Path: blob/main/contrib/elftoolchain/common/uthash.h
39483 views
/*1Copyright (c) 2003-2013, Troy D. Hanson http://uthash.sourceforge.net2All rights reserved.34Redistribution and use in source and binary forms, with or without5modification, are permitted provided that the following conditions are met:67* Redistributions of source code must retain the above copyright8notice, this list of conditions and the following disclaimer.910THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS11IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED12TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A13PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER14OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,15EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,16PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR17PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF18LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING19NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS20SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.21*/2223/* $Id: uthash.h 2682 2012-11-23 22:04:22Z kaiwang27 $ */2425#ifndef UTHASH_H26#define UTHASH_H2728#include <string.h> /* memcmp,strlen */29#include <stddef.h> /* ptrdiff_t */30#include <stdlib.h> /* exit() */3132/* These macros use decltype or the earlier __typeof GNU extension.33As decltype is only available in newer compilers (VS2010 or gcc 4.3+34when compiling c++ source) this code uses whatever method is needed35or, for VS2008 where neither is available, uses casting workarounds. */36#ifdef _MSC_VER /* MS compiler */37#if _MSC_VER >= 1600 && defined(__cplusplus) /* VS2010 or newer in C++ mode */38#define DECLTYPE(x) (decltype(x))39#else /* VS2008 or older (or VS2010 in C mode) */40#define NO_DECLTYPE41#define DECLTYPE(x)42#endif43#else /* GNU, Sun and other compilers */44#define DECLTYPE(x) (__typeof(x))45#endif4647#ifdef NO_DECLTYPE48#define DECLTYPE_ASSIGN(dst,src) \49do { \50char **_da_dst = (char**)(&(dst)); \51*_da_dst = (char*)(src); \52} while(0)53#else54#define DECLTYPE_ASSIGN(dst,src) \55do { \56(dst) = DECLTYPE(dst)(src); \57} while(0)58#endif5960/* a number of the hash function use uint32_t which isn't defined on win32 */61#ifdef _MSC_VER62typedef unsigned int uint32_t;63typedef unsigned char uint8_t;64#else65#include <inttypes.h> /* uint32_t */66#endif6768#define UTHASH_VERSION 1.9.76970#ifndef uthash_fatal71#define uthash_fatal(msg) exit(-1) /* fatal error (out of memory,etc) */72#endif73#ifndef uthash_malloc74#define uthash_malloc(sz) malloc(sz) /* malloc fcn */75#endif76#ifndef uthash_free77#define uthash_free(ptr,sz) free(ptr) /* free fcn */78#endif7980#ifndef uthash_noexpand_fyi81#define uthash_noexpand_fyi(tbl) /* can be defined to log noexpand */82#endif83#ifndef uthash_expand_fyi84#define uthash_expand_fyi(tbl) /* can be defined to log expands */85#endif8687/* initial number of buckets */88#define HASH_INITIAL_NUM_BUCKETS 32 /* initial number of buckets */89#define HASH_INITIAL_NUM_BUCKETS_LOG2 5 /* lg2 of initial number of buckets */90#define HASH_BKT_CAPACITY_THRESH 10 /* expand when bucket count reaches */9192/* calculate the element whose hash handle address is hhe */93#define ELMT_FROM_HH(tbl,hhp) ((void*)(((char*)(hhp)) - ((tbl)->hho)))9495#define HASH_FIND(hh,head,keyptr,keylen,out) \96do { \97unsigned _hf_bkt,_hf_hashv; \98out=NULL; \99if (head) { \100HASH_FCN(keyptr,keylen, (head)->hh.tbl->num_buckets, _hf_hashv, _hf_bkt); \101if (HASH_BLOOM_TEST((head)->hh.tbl, _hf_hashv)) { \102HASH_FIND_IN_BKT((head)->hh.tbl, hh, (head)->hh.tbl->buckets[ _hf_bkt ], \103keyptr,keylen,out); \104} \105} \106} while (0)107108#ifdef HASH_BLOOM109#define HASH_BLOOM_BITLEN (1ULL << HASH_BLOOM)110#define HASH_BLOOM_BYTELEN (HASH_BLOOM_BITLEN/8) + ((HASH_BLOOM_BITLEN%8) ? 1:0)111#define HASH_BLOOM_MAKE(tbl) \112do { \113(tbl)->bloom_nbits = HASH_BLOOM; \114(tbl)->bloom_bv = (uint8_t*)uthash_malloc(HASH_BLOOM_BYTELEN); \115if (!((tbl)->bloom_bv)) { uthash_fatal( "out of memory"); } \116memset((tbl)->bloom_bv, 0, HASH_BLOOM_BYTELEN); \117(tbl)->bloom_sig = HASH_BLOOM_SIGNATURE; \118} while (0)119120#define HASH_BLOOM_FREE(tbl) \121do { \122uthash_free((tbl)->bloom_bv, HASH_BLOOM_BYTELEN); \123} while (0)124125#define HASH_BLOOM_BITSET(bv,idx) (bv[(idx)/8] |= (1U << ((idx)%8)))126#define HASH_BLOOM_BITTEST(bv,idx) (bv[(idx)/8] & (1U << ((idx)%8)))127128#define HASH_BLOOM_ADD(tbl,hashv) \129HASH_BLOOM_BITSET((tbl)->bloom_bv, (hashv & (uint32_t)((1ULL << (tbl)->bloom_nbits) - 1)))130131#define HASH_BLOOM_TEST(tbl,hashv) \132HASH_BLOOM_BITTEST((tbl)->bloom_bv, (hashv & (uint32_t)((1ULL << (tbl)->bloom_nbits) - 1)))133134#else135#define HASH_BLOOM_MAKE(tbl)136#define HASH_BLOOM_FREE(tbl)137#define HASH_BLOOM_ADD(tbl,hashv)138#define HASH_BLOOM_TEST(tbl,hashv) (1)139#endif140141#define HASH_MAKE_TABLE(hh,head) \142do { \143(head)->hh.tbl = (UT_hash_table*)uthash_malloc( \144sizeof(UT_hash_table)); \145if (!((head)->hh.tbl)) { uthash_fatal( "out of memory"); } \146memset((head)->hh.tbl, 0, sizeof(UT_hash_table)); \147(head)->hh.tbl->tail = &((head)->hh); \148(head)->hh.tbl->num_buckets = HASH_INITIAL_NUM_BUCKETS; \149(head)->hh.tbl->log2_num_buckets = HASH_INITIAL_NUM_BUCKETS_LOG2; \150(head)->hh.tbl->hho = (char*)(&(head)->hh) - (char*)(head); \151(head)->hh.tbl->buckets = (UT_hash_bucket*)uthash_malloc( \152HASH_INITIAL_NUM_BUCKETS*sizeof(struct UT_hash_bucket)); \153if (! (head)->hh.tbl->buckets) { uthash_fatal( "out of memory"); } \154memset((head)->hh.tbl->buckets, 0, \155HASH_INITIAL_NUM_BUCKETS*sizeof(struct UT_hash_bucket)); \156HASH_BLOOM_MAKE((head)->hh.tbl); \157(head)->hh.tbl->signature = HASH_SIGNATURE; \158} while(0)159160#define HASH_ADD(hh,head,fieldname,keylen_in,add) \161HASH_ADD_KEYPTR(hh,head,&((add)->fieldname),keylen_in,add)162163#define HASH_ADD_KEYPTR(hh,head,keyptr,keylen_in,add) \164do { \165unsigned _ha_bkt; \166(add)->hh.next = NULL; \167(add)->hh.key = (char*)keyptr; \168(add)->hh.keylen = (unsigned)keylen_in; \169if (!(head)) { \170head = (add); \171(head)->hh.prev = NULL; \172HASH_MAKE_TABLE(hh,head); \173} else { \174(head)->hh.tbl->tail->next = (add); \175(add)->hh.prev = ELMT_FROM_HH((head)->hh.tbl, (head)->hh.tbl->tail); \176(head)->hh.tbl->tail = &((add)->hh); \177} \178(head)->hh.tbl->num_items++; \179(add)->hh.tbl = (head)->hh.tbl; \180HASH_FCN(keyptr,keylen_in, (head)->hh.tbl->num_buckets, \181(add)->hh.hashv, _ha_bkt); \182HASH_ADD_TO_BKT((head)->hh.tbl->buckets[_ha_bkt],&(add)->hh); \183HASH_BLOOM_ADD((head)->hh.tbl,(add)->hh.hashv); \184HASH_EMIT_KEY(hh,head,keyptr,keylen_in); \185HASH_FSCK(hh,head); \186} while(0)187188#define HASH_TO_BKT( hashv, num_bkts, bkt ) \189do { \190bkt = ((hashv) & ((num_bkts) - 1)); \191} while(0)192193/* delete "delptr" from the hash table.194* "the usual" patch-up process for the app-order doubly-linked-list.195* The use of _hd_hh_del below deserves special explanation.196* These used to be expressed using (delptr) but that led to a bug197* if someone used the same symbol for the head and deletee, like198* HASH_DELETE(hh,users,users);199* We want that to work, but by changing the head (users) below200* we were forfeiting our ability to further refer to the deletee (users)201* in the patch-up process. Solution: use scratch space to202* copy the deletee pointer, then the latter references are via that203* scratch pointer rather than through the repointed (users) symbol.204*/205#define HASH_DELETE(hh,head,delptr) \206do { \207unsigned _hd_bkt; \208struct UT_hash_handle *_hd_hh_del; \209if ( ((delptr)->hh.prev == NULL) && ((delptr)->hh.next == NULL) ) { \210uthash_free((head)->hh.tbl->buckets, \211(head)->hh.tbl->num_buckets*sizeof(struct UT_hash_bucket) ); \212HASH_BLOOM_FREE((head)->hh.tbl); \213uthash_free((head)->hh.tbl, sizeof(UT_hash_table)); \214head = NULL; \215} else { \216_hd_hh_del = &((delptr)->hh); \217if ((delptr) == ELMT_FROM_HH((head)->hh.tbl,(head)->hh.tbl->tail)) { \218(head)->hh.tbl->tail = \219(UT_hash_handle*)((ptrdiff_t)((delptr)->hh.prev) + \220(head)->hh.tbl->hho); \221} \222if ((delptr)->hh.prev) { \223((UT_hash_handle*)((ptrdiff_t)((delptr)->hh.prev) + \224(head)->hh.tbl->hho))->next = (delptr)->hh.next; \225} else { \226DECLTYPE_ASSIGN(head,(delptr)->hh.next); \227} \228if (_hd_hh_del->next) { \229((UT_hash_handle*)((ptrdiff_t)_hd_hh_del->next + \230(head)->hh.tbl->hho))->prev = \231_hd_hh_del->prev; \232} \233HASH_TO_BKT( _hd_hh_del->hashv, (head)->hh.tbl->num_buckets, _hd_bkt); \234HASH_DEL_IN_BKT(hh,(head)->hh.tbl->buckets[_hd_bkt], _hd_hh_del); \235(head)->hh.tbl->num_items--; \236} \237HASH_FSCK(hh,head); \238} while (0)239240241/* convenience forms of HASH_FIND/HASH_ADD/HASH_DEL */242#define HASH_FIND_STR(head,findstr,out) \243HASH_FIND(hh,head,findstr,strlen(findstr),out)244#define HASH_ADD_STR(head,strfield,add) \245HASH_ADD(hh,head,strfield,strlen(add->strfield),add)246#define HASH_FIND_INT(head,findint,out) \247HASH_FIND(hh,head,findint,sizeof(int),out)248#define HASH_ADD_INT(head,intfield,add) \249HASH_ADD(hh,head,intfield,sizeof(int),add)250#define HASH_FIND_PTR(head,findptr,out) \251HASH_FIND(hh,head,findptr,sizeof(void *),out)252#define HASH_ADD_PTR(head,ptrfield,add) \253HASH_ADD(hh,head,ptrfield,sizeof(void *),add)254#define HASH_DEL(head,delptr) \255HASH_DELETE(hh,head,delptr)256257/* HASH_FSCK checks hash integrity on every add/delete when HASH_DEBUG is defined.258* This is for uthash developer only; it compiles away if HASH_DEBUG isn't defined.259*/260#ifdef HASH_DEBUG261#define HASH_OOPS(...) do { fprintf(stderr,__VA_ARGS__); exit(-1); } while (0)262#define HASH_FSCK(hh,head) \263do { \264unsigned _bkt_i; \265unsigned _count, _bkt_count; \266char *_prev; \267struct UT_hash_handle *_thh; \268if (head) { \269_count = 0; \270for( _bkt_i = 0; _bkt_i < (head)->hh.tbl->num_buckets; _bkt_i++) { \271_bkt_count = 0; \272_thh = (head)->hh.tbl->buckets[_bkt_i].hh_head; \273_prev = NULL; \274while (_thh) { \275if (_prev != (char*)(_thh->hh_prev)) { \276HASH_OOPS("invalid hh_prev %p, actual %p\n", \277_thh->hh_prev, _prev ); \278} \279_bkt_count++; \280_prev = (char*)(_thh); \281_thh = _thh->hh_next; \282} \283_count += _bkt_count; \284if ((head)->hh.tbl->buckets[_bkt_i].count != _bkt_count) { \285HASH_OOPS("invalid bucket count %d, actual %d\n", \286(head)->hh.tbl->buckets[_bkt_i].count, _bkt_count); \287} \288} \289if (_count != (head)->hh.tbl->num_items) { \290HASH_OOPS("invalid hh item count %d, actual %d\n", \291(head)->hh.tbl->num_items, _count ); \292} \293/* traverse hh in app order; check next/prev integrity, count */ \294_count = 0; \295_prev = NULL; \296_thh = &(head)->hh; \297while (_thh) { \298_count++; \299if (_prev !=(char*)(_thh->prev)) { \300HASH_OOPS("invalid prev %p, actual %p\n", \301_thh->prev, _prev ); \302} \303_prev = (char*)ELMT_FROM_HH((head)->hh.tbl, _thh); \304_thh = ( _thh->next ? (UT_hash_handle*)((char*)(_thh->next) + \305(head)->hh.tbl->hho) : NULL ); \306} \307if (_count != (head)->hh.tbl->num_items) { \308HASH_OOPS("invalid app item count %d, actual %d\n", \309(head)->hh.tbl->num_items, _count ); \310} \311} \312} while (0)313#else314#define HASH_FSCK(hh,head)315#endif316317/* When compiled with -DHASH_EMIT_KEYS, length-prefixed keys are emitted to318* the descriptor to which this macro is defined for tuning the hash function.319* The app can #include <unistd.h> to get the prototype for write(2). */320#ifdef HASH_EMIT_KEYS321#define HASH_EMIT_KEY(hh,head,keyptr,fieldlen) \322do { \323unsigned _klen = fieldlen; \324write(HASH_EMIT_KEYS, &_klen, sizeof(_klen)); \325write(HASH_EMIT_KEYS, keyptr, fieldlen); \326} while (0)327#else328#define HASH_EMIT_KEY(hh,head,keyptr,fieldlen)329#endif330331/* default to Jenkin's hash unless overridden e.g. DHASH_FUNCTION=HASH_SAX */332#ifdef HASH_FUNCTION333#define HASH_FCN HASH_FUNCTION334#else335#define HASH_FCN HASH_JEN336#endif337338/* The Bernstein hash function, used in Perl prior to v5.6 */339#define HASH_BER(key,keylen,num_bkts,hashv,bkt) \340do { \341unsigned _hb_keylen=keylen; \342char *_hb_key=(char*)(key); \343(hashv) = 0; \344while (_hb_keylen--) { (hashv) = ((hashv) * 33) + *_hb_key++; } \345bkt = (hashv) & (num_bkts-1); \346} while (0)347348349/* SAX/FNV/OAT/JEN hash functions are macro variants of those listed at350* http://eternallyconfuzzled.com/tuts/algorithms/jsw_tut_hashing.aspx */351#define HASH_SAX(key,keylen,num_bkts,hashv,bkt) \352do { \353unsigned _sx_i; \354char *_hs_key=(char*)(key); \355hashv = 0; \356for(_sx_i=0; _sx_i < keylen; _sx_i++) \357hashv ^= (hashv << 5) + (hashv >> 2) + _hs_key[_sx_i]; \358bkt = hashv & (num_bkts-1); \359} while (0)360361#define HASH_FNV(key,keylen,num_bkts,hashv,bkt) \362do { \363unsigned _fn_i; \364char *_hf_key=(char*)(key); \365hashv = 2166136261UL; \366for(_fn_i=0; _fn_i < keylen; _fn_i++) \367hashv = (hashv * 16777619) ^ _hf_key[_fn_i]; \368bkt = hashv & (num_bkts-1); \369} while(0)370371#define HASH_OAT(key,keylen,num_bkts,hashv,bkt) \372do { \373unsigned _ho_i; \374char *_ho_key=(char*)(key); \375hashv = 0; \376for(_ho_i=0; _ho_i < keylen; _ho_i++) { \377hashv += _ho_key[_ho_i]; \378hashv += (hashv << 10); \379hashv ^= (hashv >> 6); \380} \381hashv += (hashv << 3); \382hashv ^= (hashv >> 11); \383hashv += (hashv << 15); \384bkt = hashv & (num_bkts-1); \385} while(0)386387#define HASH_JEN_MIX(a,b,c) \388do { \389a -= b; a -= c; a ^= ( c >> 13 ); \390b -= c; b -= a; b ^= ( a << 8 ); \391c -= a; c -= b; c ^= ( b >> 13 ); \392a -= b; a -= c; a ^= ( c >> 12 ); \393b -= c; b -= a; b ^= ( a << 16 ); \394c -= a; c -= b; c ^= ( b >> 5 ); \395a -= b; a -= c; a ^= ( c >> 3 ); \396b -= c; b -= a; b ^= ( a << 10 ); \397c -= a; c -= b; c ^= ( b >> 15 ); \398} while (0)399400#define HASH_JEN(key,keylen,num_bkts,hashv,bkt) \401do { \402unsigned _hj_i,_hj_j,_hj_k; \403char *_hj_key=(char*)(key); \404hashv = 0xfeedbeef; \405_hj_i = _hj_j = 0x9e3779b9; \406_hj_k = (unsigned)keylen; \407while (_hj_k >= 12) { \408_hj_i += (_hj_key[0] + ( (unsigned)_hj_key[1] << 8 ) \409+ ( (unsigned)_hj_key[2] << 16 ) \410+ ( (unsigned)_hj_key[3] << 24 ) ); \411_hj_j += (_hj_key[4] + ( (unsigned)_hj_key[5] << 8 ) \412+ ( (unsigned)_hj_key[6] << 16 ) \413+ ( (unsigned)_hj_key[7] << 24 ) ); \414hashv += (_hj_key[8] + ( (unsigned)_hj_key[9] << 8 ) \415+ ( (unsigned)_hj_key[10] << 16 ) \416+ ( (unsigned)_hj_key[11] << 24 ) ); \417\418HASH_JEN_MIX(_hj_i, _hj_j, hashv); \419\420_hj_key += 12; \421_hj_k -= 12; \422} \423hashv += keylen; \424switch ( _hj_k ) { \425case 11: hashv += ( (unsigned)_hj_key[10] << 24 ); \426case 10: hashv += ( (unsigned)_hj_key[9] << 16 ); \427case 9: hashv += ( (unsigned)_hj_key[8] << 8 ); \428case 8: _hj_j += ( (unsigned)_hj_key[7] << 24 ); \429case 7: _hj_j += ( (unsigned)_hj_key[6] << 16 ); \430case 6: _hj_j += ( (unsigned)_hj_key[5] << 8 ); \431case 5: _hj_j += _hj_key[4]; \432case 4: _hj_i += ( (unsigned)_hj_key[3] << 24 ); \433case 3: _hj_i += ( (unsigned)_hj_key[2] << 16 ); \434case 2: _hj_i += ( (unsigned)_hj_key[1] << 8 ); \435case 1: _hj_i += _hj_key[0]; \436} \437HASH_JEN_MIX(_hj_i, _hj_j, hashv); \438bkt = hashv & (num_bkts-1); \439} while(0)440441/* The Paul Hsieh hash function */442#undef get16bits443#if (defined(__GNUC__) && defined(__i386__)) || defined(__WATCOMC__) \444|| defined(_MSC_VER) || defined (__BORLANDC__) || defined (__TURBOC__)445#define get16bits(d) (*((const uint16_t *) (d)))446#endif447448#if !defined (get16bits)449#define get16bits(d) ((((uint32_t)(((const uint8_t *)(d))[1])) << 8) \450+(uint32_t)(((const uint8_t *)(d))[0]) )451#endif452#define HASH_SFH(key,keylen,num_bkts,hashv,bkt) \453do { \454char *_sfh_key=(char*)(key); \455uint32_t _sfh_tmp, _sfh_len = keylen; \456\457int _sfh_rem = _sfh_len & 3; \458_sfh_len >>= 2; \459hashv = 0xcafebabe; \460\461/* Main loop */ \462for (;_sfh_len > 0; _sfh_len--) { \463hashv += get16bits (_sfh_key); \464_sfh_tmp = (get16bits (_sfh_key+2) << 11) ^ hashv; \465hashv = (hashv << 16) ^ _sfh_tmp; \466_sfh_key += 2*sizeof (uint16_t); \467hashv += hashv >> 11; \468} \469\470/* Handle end cases */ \471switch (_sfh_rem) { \472case 3: hashv += get16bits (_sfh_key); \473hashv ^= hashv << 16; \474hashv ^= _sfh_key[sizeof (uint16_t)] << 18; \475hashv += hashv >> 11; \476break; \477case 2: hashv += get16bits (_sfh_key); \478hashv ^= hashv << 11; \479hashv += hashv >> 17; \480break; \481case 1: hashv += *_sfh_key; \482hashv ^= hashv << 10; \483hashv += hashv >> 1; \484} \485\486/* Force "avalanching" of final 127 bits */ \487hashv ^= hashv << 3; \488hashv += hashv >> 5; \489hashv ^= hashv << 4; \490hashv += hashv >> 17; \491hashv ^= hashv << 25; \492hashv += hashv >> 6; \493bkt = hashv & (num_bkts-1); \494} while(0)495496#ifdef HASH_USING_NO_STRICT_ALIASING497/* The MurmurHash exploits some CPU's (x86,x86_64) tolerance for unaligned reads.498* For other types of CPU's (e.g. Sparc) an unaligned read causes a bus error.499* MurmurHash uses the faster approach only on CPU's where we know it's safe.500*501* Note the preprocessor built-in defines can be emitted using:502*503* gcc -m64 -dM -E - < /dev/null (on gcc)504* cc -## a.c (where a.c is a simple test file) (Sun Studio)505*/506#if (defined(__i386__) || defined(__x86_64__) || defined(_M_IX86))507#define MUR_GETBLOCK(p,i) p[i]508#else /* non intel */509#define MUR_PLUS0_ALIGNED(p) (((unsigned long)p & 0x3) == 0)510#define MUR_PLUS1_ALIGNED(p) (((unsigned long)p & 0x3) == 1)511#define MUR_PLUS2_ALIGNED(p) (((unsigned long)p & 0x3) == 2)512#define MUR_PLUS3_ALIGNED(p) (((unsigned long)p & 0x3) == 3)513#define WP(p) ((uint32_t*)((unsigned long)(p) & ~3UL))514#if (defined(__BIG_ENDIAN__) || defined(SPARC) || defined(__ppc__) || defined(__ppc64__))515#define MUR_THREE_ONE(p) ((((*WP(p))&0x00ffffff) << 8) | (((*(WP(p)+1))&0xff000000) >> 24))516#define MUR_TWO_TWO(p) ((((*WP(p))&0x0000ffff) <<16) | (((*(WP(p)+1))&0xffff0000) >> 16))517#define MUR_ONE_THREE(p) ((((*WP(p))&0x000000ff) <<24) | (((*(WP(p)+1))&0xffffff00) >> 8))518#else /* assume little endian non-intel */519#define MUR_THREE_ONE(p) ((((*WP(p))&0xffffff00) >> 8) | (((*(WP(p)+1))&0x000000ff) << 24))520#define MUR_TWO_TWO(p) ((((*WP(p))&0xffff0000) >>16) | (((*(WP(p)+1))&0x0000ffff) << 16))521#define MUR_ONE_THREE(p) ((((*WP(p))&0xff000000) >>24) | (((*(WP(p)+1))&0x00ffffff) << 8))522#endif523#define MUR_GETBLOCK(p,i) (MUR_PLUS0_ALIGNED(p) ? ((p)[i]) : \524(MUR_PLUS1_ALIGNED(p) ? MUR_THREE_ONE(p) : \525(MUR_PLUS2_ALIGNED(p) ? MUR_TWO_TWO(p) : \526MUR_ONE_THREE(p))))527#endif528#define MUR_ROTL32(x,r) (((x) << (r)) | ((x) >> (32 - (r))))529#define MUR_FMIX(_h) \530do { \531_h ^= _h >> 16; \532_h *= 0x85ebca6b; \533_h ^= _h >> 13; \534_h *= 0xc2b2ae35l; \535_h ^= _h >> 16; \536} while(0)537538#define HASH_MUR(key,keylen,num_bkts,hashv,bkt) \539do { \540const uint8_t *_mur_data = (const uint8_t*)(key); \541const int _mur_nblocks = (keylen) / 4; \542uint32_t _mur_h1 = 0xf88D5353; \543uint32_t _mur_c1 = 0xcc9e2d51; \544uint32_t _mur_c2 = 0x1b873593; \545uint32_t _mur_k1 = 0; \546const uint8_t *_mur_tail; \547const uint32_t *_mur_blocks = (const uint32_t*)(_mur_data+_mur_nblocks*4); \548int _mur_i; \549for(_mur_i = -_mur_nblocks; _mur_i; _mur_i++) { \550_mur_k1 = MUR_GETBLOCK(_mur_blocks,_mur_i); \551_mur_k1 *= _mur_c1; \552_mur_k1 = MUR_ROTL32(_mur_k1,15); \553_mur_k1 *= _mur_c2; \554\555_mur_h1 ^= _mur_k1; \556_mur_h1 = MUR_ROTL32(_mur_h1,13); \557_mur_h1 = _mur_h1*5+0xe6546b64; \558} \559_mur_tail = (const uint8_t*)(_mur_data + _mur_nblocks*4); \560_mur_k1=0; \561switch((keylen) & 3) { \562case 3: _mur_k1 ^= _mur_tail[2] << 16; \563case 2: _mur_k1 ^= _mur_tail[1] << 8; \564case 1: _mur_k1 ^= _mur_tail[0]; \565_mur_k1 *= _mur_c1; \566_mur_k1 = MUR_ROTL32(_mur_k1,15); \567_mur_k1 *= _mur_c2; \568_mur_h1 ^= _mur_k1; \569} \570_mur_h1 ^= (keylen); \571MUR_FMIX(_mur_h1); \572hashv = _mur_h1; \573bkt = hashv & (num_bkts-1); \574} while(0)575#endif /* HASH_USING_NO_STRICT_ALIASING */576577/* key comparison function; return 0 if keys equal */578#define HASH_KEYCMP(a,b,len) memcmp(a,b,len)579580/* iterate over items in a known bucket to find desired item */581#define HASH_FIND_IN_BKT(tbl,hh,head,keyptr,keylen_in,out) \582do { \583if (head.hh_head) DECLTYPE_ASSIGN(out,ELMT_FROM_HH(tbl,head.hh_head)); \584else out=NULL; \585while (out) { \586if ((out)->hh.keylen == keylen_in) { \587if ((HASH_KEYCMP((out)->hh.key,keyptr,keylen_in)) == 0) break; \588} \589if ((out)->hh.hh_next) DECLTYPE_ASSIGN(out,ELMT_FROM_HH(tbl,(out)->hh.hh_next)); \590else out = NULL; \591} \592} while(0)593594/* add an item to a bucket */595#define HASH_ADD_TO_BKT(head,addhh) \596do { \597head.count++; \598(addhh)->hh_next = head.hh_head; \599(addhh)->hh_prev = NULL; \600if (head.hh_head) { (head).hh_head->hh_prev = (addhh); } \601(head).hh_head=addhh; \602if (head.count >= ((head.expand_mult+1) * HASH_BKT_CAPACITY_THRESH) \603&& (addhh)->tbl->noexpand != 1) { \604HASH_EXPAND_BUCKETS((addhh)->tbl); \605} \606} while(0)607608/* remove an item from a given bucket */609#define HASH_DEL_IN_BKT(hh,head,hh_del) \610(head).count--; \611if ((head).hh_head == hh_del) { \612(head).hh_head = hh_del->hh_next; \613} \614if (hh_del->hh_prev) { \615hh_del->hh_prev->hh_next = hh_del->hh_next; \616} \617if (hh_del->hh_next) { \618hh_del->hh_next->hh_prev = hh_del->hh_prev; \619}620621/* Bucket expansion has the effect of doubling the number of buckets622* and redistributing the items into the new buckets. Ideally the623* items will distribute more or less evenly into the new buckets624* (the extent to which this is true is a measure of the quality of625* the hash function as it applies to the key domain).626*627* With the items distributed into more buckets, the chain length628* (item count) in each bucket is reduced. Thus by expanding buckets629* the hash keeps a bound on the chain length. This bounded chain630* length is the essence of how a hash provides constant time lookup.631*632* The calculation of tbl->ideal_chain_maxlen below deserves some633* explanation. First, keep in mind that we're calculating the ideal634* maximum chain length based on the *new* (doubled) bucket count.635* In fractions this is just n/b (n=number of items,b=new num buckets).636* Since the ideal chain length is an integer, we want to calculate637* ceil(n/b). We don't depend on floating point arithmetic in this638* hash, so to calculate ceil(n/b) with integers we could write639*640* ceil(n/b) = (n/b) + ((n%b)?1:0)641*642* and in fact a previous version of this hash did just that.643* But now we have improved things a bit by recognizing that b is644* always a power of two. We keep its base 2 log handy (call it lb),645* so now we can write this with a bit shift and logical AND:646*647* ceil(n/b) = (n>>lb) + ( (n & (b-1)) ? 1:0)648*649*/650#define HASH_EXPAND_BUCKETS(tbl) \651do { \652unsigned _he_bkt; \653unsigned _he_bkt_i; \654struct UT_hash_handle *_he_thh, *_he_hh_nxt; \655UT_hash_bucket *_he_new_buckets, *_he_newbkt; \656_he_new_buckets = (UT_hash_bucket*)uthash_malloc( \6572 * tbl->num_buckets * sizeof(struct UT_hash_bucket)); \658if (!_he_new_buckets) { uthash_fatal( "out of memory"); } \659memset(_he_new_buckets, 0, \6602 * tbl->num_buckets * sizeof(struct UT_hash_bucket)); \661tbl->ideal_chain_maxlen = \662(tbl->num_items >> (tbl->log2_num_buckets+1)) + \663((tbl->num_items & ((tbl->num_buckets*2)-1)) ? 1 : 0); \664tbl->nonideal_items = 0; \665for(_he_bkt_i = 0; _he_bkt_i < tbl->num_buckets; _he_bkt_i++) \666{ \667_he_thh = tbl->buckets[ _he_bkt_i ].hh_head; \668while (_he_thh) { \669_he_hh_nxt = _he_thh->hh_next; \670HASH_TO_BKT( _he_thh->hashv, tbl->num_buckets*2, _he_bkt); \671_he_newbkt = &(_he_new_buckets[ _he_bkt ]); \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} \684} \685uthash_free( tbl->buckets, tbl->num_buckets*sizeof(struct UT_hash_bucket) ); \686tbl->num_buckets *= 2; \687tbl->log2_num_buckets++; \688tbl->buckets = _he_new_buckets; \689tbl->ineff_expands = (tbl->nonideal_items > (tbl->num_items >> 1)) ? \690(tbl->ineff_expands+1) : 0; \691if (tbl->ineff_expands > 1) { \692tbl->noexpand=1; \693uthash_noexpand_fyi(tbl); \694} \695uthash_expand_fyi(tbl); \696} while(0)697698699/* This is an adaptation of Simon Tatham's O(n log(n)) mergesort */700/* Note that HASH_SORT assumes the hash handle name to be hh.701* HASH_SRT was added to allow the hash handle name to be passed in. */702#define HASH_SORT(head,cmpfcn) HASH_SRT(hh,head,cmpfcn)703#define HASH_SRT(hh,head,cmpfcn) \704do { \705unsigned _hs_i; \706unsigned _hs_looping,_hs_nmerges,_hs_insize,_hs_psize,_hs_qsize; \707struct UT_hash_handle *_hs_p, *_hs_q, *_hs_e, *_hs_list, *_hs_tail; \708if (head) { \709_hs_insize = 1; \710_hs_looping = 1; \711_hs_list = &((head)->hh); \712while (_hs_looping) { \713_hs_p = _hs_list; \714_hs_list = NULL; \715_hs_tail = NULL; \716_hs_nmerges = 0; \717while (_hs_p) { \718_hs_nmerges++; \719_hs_q = _hs_p; \720_hs_psize = 0; \721for ( _hs_i = 0; _hs_i < _hs_insize; _hs_i++ ) { \722_hs_psize++; \723_hs_q = (UT_hash_handle*)((_hs_q->next) ? \724((void*)((char*)(_hs_q->next) + \725(head)->hh.tbl->hho)) : NULL); \726if (! (_hs_q) ) break; \727} \728_hs_qsize = _hs_insize; \729while ((_hs_psize > 0) || ((_hs_qsize > 0) && _hs_q )) { \730if (_hs_psize == 0) { \731_hs_e = _hs_q; \732_hs_q = (UT_hash_handle*)((_hs_q->next) ? \733((void*)((char*)(_hs_q->next) + \734(head)->hh.tbl->hho)) : NULL); \735_hs_qsize--; \736} else if ( (_hs_qsize == 0) || !(_hs_q) ) { \737_hs_e = _hs_p; \738_hs_p = (UT_hash_handle*)((_hs_p->next) ? \739((void*)((char*)(_hs_p->next) + \740(head)->hh.tbl->hho)) : NULL); \741_hs_psize--; \742} else if (( \743cmpfcn(DECLTYPE(head)(ELMT_FROM_HH((head)->hh.tbl,_hs_p)), \744DECLTYPE(head)(ELMT_FROM_HH((head)->hh.tbl,_hs_q))) \745) <= 0) { \746_hs_e = _hs_p; \747_hs_p = (UT_hash_handle*)((_hs_p->next) ? \748((void*)((char*)(_hs_p->next) + \749(head)->hh.tbl->hho)) : NULL); \750_hs_psize--; \751} else { \752_hs_e = _hs_q; \753_hs_q = (UT_hash_handle*)((_hs_q->next) ? \754((void*)((char*)(_hs_q->next) + \755(head)->hh.tbl->hho)) : NULL); \756_hs_qsize--; \757} \758if ( _hs_tail ) { \759_hs_tail->next = ((_hs_e) ? \760ELMT_FROM_HH((head)->hh.tbl,_hs_e) : NULL); \761} else { \762_hs_list = _hs_e; \763} \764_hs_e->prev = ((_hs_tail) ? \765ELMT_FROM_HH((head)->hh.tbl,_hs_tail) : NULL); \766_hs_tail = _hs_e; \767} \768_hs_p = _hs_q; \769} \770_hs_tail->next = NULL; \771if ( _hs_nmerges <= 1 ) { \772_hs_looping=0; \773(head)->hh.tbl->tail = _hs_tail; \774DECLTYPE_ASSIGN(head,ELMT_FROM_HH((head)->hh.tbl, _hs_list)); \775} \776_hs_insize *= 2; \777} \778HASH_FSCK(hh,head); \779} \780} while (0)781782/* This function selects items from one hash into another hash.783* The end result is that the selected items have dual presence784* in both hashes. There is no copy of the items made; rather785* they are added into the new hash through a secondary hash786* hash handle that must be present in the structure. */787#define HASH_SELECT(hh_dst, dst, hh_src, src, cond) \788do { \789unsigned _src_bkt, _dst_bkt; \790void *_last_elt=NULL, *_elt; \791UT_hash_handle *_src_hh, *_dst_hh, *_last_elt_hh=NULL; \792ptrdiff_t _dst_hho = ((char*)(&(dst)->hh_dst) - (char*)(dst)); \793if (src) { \794for(_src_bkt=0; _src_bkt < (src)->hh_src.tbl->num_buckets; _src_bkt++) { \795for(_src_hh = (src)->hh_src.tbl->buckets[_src_bkt].hh_head; \796_src_hh; \797_src_hh = _src_hh->hh_next) { \798_elt = ELMT_FROM_HH((src)->hh_src.tbl, _src_hh); \799if (cond(_elt)) { \800_dst_hh = (UT_hash_handle*)(((char*)_elt) + _dst_hho); \801_dst_hh->key = _src_hh->key; \802_dst_hh->keylen = _src_hh->keylen; \803_dst_hh->hashv = _src_hh->hashv; \804_dst_hh->prev = _last_elt; \805_dst_hh->next = NULL; \806if (_last_elt_hh) { _last_elt_hh->next = _elt; } \807if (!dst) { \808DECLTYPE_ASSIGN(dst,_elt); \809HASH_MAKE_TABLE(hh_dst,dst); \810} else { \811_dst_hh->tbl = (dst)->hh_dst.tbl; \812} \813HASH_TO_BKT(_dst_hh->hashv, _dst_hh->tbl->num_buckets, _dst_bkt); \814HASH_ADD_TO_BKT(_dst_hh->tbl->buckets[_dst_bkt],_dst_hh); \815(dst)->hh_dst.tbl->num_items++; \816_last_elt = _elt; \817_last_elt_hh = _dst_hh; \818} \819} \820} \821} \822HASH_FSCK(hh_dst,dst); \823} while (0)824825#define HASH_CLEAR(hh,head) \826do { \827if (head) { \828uthash_free((head)->hh.tbl->buckets, \829(head)->hh.tbl->num_buckets*sizeof(struct UT_hash_bucket)); \830HASH_BLOOM_FREE((head)->hh.tbl); \831uthash_free((head)->hh.tbl, sizeof(UT_hash_table)); \832(head)=NULL; \833} \834} while(0)835836#ifdef NO_DECLTYPE837#define HASH_ITER(hh,head,el,tmp) \838for((el)=(head), (*(char**)(&(tmp)))=(char*)((head)?(head)->hh.next:NULL); \839el; (el)=(tmp),(*(char**)(&(tmp)))=(char*)((tmp)?(tmp)->hh.next:NULL))840#else841#define HASH_ITER(hh,head,el,tmp) \842for((el)=(head),(tmp)=DECLTYPE(el)((head)?(head)->hh.next:NULL); \843el; (el)=(tmp),(tmp)=DECLTYPE(el)((tmp)?(tmp)->hh.next:NULL))844#endif845846/* obtain a count of items in the hash */847#define HASH_COUNT(head) HASH_CNT(hh,head)848#define HASH_CNT(hh,head) ((head)?((head)->hh.tbl->num_items):0)849850typedef struct UT_hash_bucket {851struct UT_hash_handle *hh_head;852unsigned count;853854/* expand_mult is normally set to 0. In this situation, the max chain length855* threshold is enforced at its default value, HASH_BKT_CAPACITY_THRESH. (If856* the bucket's chain exceeds this length, bucket expansion is triggered).857* However, setting expand_mult to a non-zero value delays bucket expansion858* (that would be triggered by additions to this particular bucket)859* until its chain length reaches a *multiple* of HASH_BKT_CAPACITY_THRESH.860* (The multiplier is simply expand_mult+1). The whole idea of this861* multiplier is to reduce bucket expansions, since they are expensive, in862* situations where we know that a particular bucket tends to be overused.863* It is better to let its chain length grow to a longer yet-still-bounded864* value, than to do an O(n) bucket expansion too often.865*/866unsigned expand_mult;867868} UT_hash_bucket;869870/* random signature used only to find hash tables in external analysis */871#define HASH_SIGNATURE 0xa0111fe1872#define HASH_BLOOM_SIGNATURE 0xb12220f2873874typedef struct UT_hash_table {875UT_hash_bucket *buckets;876unsigned num_buckets, log2_num_buckets;877unsigned num_items;878struct UT_hash_handle *tail; /* tail hh in app order, for fast append */879ptrdiff_t hho; /* hash handle offset (byte pos of hash handle in element */880881/* in an ideal situation (all buckets used equally), no bucket would have882* more than ceil(#items/#buckets) items. that's the ideal chain length. */883unsigned ideal_chain_maxlen;884885/* nonideal_items is the number of items in the hash whose chain position886* exceeds the ideal chain maxlen. these items pay the penalty for an uneven887* hash distribution; reaching them in a chain traversal takes >ideal steps */888unsigned nonideal_items;889890/* ineffective expands occur when a bucket doubling was performed, but891* afterward, more than half the items in the hash had nonideal chain892* positions. If this happens on two consecutive expansions we inhibit any893* further expansion, as it's not helping; this happens when the hash894* function isn't a good fit for the key domain. When expansion is inhibited895* the hash will still work, albeit no longer in constant time. */896unsigned ineff_expands, noexpand;897898uint32_t signature; /* used only to find hash tables in external analysis */899#ifdef HASH_BLOOM900uint32_t bloom_sig; /* used only to test bloom exists in external analysis */901uint8_t *bloom_bv;902char bloom_nbits;903#endif904905} UT_hash_table;906907typedef struct UT_hash_handle {908struct UT_hash_table *tbl;909void *prev; /* prev element in app order */910void *next; /* next element in app order */911struct UT_hash_handle *hh_prev; /* previous hh in bucket order */912struct UT_hash_handle *hh_next; /* next hh in bucket order */913void *key; /* ptr to enclosing struct's key */914unsigned keylen; /* enclosing struct's key len */915unsigned hashv; /* result of hash-fcn(key) */916} UT_hash_handle;917918#endif /* UTHASH_H */919920921