next ? (UT_hash_handle*)((char*)(_thh->next) + \
(head)->hh.tbl->hho) : NULL ); \
} \ if (_count != (head)->hh.tbl->num_items) { \
HASH_OOPS("invalid app item count %d, actual %u\n", \
(head)->hh.tbl->num_items, _count ); \
} \
} \
} while (0) #else #define HASH_FSCK(hh,head) #endif
/* When compiled with -DHASH_EMIT_KEYS, length-prefixed keys are emitted to * the descriptor to which this macro is defined for tuning the hash function.
* The app can #include <unistd.h> to get the prototype for write(2). */ #ifdef HASH_EMIT_KEYS #define HASH_EMIT_KEY(hh,head,keyptr,fieldlen) \ do { \ unsigned _klen = fieldlen; \
write(HASH_EMIT_KEYS, &_klen, sizeof(_klen)); \
write(HASH_EMIT_KEYS, keyptr, fieldlen); \
} while (0) #else #define HASH_EMIT_KEY(hh,head,keyptr,fieldlen) #endif
/* default to Jenkin's hash unless overridden e.g. DHASH_FUNCTION=HASH_SAX */ #ifdef HASH_FUNCTION #define HASH_FCN HASH_FUNCTION #else #define HASH_FCN HASH_JEN #endif
/* The Bernstein hash function, used in Perl prior to v5.6 */ #define HASH_BER(key,keylen,num_bkts,hashv,bkt) \ do { \ unsigned _hb_keylen=keylen; \ char *_hb_key=(char*)(key); \
(hashv) = 0; \ while (_hb_keylen--) { (hashv) = ((hashv) * 33) + *_hb_key++; } \
bkt = (hashv) & (num_bkts-1); \
} while (0)
/* SAX/FNV/OAT/JEN hash functions are macro variants of those listed at
* http://eternallyconfuzzled.com/tuts/algorithms/jsw_tut_hashing.aspx */ #define HASH_SAX(key,keylen,num_bkts,hashv,bkt) \ do { \ unsigned _sx_i; \ char *_hs_key=(char*)(key); \
hashv = 0; \ for(_sx_i=0; _sx_i < keylen; _sx_i++) \
hashv ^= (hashv << 5) + (hashv >> 2) + _hs_key[_sx_i]; \
bkt = hashv & (num_bkts-1); \
} while (0)
#define HASH_JEN_MIX(a,b,c) \ do { \
a -= b; a -= c; a ^= ( c >> 13 ); \
b -= c; b -= a; b ^= ( a << 8 ); \
c -= a; c -= b; c ^= ( b >> 13 ); \
a -= b; a -= c; a ^= ( c >> 12 ); \
b -= c; b -= a; b ^= ( a << 16 ); \
c -= a; c -= b; c ^= ( b >> 5 ); \
a -= b; a -= c; a ^= ( c >> 3 ); \
b -= c; b -= a; b ^= ( a << 10 ); \
c -= a; c -= b; c ^= ( b >> 15 ); \
} while (0)
#ifdef HASH_USING_NO_STRICT_ALIASING /* The MurmurHash exploits some CPU's (e.g. x86) tolerance for unaligned reads. * For other types of CPU's (e.g. Sparc) an unaligned read causes a bus error. * So MurmurHash comes in two versions, the faster unaligned one and the slower * aligned one. We only use the faster one on CPU's where we know it's safe. * * Note the preprocessor built-in defines can be emitted using: * * gcc -m64 -dM -E - < /dev/null (on gcc) * cc -## a.c (where a.c is a simple test file) (Sun Studio)
*/ #if (defined(__i386__) || defined(__x86_64__)) #define HASH_MUR HASH_MUR_UNALIGNED #else #define HASH_MUR HASH_MUR_ALIGNED #endif
/* iterate over items in a known bucket to find desired item */ #define HASH_FIND_IN_BKT(tbl,hh,head,keyptr,keylen_in,out) \ do { \ if (head.hh_head) DECLTYPE_ASSIGN(out,ELMT_FROM_HH(tbl,head.hh_head)); \ else out=NULL; \ while (out) { \ if (out->hh.keylen == keylen_in) { \ if ((HASH_KEYCMP(out->hh.key,keyptr,keylen_in)) == 0) break; \
} \ if (out->hh.hh_next) DECLTYPE_ASSIGN(out,ELMT_FROM_HH(tbl,out->hh.hh_next)); \ else out = NULL; \
} \
} while(0)
/* add an item to a bucket */ #define HASH_ADD_TO_BKT(head,addhh) \ do { \
head.count++; \
(addhh)->hh_next = head.hh_head; \
(addhh)->hh_prev = NULL; \ if (head.hh_head) { (head).hh_head->hh_prev = (addhh); } \
(head).hh_head=addhh; \ if (head.count >= ((head.expand_mult+1) * HASH_BKT_CAPACITY_THRESH) \
&& (addhh)->tbl->noexpand != 1) { \
HASH_EXPAND_BUCKETS((addhh)->tbl); \
} \
} while(0)
/* remove an item from a given bucket */ #define HASH_DEL_IN_BKT(hh,head,hh_del) \
(head).count--; \ if ((head).hh_head == hh_del) { \
(head).hh_head = hh_del->hh_next; \
} \ if (hh_del->hh_prev) { \
hh_del->hh_prev->hh_next = hh_del->hh_next; \
} \ if (hh_del->hh_next) { \
hh_del->hh_next->hh_prev = hh_del->hh_prev; \
}
/* Bucket expansion has the effect of doubling the number of buckets * and redistributing the items into the new buckets. Ideally the * items will distribute more or less evenly into the new buckets * (the extent to which this is true is a measure of the quality of * the hash function as it applies to the key domain). * * With the items distributed into more buckets, the chain length * (item count) in each bucket is reduced. Thus by expanding buckets * the hash keeps a bound on the chain length. This bounded chain * length is the essence of how a hash provides constant time lookup. * * The calculation of tbl->ideal_chain_maxlen below deserves some * explanation. First, keep in mind that we're calculating the ideal * maximum chain length based on the *new* (doubled) bucket count. * In fractions this is just n/b (n=number of items,b=new num buckets). * Since the ideal chain length is an integer, we want to calculate * ceil(n/b). We don't depend on floating point arithmetic in this * hash, so to calculate ceil(n/b) with integers we could write * * ceil(n/b) = (n/b) + ((n%b)?1:0) * * and in fact a previous version of this hash did just that. * But now we have improved things a bit by recognizing that b is * always a power of two. We keep its base 2 log handy (call it lb), * so now we can write this with a bit shift and logical AND: * * ceil(n/b) = (n>>lb) + ( (n & (b-1)) ? 1:0) *
*/ #define HASH_EXPAND_BUCKETS(tbl) \ do { \ unsigned _he_bkt; \ unsigned _he_bkt_i; \ struct UT_hash_handle *_he_thh, *_he_hh_nxt; \
UT_hash_bucket *_he_new_buckets, *_he_newbkt; \
_he_new_buckets = (UT_hash_bucket*)uthash_malloc( \
2 * tbl->num_buckets * sizeof(struct UT_hash_bucket)); \ if (!_he_new_buckets) { uthash_fatal( "out of memory"); } \
memset(_he_new_buckets, 0, \
2 * tbl->num_buckets * sizeof(struct UT_hash_bucket)); \
tbl->ideal_chain_maxlen = \
(tbl->num_items >> (tbl->log2_num_buckets+1)) + \
((tbl->num_items & ((tbl->num_buckets*2)-1)) ? 1 : 0); \
tbl->nonideal_items = 0; \ for(_he_bkt_i = 0; _he_bkt_i < tbl->num_buckets; _he_bkt_i++) \
{ \
_he_thh = tbl->buckets[ _he_bkt_i ].hh_head; \ while (_he_thh) { \
_he_hh_nxt = _he_thh->hh_next; \
HASH_TO_BKT( _he_thh->hashv, tbl->num_buckets*2, _he_bkt); \
_he_newbkt = &(_he_new_buckets[ _he_bkt ]); \ if (++(_he_newbkt->count) > tbl->ideal_chain_maxlen) { \
tbl->nonideal_items++; \
_he_newbkt->expand_mult = _he_newbkt->count / \
tbl->ideal_chain_maxlen; \
} \
_he_thh->hh_prev = NULL; \
_he_thh->hh_next = _he_newbkt->hh_head; \ if (_he_newbkt->hh_head) _he_newbkt->hh_head->hh_prev = \
_he_thh; \
_he_newbkt->hh_head = _he_thh; \
_he_thh = _he_hh_nxt; \
} \
} \
uthash_free( tbl->buckets, tbl->num_buckets*sizeof(struct UT_hash_bucket) ); \
tbl->num_buckets *= 2; \
tbl->log2_num_buckets++; \
tbl->buckets = _he_new_buckets; \
tbl->ineff_expands = (tbl->nonideal_items > (tbl->num_items >> 1)) ? \
(tbl->ineff_expands+1) : 0; \ if (tbl->ineff_expands > 1) { \
tbl->noexpand=1; \
uthash_noexpand_fyi(tbl); \
} \
uthash_expand_fyi(tbl); \
} while(0)
/* expand_mult is normally set to 0. In this situation, the max chain length * threshold is enforced at its default value, HASH_BKT_CAPACITY_THRESH. (If * the bucket's chain exceeds this length, bucket expansion is triggered). * However, setting expand_mult to a non-zero value delays bucket expansion * (that would be triggered by additions to this particular bucket) * until its chain length reaches a *multiple* of HASH_BKT_CAPACITY_THRESH. * (The multiplier is simply expand_mult+1). The whole idea of this * multiplier is to reduce bucket expansions, since they are expensive, in * situations where we know that a particular bucket tends to be overused. * It is better to let its chain length grow to a longer yet-still-bounded * value, than to do an O(n) bucket expansion too often.
*/ unsigned expand_mult;
} UT_hash_bucket;
/* random signature used only to find hash tables in external analysis */ #define HASH_SIGNATURE 0xa0111fe1 #define HASH_BLOOM_SIGNATURE 0xb12220f2
typedefstruct UT_hash_table {
UT_hash_bucket *buckets; unsigned num_buckets, log2_num_buckets; unsigned num_items; struct UT_hash_handle *tail; /* tail hh in app order, for fast append */
ptrdiff_t hho; /* hash handle offset (byte pos of hash handle in element */
/* in an ideal situation (all buckets used equally), no bucket would have
* more than ceil(#items/#buckets) items. that's the ideal chain length. */ unsigned ideal_chain_maxlen;
/* nonideal_items is the number of items in the hash whose chain position * exceeds the ideal chain maxlen. these items pay the penalty for an uneven
* hash distribution; reaching them in a chain traversal takes >ideal steps */ unsigned nonideal_items;
/* ineffective expands occur when a bucket doubling was performed, but * afterward, more than half the items in the hash had nonideal chain * positions. If this happens on two consecutive expansions we inhibit any * further expansion, as it's not helping; this happens when the hash * function isn't a good fit for the key domain. When expansion is inhibited
* the hash will still work, albeit no longer in constant time. */ unsigned ineff_expands, noexpand;
uint32_t signature; /* used only to find hash tables in external analysis */ #ifdef HASH_BLOOM
uint32_t bloom_sig; /* used only to test bloom exists in external analysis */
uint8_t *bloom_bv; char bloom_nbits; #endif
} UT_hash_table;
typedefstruct UT_hash_handle { struct UT_hash_table *tbl; void *prev; /* prev element in app order */ void *next; /* next element in app order */ struct UT_hash_handle *hh_prev; /* previous hh in bucket order */ struct UT_hash_handle *hh_next; /* next hh in bucket order */ void *key; /* ptr to enclosing struct's key */ unsigned keylen; /* enclosing struct's key len */ unsigned hashv; /* result of hash-fcn(key) */
} UT_hash_handle;
Die Informationen auf dieser Webseite wurden
nach bestem Wissen sorgfältig zusammengestellt. Es wird jedoch weder Vollständigkeit, noch Richtigkeit,
noch Qualität der bereit gestellten Informationen zugesichert.
Bemerkung:
Die farbliche Syntaxdarstellung und die Messung sind noch experimentell.