1*b1d06847SJean-Baptiste Boric /* 2*b1d06847SJean-Baptiste Boric Copyright (c) 2003-2014, Troy D. Hanson http://troydhanson.github.com/uthash/ 3*b1d06847SJean-Baptiste Boric All rights reserved. 4*b1d06847SJean-Baptiste Boric 5*b1d06847SJean-Baptiste Boric Redistribution and use in source and binary forms, with or without 6*b1d06847SJean-Baptiste Boric modification, are permitted provided that the following conditions are met: 7*b1d06847SJean-Baptiste Boric 8*b1d06847SJean-Baptiste Boric * Redistributions of source code must retain the above copyright 9*b1d06847SJean-Baptiste Boric notice, this list of conditions and the following disclaimer. 10*b1d06847SJean-Baptiste Boric 11*b1d06847SJean-Baptiste Boric THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS 12*b1d06847SJean-Baptiste Boric IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 13*b1d06847SJean-Baptiste Boric TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A 14*b1d06847SJean-Baptiste Boric PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER 15*b1d06847SJean-Baptiste Boric OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, 16*b1d06847SJean-Baptiste Boric EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 17*b1d06847SJean-Baptiste Boric PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR 18*b1d06847SJean-Baptiste Boric PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF 19*b1d06847SJean-Baptiste Boric LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING 20*b1d06847SJean-Baptiste Boric NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS 21*b1d06847SJean-Baptiste Boric SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 22*b1d06847SJean-Baptiste Boric */ 23*b1d06847SJean-Baptiste Boric 24*b1d06847SJean-Baptiste Boric #ifndef UTHASH_H 25*b1d06847SJean-Baptiste Boric #define UTHASH_H 26*b1d06847SJean-Baptiste Boric 27*b1d06847SJean-Baptiste Boric #include <string.h> /* memcmp,strlen */ 28*b1d06847SJean-Baptiste Boric #include <stddef.h> /* ptrdiff_t */ 29*b1d06847SJean-Baptiste Boric #include <stdlib.h> /* exit() */ 30*b1d06847SJean-Baptiste Boric 31*b1d06847SJean-Baptiste Boric /* These macros use decltype or the earlier __typeof GNU extension. 32*b1d06847SJean-Baptiste Boric As decltype is only available in newer compilers (VS2010 or gcc 4.3+ 33*b1d06847SJean-Baptiste Boric when compiling c++ source) this code uses whatever method is needed 34*b1d06847SJean-Baptiste Boric or, for VS2008 where neither is available, uses casting workarounds. */ 35*b1d06847SJean-Baptiste Boric #if defined(_MSC_VER) /* MS compiler */ 36*b1d06847SJean-Baptiste Boric #if _MSC_VER >= 1600 && defined(__cplusplus) /* VS2010 or newer in C++ mode */ 37*b1d06847SJean-Baptiste Boric #define DECLTYPE(x) (decltype(x)) 38*b1d06847SJean-Baptiste Boric #else /* VS2008 or older (or VS2010 in C mode) */ 39*b1d06847SJean-Baptiste Boric #define NO_DECLTYPE 40*b1d06847SJean-Baptiste Boric #define DECLTYPE(x) 41*b1d06847SJean-Baptiste Boric #endif 42*b1d06847SJean-Baptiste Boric #elif defined(__BORLANDC__) || defined(__LCC__) || defined(__WATCOMC__) 43*b1d06847SJean-Baptiste Boric #define NO_DECLTYPE 44*b1d06847SJean-Baptiste Boric #define DECLTYPE(x) 45*b1d06847SJean-Baptiste Boric #else /* GNU, Sun and other compilers */ 46*b1d06847SJean-Baptiste Boric #define DECLTYPE(x) (__typeof(x)) 47*b1d06847SJean-Baptiste Boric #endif 48*b1d06847SJean-Baptiste Boric 49*b1d06847SJean-Baptiste Boric #ifdef NO_DECLTYPE 50*b1d06847SJean-Baptiste Boric #define DECLTYPE_ASSIGN(dst,src) \ 51*b1d06847SJean-Baptiste Boric do { \ 52*b1d06847SJean-Baptiste Boric char **_da_dst = (char**)(&(dst)); \ 53*b1d06847SJean-Baptiste Boric *_da_dst = (char*)(src); \ 54*b1d06847SJean-Baptiste Boric } while(0) 55*b1d06847SJean-Baptiste Boric #else 56*b1d06847SJean-Baptiste Boric #define DECLTYPE_ASSIGN(dst,src) \ 57*b1d06847SJean-Baptiste Boric do { \ 58*b1d06847SJean-Baptiste Boric (dst) = DECLTYPE(dst)(src); \ 59*b1d06847SJean-Baptiste Boric } while(0) 60*b1d06847SJean-Baptiste Boric #endif 61*b1d06847SJean-Baptiste Boric 62*b1d06847SJean-Baptiste Boric /* a number of the hash function use uint32_t which isn't defined on Pre VS2010 */ 63*b1d06847SJean-Baptiste Boric #if defined (_WIN32) 64*b1d06847SJean-Baptiste Boric #if defined(_MSC_VER) && _MSC_VER >= 1600 65*b1d06847SJean-Baptiste Boric #include <stdint.h> 66*b1d06847SJean-Baptiste Boric #elif defined(__WATCOMC__) 67*b1d06847SJean-Baptiste Boric #include <stdint.h> 68*b1d06847SJean-Baptiste Boric #else 69*b1d06847SJean-Baptiste Boric typedef unsigned int uint32_t; 70*b1d06847SJean-Baptiste Boric typedef unsigned char uint8_t; 71*b1d06847SJean-Baptiste Boric #endif 72*b1d06847SJean-Baptiste Boric #else 73*b1d06847SJean-Baptiste Boric #include <stdint.h> 74*b1d06847SJean-Baptiste Boric #endif 75*b1d06847SJean-Baptiste Boric 76*b1d06847SJean-Baptiste Boric #define UTHASH_VERSION 1.9.9 77*b1d06847SJean-Baptiste Boric 78*b1d06847SJean-Baptiste Boric #ifndef uthash_fatal 79*b1d06847SJean-Baptiste Boric #define uthash_fatal(msg) exit(-1) /* fatal error (out of memory,etc) */ 80*b1d06847SJean-Baptiste Boric #endif 81*b1d06847SJean-Baptiste Boric #ifndef uthash_malloc 82*b1d06847SJean-Baptiste Boric #define uthash_malloc(sz) malloc(sz) /* malloc fcn */ 83*b1d06847SJean-Baptiste Boric #endif 84*b1d06847SJean-Baptiste Boric #ifndef uthash_free 85*b1d06847SJean-Baptiste Boric #define uthash_free(ptr,sz) free(ptr) /* free fcn */ 86*b1d06847SJean-Baptiste Boric #endif 87*b1d06847SJean-Baptiste Boric 88*b1d06847SJean-Baptiste Boric #ifndef uthash_noexpand_fyi 89*b1d06847SJean-Baptiste Boric #define uthash_noexpand_fyi(tbl) /* can be defined to log noexpand */ 90*b1d06847SJean-Baptiste Boric #endif 91*b1d06847SJean-Baptiste Boric #ifndef uthash_expand_fyi 92*b1d06847SJean-Baptiste Boric #define uthash_expand_fyi(tbl) /* can be defined to log expands */ 93*b1d06847SJean-Baptiste Boric #endif 94*b1d06847SJean-Baptiste Boric 95*b1d06847SJean-Baptiste Boric /* initial number of buckets */ 96*b1d06847SJean-Baptiste Boric #define HASH_INITIAL_NUM_BUCKETS 32 /* initial number of buckets */ 97*b1d06847SJean-Baptiste Boric #define HASH_INITIAL_NUM_BUCKETS_LOG2 5 /* lg2 of initial number of buckets */ 98*b1d06847SJean-Baptiste Boric #define HASH_BKT_CAPACITY_THRESH 10 /* expand when bucket count reaches */ 99*b1d06847SJean-Baptiste Boric 100*b1d06847SJean-Baptiste Boric /* calculate the element whose hash handle address is hhe */ 101*b1d06847SJean-Baptiste Boric #define ELMT_FROM_HH(tbl,hhp) ((void*)(((char*)(hhp)) - ((tbl)->hho))) 102*b1d06847SJean-Baptiste Boric 103*b1d06847SJean-Baptiste Boric #define HASH_FIND(hh,head,keyptr,keylen,out) \ 104*b1d06847SJean-Baptiste Boric do { \ 105*b1d06847SJean-Baptiste Boric out=NULL; \ 106*b1d06847SJean-Baptiste Boric if (head) { \ 107*b1d06847SJean-Baptiste Boric unsigned _hf_bkt,_hf_hashv; \ 108*b1d06847SJean-Baptiste Boric HASH_FCN(keyptr,keylen, (head)->hh.tbl->num_buckets, _hf_hashv, _hf_bkt); \ 109*b1d06847SJean-Baptiste Boric if (HASH_BLOOM_TEST((head)->hh.tbl, _hf_hashv)) { \ 110*b1d06847SJean-Baptiste Boric HASH_FIND_IN_BKT((head)->hh.tbl, hh, (head)->hh.tbl->buckets[ _hf_bkt ], \ 111*b1d06847SJean-Baptiste Boric keyptr,keylen,out); \ 112*b1d06847SJean-Baptiste Boric } \ 113*b1d06847SJean-Baptiste Boric } \ 114*b1d06847SJean-Baptiste Boric } while (0) 115*b1d06847SJean-Baptiste Boric 116*b1d06847SJean-Baptiste Boric #ifdef HASH_BLOOM 117*b1d06847SJean-Baptiste Boric #define HASH_BLOOM_BITLEN (1ULL << HASH_BLOOM) 118*b1d06847SJean-Baptiste Boric #define HASH_BLOOM_BYTELEN (HASH_BLOOM_BITLEN/8) + ((HASH_BLOOM_BITLEN%8) ? 1:0) 119*b1d06847SJean-Baptiste Boric #define HASH_BLOOM_MAKE(tbl) \ 120*b1d06847SJean-Baptiste Boric do { \ 121*b1d06847SJean-Baptiste Boric (tbl)->bloom_nbits = HASH_BLOOM; \ 122*b1d06847SJean-Baptiste Boric (tbl)->bloom_bv = (uint8_t*)uthash_malloc(HASH_BLOOM_BYTELEN); \ 123*b1d06847SJean-Baptiste Boric if (!((tbl)->bloom_bv)) { uthash_fatal( "out of memory"); } \ 124*b1d06847SJean-Baptiste Boric memset((tbl)->bloom_bv, 0, HASH_BLOOM_BYTELEN); \ 125*b1d06847SJean-Baptiste Boric (tbl)->bloom_sig = HASH_BLOOM_SIGNATURE; \ 126*b1d06847SJean-Baptiste Boric } while (0) 127*b1d06847SJean-Baptiste Boric 128*b1d06847SJean-Baptiste Boric #define HASH_BLOOM_FREE(tbl) \ 129*b1d06847SJean-Baptiste Boric do { \ 130*b1d06847SJean-Baptiste Boric uthash_free((tbl)->bloom_bv, HASH_BLOOM_BYTELEN); \ 131*b1d06847SJean-Baptiste Boric } while (0) 132*b1d06847SJean-Baptiste Boric 133*b1d06847SJean-Baptiste Boric #define HASH_BLOOM_BITSET(bv,idx) (bv[(idx)/8] |= (1U << ((idx)%8))) 134*b1d06847SJean-Baptiste Boric #define HASH_BLOOM_BITTEST(bv,idx) (bv[(idx)/8] & (1U << ((idx)%8))) 135*b1d06847SJean-Baptiste Boric 136*b1d06847SJean-Baptiste Boric #define HASH_BLOOM_ADD(tbl,hashv) \ 137*b1d06847SJean-Baptiste Boric HASH_BLOOM_BITSET((tbl)->bloom_bv, (hashv & (uint32_t)((1ULL << (tbl)->bloom_nbits) - 1))) 138*b1d06847SJean-Baptiste Boric 139*b1d06847SJean-Baptiste Boric #define HASH_BLOOM_TEST(tbl,hashv) \ 140*b1d06847SJean-Baptiste Boric HASH_BLOOM_BITTEST((tbl)->bloom_bv, (hashv & (uint32_t)((1ULL << (tbl)->bloom_nbits) - 1))) 141*b1d06847SJean-Baptiste Boric 142*b1d06847SJean-Baptiste Boric #else 143*b1d06847SJean-Baptiste Boric #define HASH_BLOOM_MAKE(tbl) 144*b1d06847SJean-Baptiste Boric #define HASH_BLOOM_FREE(tbl) 145*b1d06847SJean-Baptiste Boric #define HASH_BLOOM_ADD(tbl,hashv) 146*b1d06847SJean-Baptiste Boric #define HASH_BLOOM_TEST(tbl,hashv) (1) 147*b1d06847SJean-Baptiste Boric #define HASH_BLOOM_BYTELEN 0 148*b1d06847SJean-Baptiste Boric #endif 149*b1d06847SJean-Baptiste Boric 150*b1d06847SJean-Baptiste Boric #define HASH_MAKE_TABLE(hh,head) \ 151*b1d06847SJean-Baptiste Boric do { \ 152*b1d06847SJean-Baptiste Boric (head)->hh.tbl = (UT_hash_table*)uthash_malloc( \ 153*b1d06847SJean-Baptiste Boric sizeof(UT_hash_table)); \ 154*b1d06847SJean-Baptiste Boric if (!((head)->hh.tbl)) { uthash_fatal( "out of memory"); } \ 155*b1d06847SJean-Baptiste Boric memset((head)->hh.tbl, 0, sizeof(UT_hash_table)); \ 156*b1d06847SJean-Baptiste Boric (head)->hh.tbl->tail = &((head)->hh); \ 157*b1d06847SJean-Baptiste Boric (head)->hh.tbl->num_buckets = HASH_INITIAL_NUM_BUCKETS; \ 158*b1d06847SJean-Baptiste Boric (head)->hh.tbl->log2_num_buckets = HASH_INITIAL_NUM_BUCKETS_LOG2; \ 159*b1d06847SJean-Baptiste Boric (head)->hh.tbl->hho = (char*)(&(head)->hh) - (char*)(head); \ 160*b1d06847SJean-Baptiste Boric (head)->hh.tbl->buckets = (UT_hash_bucket*)uthash_malloc( \ 161*b1d06847SJean-Baptiste Boric HASH_INITIAL_NUM_BUCKETS*sizeof(struct UT_hash_bucket)); \ 162*b1d06847SJean-Baptiste Boric if (! (head)->hh.tbl->buckets) { uthash_fatal( "out of memory"); } \ 163*b1d06847SJean-Baptiste Boric memset((head)->hh.tbl->buckets, 0, \ 164*b1d06847SJean-Baptiste Boric HASH_INITIAL_NUM_BUCKETS*sizeof(struct UT_hash_bucket)); \ 165*b1d06847SJean-Baptiste Boric HASH_BLOOM_MAKE((head)->hh.tbl); \ 166*b1d06847SJean-Baptiste Boric (head)->hh.tbl->signature = HASH_SIGNATURE; \ 167*b1d06847SJean-Baptiste Boric } while(0) 168*b1d06847SJean-Baptiste Boric 169*b1d06847SJean-Baptiste Boric #define HASH_ADD(hh,head,fieldname,keylen_in,add) \ 170*b1d06847SJean-Baptiste Boric HASH_ADD_KEYPTR(hh,head,&((add)->fieldname),keylen_in,add) 171*b1d06847SJean-Baptiste Boric 172*b1d06847SJean-Baptiste Boric #define HASH_REPLACE(hh,head,fieldname,keylen_in,add,replaced) \ 173*b1d06847SJean-Baptiste Boric do { \ 174*b1d06847SJean-Baptiste Boric replaced=NULL; \ 175*b1d06847SJean-Baptiste Boric HASH_FIND(hh,head,&((add)->fieldname),keylen_in,replaced); \ 176*b1d06847SJean-Baptiste Boric if (replaced!=NULL) { \ 177*b1d06847SJean-Baptiste Boric HASH_DELETE(hh,head,replaced); \ 178*b1d06847SJean-Baptiste Boric } \ 179*b1d06847SJean-Baptiste Boric HASH_ADD(hh,head,fieldname,keylen_in,add); \ 180*b1d06847SJean-Baptiste Boric } while(0) 181*b1d06847SJean-Baptiste Boric 182*b1d06847SJean-Baptiste Boric #define HASH_ADD_KEYPTR(hh,head,keyptr,keylen_in,add) \ 183*b1d06847SJean-Baptiste Boric do { \ 184*b1d06847SJean-Baptiste Boric unsigned _ha_bkt; \ 185*b1d06847SJean-Baptiste Boric (add)->hh.next = NULL; \ 186*b1d06847SJean-Baptiste Boric (add)->hh.key = (char*)(keyptr); \ 187*b1d06847SJean-Baptiste Boric (add)->hh.keylen = (unsigned)(keylen_in); \ 188*b1d06847SJean-Baptiste Boric if (!(head)) { \ 189*b1d06847SJean-Baptiste Boric head = (add); \ 190*b1d06847SJean-Baptiste Boric (head)->hh.prev = NULL; \ 191*b1d06847SJean-Baptiste Boric HASH_MAKE_TABLE(hh,head); \ 192*b1d06847SJean-Baptiste Boric } else { \ 193*b1d06847SJean-Baptiste Boric (head)->hh.tbl->tail->next = (add); \ 194*b1d06847SJean-Baptiste Boric (add)->hh.prev = ELMT_FROM_HH((head)->hh.tbl, (head)->hh.tbl->tail); \ 195*b1d06847SJean-Baptiste Boric (head)->hh.tbl->tail = &((add)->hh); \ 196*b1d06847SJean-Baptiste Boric } \ 197*b1d06847SJean-Baptiste Boric (head)->hh.tbl->num_items++; \ 198*b1d06847SJean-Baptiste Boric (add)->hh.tbl = (head)->hh.tbl; \ 199*b1d06847SJean-Baptiste Boric HASH_FCN(keyptr,keylen_in, (head)->hh.tbl->num_buckets, \ 200*b1d06847SJean-Baptiste Boric (add)->hh.hashv, _ha_bkt); \ 201*b1d06847SJean-Baptiste Boric HASH_ADD_TO_BKT((head)->hh.tbl->buckets[_ha_bkt],&(add)->hh); \ 202*b1d06847SJean-Baptiste Boric HASH_BLOOM_ADD((head)->hh.tbl,(add)->hh.hashv); \ 203*b1d06847SJean-Baptiste Boric HASH_EMIT_KEY(hh,head,keyptr,keylen_in); \ 204*b1d06847SJean-Baptiste Boric HASH_FSCK(hh,head); \ 205*b1d06847SJean-Baptiste Boric } while(0) 206*b1d06847SJean-Baptiste Boric 207*b1d06847SJean-Baptiste Boric #define HASH_TO_BKT( hashv, num_bkts, bkt ) \ 208*b1d06847SJean-Baptiste Boric do { \ 209*b1d06847SJean-Baptiste Boric bkt = ((hashv) & ((num_bkts) - 1)); \ 210*b1d06847SJean-Baptiste Boric } while(0) 211*b1d06847SJean-Baptiste Boric 212*b1d06847SJean-Baptiste Boric /* delete "delptr" from the hash table. 213*b1d06847SJean-Baptiste Boric * "the usual" patch-up process for the app-order doubly-linked-list. 214*b1d06847SJean-Baptiste Boric * The use of _hd_hh_del below deserves special explanation. 215*b1d06847SJean-Baptiste Boric * These used to be expressed using (delptr) but that led to a bug 216*b1d06847SJean-Baptiste Boric * if someone used the same symbol for the head and deletee, like 217*b1d06847SJean-Baptiste Boric * HASH_DELETE(hh,users,users); 218*b1d06847SJean-Baptiste Boric * We want that to work, but by changing the head (users) below 219*b1d06847SJean-Baptiste Boric * we were forfeiting our ability to further refer to the deletee (users) 220*b1d06847SJean-Baptiste Boric * in the patch-up process. Solution: use scratch space to 221*b1d06847SJean-Baptiste Boric * copy the deletee pointer, then the latter references are via that 222*b1d06847SJean-Baptiste Boric * scratch pointer rather than through the repointed (users) symbol. 223*b1d06847SJean-Baptiste Boric */ 224*b1d06847SJean-Baptiste Boric #define HASH_DELETE(hh,head,delptr) \ 225*b1d06847SJean-Baptiste Boric do { \ 226*b1d06847SJean-Baptiste Boric struct UT_hash_handle *_hd_hh_del; \ 227*b1d06847SJean-Baptiste Boric if ( ((delptr)->hh.prev == NULL) && ((delptr)->hh.next == NULL) ) { \ 228*b1d06847SJean-Baptiste Boric uthash_free((head)->hh.tbl->buckets, \ 229*b1d06847SJean-Baptiste Boric (head)->hh.tbl->num_buckets*sizeof(struct UT_hash_bucket) ); \ 230*b1d06847SJean-Baptiste Boric HASH_BLOOM_FREE((head)->hh.tbl); \ 231*b1d06847SJean-Baptiste Boric uthash_free((head)->hh.tbl, sizeof(UT_hash_table)); \ 232*b1d06847SJean-Baptiste Boric head = NULL; \ 233*b1d06847SJean-Baptiste Boric } else { \ 234*b1d06847SJean-Baptiste Boric unsigned _hd_bkt; \ 235*b1d06847SJean-Baptiste Boric _hd_hh_del = &((delptr)->hh); \ 236*b1d06847SJean-Baptiste Boric if ((delptr) == ELMT_FROM_HH((head)->hh.tbl,(head)->hh.tbl->tail)) { \ 237*b1d06847SJean-Baptiste Boric (head)->hh.tbl->tail = \ 238*b1d06847SJean-Baptiste Boric (UT_hash_handle*)((ptrdiff_t)((delptr)->hh.prev) + \ 239*b1d06847SJean-Baptiste Boric (head)->hh.tbl->hho); \ 240*b1d06847SJean-Baptiste Boric } \ 241*b1d06847SJean-Baptiste Boric if ((delptr)->hh.prev) { \ 242*b1d06847SJean-Baptiste Boric ((UT_hash_handle*)((ptrdiff_t)((delptr)->hh.prev) + \ 243*b1d06847SJean-Baptiste Boric (head)->hh.tbl->hho))->next = (delptr)->hh.next; \ 244*b1d06847SJean-Baptiste Boric } else { \ 245*b1d06847SJean-Baptiste Boric DECLTYPE_ASSIGN(head,(delptr)->hh.next); \ 246*b1d06847SJean-Baptiste Boric } \ 247*b1d06847SJean-Baptiste Boric if (_hd_hh_del->next) { \ 248*b1d06847SJean-Baptiste Boric ((UT_hash_handle*)((ptrdiff_t)_hd_hh_del->next + \ 249*b1d06847SJean-Baptiste Boric (head)->hh.tbl->hho))->prev = \ 250*b1d06847SJean-Baptiste Boric _hd_hh_del->prev; \ 251*b1d06847SJean-Baptiste Boric } \ 252*b1d06847SJean-Baptiste Boric HASH_TO_BKT( _hd_hh_del->hashv, (head)->hh.tbl->num_buckets, _hd_bkt); \ 253*b1d06847SJean-Baptiste Boric HASH_DEL_IN_BKT(hh,(head)->hh.tbl->buckets[_hd_bkt], _hd_hh_del); \ 254*b1d06847SJean-Baptiste Boric (head)->hh.tbl->num_items--; \ 255*b1d06847SJean-Baptiste Boric } \ 256*b1d06847SJean-Baptiste Boric HASH_FSCK(hh,head); \ 257*b1d06847SJean-Baptiste Boric } while (0) 258*b1d06847SJean-Baptiste Boric 259*b1d06847SJean-Baptiste Boric 260*b1d06847SJean-Baptiste Boric /* convenience forms of HASH_FIND/HASH_ADD/HASH_DEL */ 261*b1d06847SJean-Baptiste Boric #define HASH_FIND_STR(head,findstr,out) \ 262*b1d06847SJean-Baptiste Boric HASH_FIND(hh,head,findstr,(unsigned)strlen(findstr),out) 263*b1d06847SJean-Baptiste Boric #define HASH_ADD_STR(head,strfield,add) \ 264*b1d06847SJean-Baptiste Boric HASH_ADD(hh,head,strfield[0],strlen(add->strfield),add) 265*b1d06847SJean-Baptiste Boric #define HASH_REPLACE_STR(head,strfield,add,replaced) \ 266*b1d06847SJean-Baptiste Boric HASH_REPLACE(hh,head,strfield[0],(unsigned)strlen(add->strfield),add,replaced) 267*b1d06847SJean-Baptiste Boric #define HASH_FIND_INT(head,findint,out) \ 268*b1d06847SJean-Baptiste Boric HASH_FIND(hh,head,findint,sizeof(int),out) 269*b1d06847SJean-Baptiste Boric #define HASH_ADD_INT(head,intfield,add) \ 270*b1d06847SJean-Baptiste Boric HASH_ADD(hh,head,intfield,sizeof(int),add) 271*b1d06847SJean-Baptiste Boric #define HASH_REPLACE_INT(head,intfield,add,replaced) \ 272*b1d06847SJean-Baptiste Boric HASH_REPLACE(hh,head,intfield,sizeof(int),add,replaced) 273*b1d06847SJean-Baptiste Boric #define HASH_FIND_PTR(head,findptr,out) \ 274*b1d06847SJean-Baptiste Boric HASH_FIND(hh,head,findptr,sizeof(void *),out) 275*b1d06847SJean-Baptiste Boric #define HASH_ADD_PTR(head,ptrfield,add) \ 276*b1d06847SJean-Baptiste Boric HASH_ADD(hh,head,ptrfield,sizeof(void *),add) 277*b1d06847SJean-Baptiste Boric #define HASH_REPLACE_PTR(head,ptrfield,add,replaced) \ 278*b1d06847SJean-Baptiste Boric HASH_REPLACE(hh,head,ptrfield,sizeof(void *),add,replaced) 279*b1d06847SJean-Baptiste Boric #define HASH_DEL(head,delptr) \ 280*b1d06847SJean-Baptiste Boric HASH_DELETE(hh,head,delptr) 281*b1d06847SJean-Baptiste Boric 282*b1d06847SJean-Baptiste Boric /* HASH_FSCK checks hash integrity on every add/delete when HASH_DEBUG is defined. 283*b1d06847SJean-Baptiste Boric * This is for uthash developer only; it compiles away if HASH_DEBUG isn't defined. 284*b1d06847SJean-Baptiste Boric */ 285*b1d06847SJean-Baptiste Boric #ifdef HASH_DEBUG 286*b1d06847SJean-Baptiste Boric #define HASH_OOPS(...) do { fprintf(stderr,__VA_ARGS__); exit(-1); } while (0) 287*b1d06847SJean-Baptiste Boric #define HASH_FSCK(hh,head) \ 288*b1d06847SJean-Baptiste Boric do { \ 289*b1d06847SJean-Baptiste Boric struct UT_hash_handle *_thh; \ 290*b1d06847SJean-Baptiste Boric if (head) { \ 291*b1d06847SJean-Baptiste Boric unsigned _bkt_i; \ 292*b1d06847SJean-Baptiste Boric unsigned _count; \ 293*b1d06847SJean-Baptiste Boric char *_prev; \ 294*b1d06847SJean-Baptiste Boric _count = 0; \ 295*b1d06847SJean-Baptiste Boric for( _bkt_i = 0; _bkt_i < (head)->hh.tbl->num_buckets; _bkt_i++) { \ 296*b1d06847SJean-Baptiste Boric unsigned _bkt_count = 0; \ 297*b1d06847SJean-Baptiste Boric _thh = (head)->hh.tbl->buckets[_bkt_i].hh_head; \ 298*b1d06847SJean-Baptiste Boric _prev = NULL; \ 299*b1d06847SJean-Baptiste Boric while (_thh) { \ 300*b1d06847SJean-Baptiste Boric if (_prev != (char*)(_thh->hh_prev)) { \ 301*b1d06847SJean-Baptiste Boric HASH_OOPS("invalid hh_prev %p, actual %p\n", \ 302*b1d06847SJean-Baptiste Boric _thh->hh_prev, _prev ); \ 303*b1d06847SJean-Baptiste Boric } \ 304*b1d06847SJean-Baptiste Boric _bkt_count++; \ 305*b1d06847SJean-Baptiste Boric _prev = (char*)(_thh); \ 306*b1d06847SJean-Baptiste Boric _thh = _thh->hh_next; \ 307*b1d06847SJean-Baptiste Boric } \ 308*b1d06847SJean-Baptiste Boric _count += _bkt_count; \ 309*b1d06847SJean-Baptiste Boric if ((head)->hh.tbl->buckets[_bkt_i].count != _bkt_count) { \ 310*b1d06847SJean-Baptiste Boric HASH_OOPS("invalid bucket count %u, actual %u\n", \ 311*b1d06847SJean-Baptiste Boric (head)->hh.tbl->buckets[_bkt_i].count, _bkt_count); \ 312*b1d06847SJean-Baptiste Boric } \ 313*b1d06847SJean-Baptiste Boric } \ 314*b1d06847SJean-Baptiste Boric if (_count != (head)->hh.tbl->num_items) { \ 315*b1d06847SJean-Baptiste Boric HASH_OOPS("invalid hh item count %u, actual %u\n", \ 316*b1d06847SJean-Baptiste Boric (head)->hh.tbl->num_items, _count ); \ 317*b1d06847SJean-Baptiste Boric } \ 318*b1d06847SJean-Baptiste Boric /* traverse hh in app order; check next/prev integrity, count */ \ 319*b1d06847SJean-Baptiste Boric _count = 0; \ 320*b1d06847SJean-Baptiste Boric _prev = NULL; \ 321*b1d06847SJean-Baptiste Boric _thh = &(head)->hh; \ 322*b1d06847SJean-Baptiste Boric while (_thh) { \ 323*b1d06847SJean-Baptiste Boric _count++; \ 324*b1d06847SJean-Baptiste Boric if (_prev !=(char*)(_thh->prev)) { \ 325*b1d06847SJean-Baptiste Boric HASH_OOPS("invalid prev %p, actual %p\n", \ 326*b1d06847SJean-Baptiste Boric _thh->prev, _prev ); \ 327*b1d06847SJean-Baptiste Boric } \ 328*b1d06847SJean-Baptiste Boric _prev = (char*)ELMT_FROM_HH((head)->hh.tbl, _thh); \ 329*b1d06847SJean-Baptiste Boric _thh = ( _thh->next ? (UT_hash_handle*)((char*)(_thh->next) + \ 330*b1d06847SJean-Baptiste Boric (head)->hh.tbl->hho) : NULL ); \ 331*b1d06847SJean-Baptiste Boric } \ 332*b1d06847SJean-Baptiste Boric if (_count != (head)->hh.tbl->num_items) { \ 333*b1d06847SJean-Baptiste Boric HASH_OOPS("invalid app item count %u, actual %u\n", \ 334*b1d06847SJean-Baptiste Boric (head)->hh.tbl->num_items, _count ); \ 335*b1d06847SJean-Baptiste Boric } \ 336*b1d06847SJean-Baptiste Boric } \ 337*b1d06847SJean-Baptiste Boric } while (0) 338*b1d06847SJean-Baptiste Boric #else 339*b1d06847SJean-Baptiste Boric #define HASH_FSCK(hh,head) 340*b1d06847SJean-Baptiste Boric #endif 341*b1d06847SJean-Baptiste Boric 342*b1d06847SJean-Baptiste Boric /* When compiled with -DHASH_EMIT_KEYS, length-prefixed keys are emitted to 343*b1d06847SJean-Baptiste Boric * the descriptor to which this macro is defined for tuning the hash function. 344*b1d06847SJean-Baptiste Boric * The app can #include <unistd.h> to get the prototype for write(2). */ 345*b1d06847SJean-Baptiste Boric #ifdef HASH_EMIT_KEYS 346*b1d06847SJean-Baptiste Boric #define HASH_EMIT_KEY(hh,head,keyptr,fieldlen) \ 347*b1d06847SJean-Baptiste Boric do { \ 348*b1d06847SJean-Baptiste Boric unsigned _klen = fieldlen; \ 349*b1d06847SJean-Baptiste Boric write(HASH_EMIT_KEYS, &_klen, sizeof(_klen)); \ 350*b1d06847SJean-Baptiste Boric write(HASH_EMIT_KEYS, keyptr, fieldlen); \ 351*b1d06847SJean-Baptiste Boric } while (0) 352*b1d06847SJean-Baptiste Boric #else 353*b1d06847SJean-Baptiste Boric #define HASH_EMIT_KEY(hh,head,keyptr,fieldlen) 354*b1d06847SJean-Baptiste Boric #endif 355*b1d06847SJean-Baptiste Boric 356*b1d06847SJean-Baptiste Boric /* default to Jenkin's hash unless overridden e.g. DHASH_FUNCTION=HASH_SAX */ 357*b1d06847SJean-Baptiste Boric #ifdef HASH_FUNCTION 358*b1d06847SJean-Baptiste Boric #define HASH_FCN HASH_FUNCTION 359*b1d06847SJean-Baptiste Boric #else 360*b1d06847SJean-Baptiste Boric #define HASH_FCN HASH_JEN 361*b1d06847SJean-Baptiste Boric #endif 362*b1d06847SJean-Baptiste Boric 363*b1d06847SJean-Baptiste Boric /* The Bernstein hash function, used in Perl prior to v5.6. Note (x<<5+x)=x*33. */ 364*b1d06847SJean-Baptiste Boric #define HASH_BER(key,keylen,num_bkts,hashv,bkt) \ 365*b1d06847SJean-Baptiste Boric do { \ 366*b1d06847SJean-Baptiste Boric unsigned _hb_keylen=keylen; \ 367*b1d06847SJean-Baptiste Boric char *_hb_key=(char*)(key); \ 368*b1d06847SJean-Baptiste Boric (hashv) = 0; \ 369*b1d06847SJean-Baptiste Boric while (_hb_keylen--) { (hashv) = (((hashv) << 5) + (hashv)) + *_hb_key++; } \ 370*b1d06847SJean-Baptiste Boric bkt = (hashv) & (num_bkts-1); \ 371*b1d06847SJean-Baptiste Boric } while (0) 372*b1d06847SJean-Baptiste Boric 373*b1d06847SJean-Baptiste Boric 374*b1d06847SJean-Baptiste Boric /* SAX/FNV/OAT/JEN hash functions are macro variants of those listed at 375*b1d06847SJean-Baptiste Boric * http://eternallyconfuzzled.com/tuts/algorithms/jsw_tut_hashing.aspx */ 376*b1d06847SJean-Baptiste Boric #define HASH_SAX(key,keylen,num_bkts,hashv,bkt) \ 377*b1d06847SJean-Baptiste Boric do { \ 378*b1d06847SJean-Baptiste Boric unsigned _sx_i; \ 379*b1d06847SJean-Baptiste Boric char *_hs_key=(char*)(key); \ 380*b1d06847SJean-Baptiste Boric hashv = 0; \ 381*b1d06847SJean-Baptiste Boric for(_sx_i=0; _sx_i < keylen; _sx_i++) \ 382*b1d06847SJean-Baptiste Boric hashv ^= (hashv << 5) + (hashv >> 2) + _hs_key[_sx_i]; \ 383*b1d06847SJean-Baptiste Boric bkt = hashv & (num_bkts-1); \ 384*b1d06847SJean-Baptiste Boric } while (0) 385*b1d06847SJean-Baptiste Boric /* FNV-1a variation */ 386*b1d06847SJean-Baptiste Boric #define HASH_FNV(key,keylen,num_bkts,hashv,bkt) \ 387*b1d06847SJean-Baptiste Boric do { \ 388*b1d06847SJean-Baptiste Boric unsigned _fn_i; \ 389*b1d06847SJean-Baptiste Boric char *_hf_key=(char*)(key); \ 390*b1d06847SJean-Baptiste Boric hashv = 2166136261UL; \ 391*b1d06847SJean-Baptiste Boric for(_fn_i=0; _fn_i < keylen; _fn_i++) { \ 392*b1d06847SJean-Baptiste Boric hashv = hashv ^ _hf_key[_fn_i]; \ 393*b1d06847SJean-Baptiste Boric hashv = hashv * 16777619; \ 394*b1d06847SJean-Baptiste Boric } \ 395*b1d06847SJean-Baptiste Boric bkt = hashv & (num_bkts-1); \ 396*b1d06847SJean-Baptiste Boric } while(0) 397*b1d06847SJean-Baptiste Boric 398*b1d06847SJean-Baptiste Boric #define HASH_OAT(key,keylen,num_bkts,hashv,bkt) \ 399*b1d06847SJean-Baptiste Boric do { \ 400*b1d06847SJean-Baptiste Boric unsigned _ho_i; \ 401*b1d06847SJean-Baptiste Boric char *_ho_key=(char*)(key); \ 402*b1d06847SJean-Baptiste Boric hashv = 0; \ 403*b1d06847SJean-Baptiste Boric for(_ho_i=0; _ho_i < keylen; _ho_i++) { \ 404*b1d06847SJean-Baptiste Boric hashv += _ho_key[_ho_i]; \ 405*b1d06847SJean-Baptiste Boric hashv += (hashv << 10); \ 406*b1d06847SJean-Baptiste Boric hashv ^= (hashv >> 6); \ 407*b1d06847SJean-Baptiste Boric } \ 408*b1d06847SJean-Baptiste Boric hashv += (hashv << 3); \ 409*b1d06847SJean-Baptiste Boric hashv ^= (hashv >> 11); \ 410*b1d06847SJean-Baptiste Boric hashv += (hashv << 15); \ 411*b1d06847SJean-Baptiste Boric bkt = hashv & (num_bkts-1); \ 412*b1d06847SJean-Baptiste Boric } while(0) 413*b1d06847SJean-Baptiste Boric 414*b1d06847SJean-Baptiste Boric #define HASH_JEN_MIX(a,b,c) \ 415*b1d06847SJean-Baptiste Boric do { \ 416*b1d06847SJean-Baptiste Boric a -= b; a -= c; a ^= ( c >> 13 ); \ 417*b1d06847SJean-Baptiste Boric b -= c; b -= a; b ^= ( a << 8 ); \ 418*b1d06847SJean-Baptiste Boric c -= a; c -= b; c ^= ( b >> 13 ); \ 419*b1d06847SJean-Baptiste Boric a -= b; a -= c; a ^= ( c >> 12 ); \ 420*b1d06847SJean-Baptiste Boric b -= c; b -= a; b ^= ( a << 16 ); \ 421*b1d06847SJean-Baptiste Boric c -= a; c -= b; c ^= ( b >> 5 ); \ 422*b1d06847SJean-Baptiste Boric a -= b; a -= c; a ^= ( c >> 3 ); \ 423*b1d06847SJean-Baptiste Boric b -= c; b -= a; b ^= ( a << 10 ); \ 424*b1d06847SJean-Baptiste Boric c -= a; c -= b; c ^= ( b >> 15 ); \ 425*b1d06847SJean-Baptiste Boric } while (0) 426*b1d06847SJean-Baptiste Boric 427*b1d06847SJean-Baptiste Boric #define HASH_JEN(key,keylen,num_bkts,hashv,bkt) \ 428*b1d06847SJean-Baptiste Boric do { \ 429*b1d06847SJean-Baptiste Boric unsigned _hj_i,_hj_j,_hj_k; \ 430*b1d06847SJean-Baptiste Boric unsigned char *_hj_key=(unsigned char*)(key); \ 431*b1d06847SJean-Baptiste Boric hashv = 0xfeedbeef; \ 432*b1d06847SJean-Baptiste Boric _hj_i = _hj_j = 0x9e3779b9; \ 433*b1d06847SJean-Baptiste Boric _hj_k = (unsigned)(keylen); \ 434*b1d06847SJean-Baptiste Boric while (_hj_k >= 12) { \ 435*b1d06847SJean-Baptiste Boric _hj_i += (_hj_key[0] + ( (unsigned)_hj_key[1] << 8 ) \ 436*b1d06847SJean-Baptiste Boric + ( (unsigned)_hj_key[2] << 16 ) \ 437*b1d06847SJean-Baptiste Boric + ( (unsigned)_hj_key[3] << 24 ) ); \ 438*b1d06847SJean-Baptiste Boric _hj_j += (_hj_key[4] + ( (unsigned)_hj_key[5] << 8 ) \ 439*b1d06847SJean-Baptiste Boric + ( (unsigned)_hj_key[6] << 16 ) \ 440*b1d06847SJean-Baptiste Boric + ( (unsigned)_hj_key[7] << 24 ) ); \ 441*b1d06847SJean-Baptiste Boric hashv += (_hj_key[8] + ( (unsigned)_hj_key[9] << 8 ) \ 442*b1d06847SJean-Baptiste Boric + ( (unsigned)_hj_key[10] << 16 ) \ 443*b1d06847SJean-Baptiste Boric + ( (unsigned)_hj_key[11] << 24 ) ); \ 444*b1d06847SJean-Baptiste Boric \ 445*b1d06847SJean-Baptiste Boric HASH_JEN_MIX(_hj_i, _hj_j, hashv); \ 446*b1d06847SJean-Baptiste Boric \ 447*b1d06847SJean-Baptiste Boric _hj_key += 12; \ 448*b1d06847SJean-Baptiste Boric _hj_k -= 12; \ 449*b1d06847SJean-Baptiste Boric } \ 450*b1d06847SJean-Baptiste Boric hashv += keylen; \ 451*b1d06847SJean-Baptiste Boric switch ( _hj_k ) { \ 452*b1d06847SJean-Baptiste Boric case 11: hashv += ( (unsigned)_hj_key[10] << 24 ); \ 453*b1d06847SJean-Baptiste Boric case 10: hashv += ( (unsigned)_hj_key[9] << 16 ); \ 454*b1d06847SJean-Baptiste Boric case 9: hashv += ( (unsigned)_hj_key[8] << 8 ); \ 455*b1d06847SJean-Baptiste Boric case 8: _hj_j += ( (unsigned)_hj_key[7] << 24 ); \ 456*b1d06847SJean-Baptiste Boric case 7: _hj_j += ( (unsigned)_hj_key[6] << 16 ); \ 457*b1d06847SJean-Baptiste Boric case 6: _hj_j += ( (unsigned)_hj_key[5] << 8 ); \ 458*b1d06847SJean-Baptiste Boric case 5: _hj_j += _hj_key[4]; \ 459*b1d06847SJean-Baptiste Boric case 4: _hj_i += ( (unsigned)_hj_key[3] << 24 ); \ 460*b1d06847SJean-Baptiste Boric case 3: _hj_i += ( (unsigned)_hj_key[2] << 16 ); \ 461*b1d06847SJean-Baptiste Boric case 2: _hj_i += ( (unsigned)_hj_key[1] << 8 ); \ 462*b1d06847SJean-Baptiste Boric case 1: _hj_i += _hj_key[0]; \ 463*b1d06847SJean-Baptiste Boric } \ 464*b1d06847SJean-Baptiste Boric HASH_JEN_MIX(_hj_i, _hj_j, hashv); \ 465*b1d06847SJean-Baptiste Boric bkt = hashv & (num_bkts-1); \ 466*b1d06847SJean-Baptiste Boric } while(0) 467*b1d06847SJean-Baptiste Boric 468*b1d06847SJean-Baptiste Boric /* The Paul Hsieh hash function */ 469*b1d06847SJean-Baptiste Boric #undef get16bits 470*b1d06847SJean-Baptiste Boric #if (defined(__GNUC__) && defined(__i386__)) || defined(__WATCOMC__) \ 471*b1d06847SJean-Baptiste Boric || defined(_MSC_VER) || defined (__BORLANDC__) || defined (__TURBOC__) 472*b1d06847SJean-Baptiste Boric #define get16bits(d) (*((const uint16_t *) (d))) 473*b1d06847SJean-Baptiste Boric #endif 474*b1d06847SJean-Baptiste Boric 475*b1d06847SJean-Baptiste Boric #if !defined (get16bits) 476*b1d06847SJean-Baptiste Boric #define get16bits(d) ((((uint32_t)(((const uint8_t *)(d))[1])) << 8) \ 477*b1d06847SJean-Baptiste Boric +(uint32_t)(((const uint8_t *)(d))[0]) ) 478*b1d06847SJean-Baptiste Boric #endif 479*b1d06847SJean-Baptiste Boric #define HASH_SFH(key,keylen,num_bkts,hashv,bkt) \ 480*b1d06847SJean-Baptiste Boric do { \ 481*b1d06847SJean-Baptiste Boric unsigned char *_sfh_key=(unsigned char*)(key); \ 482*b1d06847SJean-Baptiste Boric uint32_t _sfh_tmp, _sfh_len = keylen; \ 483*b1d06847SJean-Baptiste Boric \ 484*b1d06847SJean-Baptiste Boric int _sfh_rem = _sfh_len & 3; \ 485*b1d06847SJean-Baptiste Boric _sfh_len >>= 2; \ 486*b1d06847SJean-Baptiste Boric hashv = 0xcafebabe; \ 487*b1d06847SJean-Baptiste Boric \ 488*b1d06847SJean-Baptiste Boric /* Main loop */ \ 489*b1d06847SJean-Baptiste Boric for (;_sfh_len > 0; _sfh_len--) { \ 490*b1d06847SJean-Baptiste Boric hashv += get16bits (_sfh_key); \ 491*b1d06847SJean-Baptiste Boric _sfh_tmp = (uint32_t)(get16bits (_sfh_key+2)) << 11 ^ hashv; \ 492*b1d06847SJean-Baptiste Boric hashv = (hashv << 16) ^ _sfh_tmp; \ 493*b1d06847SJean-Baptiste Boric _sfh_key += 2*sizeof (uint16_t); \ 494*b1d06847SJean-Baptiste Boric hashv += hashv >> 11; \ 495*b1d06847SJean-Baptiste Boric } \ 496*b1d06847SJean-Baptiste Boric \ 497*b1d06847SJean-Baptiste Boric /* Handle end cases */ \ 498*b1d06847SJean-Baptiste Boric switch (_sfh_rem) { \ 499*b1d06847SJean-Baptiste Boric case 3: hashv += get16bits (_sfh_key); \ 500*b1d06847SJean-Baptiste Boric hashv ^= hashv << 16; \ 501*b1d06847SJean-Baptiste Boric hashv ^= (uint32_t)(_sfh_key[sizeof (uint16_t)] << 18); \ 502*b1d06847SJean-Baptiste Boric hashv += hashv >> 11; \ 503*b1d06847SJean-Baptiste Boric break; \ 504*b1d06847SJean-Baptiste Boric case 2: hashv += get16bits (_sfh_key); \ 505*b1d06847SJean-Baptiste Boric hashv ^= hashv << 11; \ 506*b1d06847SJean-Baptiste Boric hashv += hashv >> 17; \ 507*b1d06847SJean-Baptiste Boric break; \ 508*b1d06847SJean-Baptiste Boric case 1: hashv += *_sfh_key; \ 509*b1d06847SJean-Baptiste Boric hashv ^= hashv << 10; \ 510*b1d06847SJean-Baptiste Boric hashv += hashv >> 1; \ 511*b1d06847SJean-Baptiste Boric } \ 512*b1d06847SJean-Baptiste Boric \ 513*b1d06847SJean-Baptiste Boric /* Force "avalanching" of final 127 bits */ \ 514*b1d06847SJean-Baptiste Boric hashv ^= hashv << 3; \ 515*b1d06847SJean-Baptiste Boric hashv += hashv >> 5; \ 516*b1d06847SJean-Baptiste Boric hashv ^= hashv << 4; \ 517*b1d06847SJean-Baptiste Boric hashv += hashv >> 17; \ 518*b1d06847SJean-Baptiste Boric hashv ^= hashv << 25; \ 519*b1d06847SJean-Baptiste Boric hashv += hashv >> 6; \ 520*b1d06847SJean-Baptiste Boric bkt = hashv & (num_bkts-1); \ 521*b1d06847SJean-Baptiste Boric } while(0) 522*b1d06847SJean-Baptiste Boric 523*b1d06847SJean-Baptiste Boric #ifdef HASH_USING_NO_STRICT_ALIASING 524*b1d06847SJean-Baptiste Boric /* The MurmurHash exploits some CPU's (x86,x86_64) tolerance for unaligned reads. 525*b1d06847SJean-Baptiste Boric * For other types of CPU's (e.g. Sparc) an unaligned read causes a bus error. 526*b1d06847SJean-Baptiste Boric * MurmurHash uses the faster approach only on CPU's where we know it's safe. 527*b1d06847SJean-Baptiste Boric * 528*b1d06847SJean-Baptiste Boric * Note the preprocessor built-in defines can be emitted using: 529*b1d06847SJean-Baptiste Boric * 530*b1d06847SJean-Baptiste Boric * gcc -m64 -dM -E - < /dev/null (on gcc) 531*b1d06847SJean-Baptiste Boric * cc -## a.c (where a.c is a simple test file) (Sun Studio) 532*b1d06847SJean-Baptiste Boric */ 533*b1d06847SJean-Baptiste Boric #if (defined(__i386__) || defined(__x86_64__) || defined(_M_IX86)) 534*b1d06847SJean-Baptiste Boric #define MUR_GETBLOCK(p,i) p[i] 535*b1d06847SJean-Baptiste Boric #else /* non intel */ 536*b1d06847SJean-Baptiste Boric #define MUR_PLUS0_ALIGNED(p) (((unsigned long)p & 0x3) == 0) 537*b1d06847SJean-Baptiste Boric #define MUR_PLUS1_ALIGNED(p) (((unsigned long)p & 0x3) == 1) 538*b1d06847SJean-Baptiste Boric #define MUR_PLUS2_ALIGNED(p) (((unsigned long)p & 0x3) == 2) 539*b1d06847SJean-Baptiste Boric #define MUR_PLUS3_ALIGNED(p) (((unsigned long)p & 0x3) == 3) 540*b1d06847SJean-Baptiste Boric #define WP(p) ((uint32_t*)((unsigned long)(p) & ~3UL)) 541*b1d06847SJean-Baptiste Boric #if (defined(__BIG_ENDIAN__) || defined(SPARC) || defined(__ppc__) || defined(__ppc64__)) 542*b1d06847SJean-Baptiste Boric #define MUR_THREE_ONE(p) ((((*WP(p))&0x00ffffff) << 8) | (((*(WP(p)+1))&0xff000000) >> 24)) 543*b1d06847SJean-Baptiste Boric #define MUR_TWO_TWO(p) ((((*WP(p))&0x0000ffff) <<16) | (((*(WP(p)+1))&0xffff0000) >> 16)) 544*b1d06847SJean-Baptiste Boric #define MUR_ONE_THREE(p) ((((*WP(p))&0x000000ff) <<24) | (((*(WP(p)+1))&0xffffff00) >> 8)) 545*b1d06847SJean-Baptiste Boric #else /* assume little endian non-intel */ 546*b1d06847SJean-Baptiste Boric #define MUR_THREE_ONE(p) ((((*WP(p))&0xffffff00) >> 8) | (((*(WP(p)+1))&0x000000ff) << 24)) 547*b1d06847SJean-Baptiste Boric #define MUR_TWO_TWO(p) ((((*WP(p))&0xffff0000) >>16) | (((*(WP(p)+1))&0x0000ffff) << 16)) 548*b1d06847SJean-Baptiste Boric #define MUR_ONE_THREE(p) ((((*WP(p))&0xff000000) >>24) | (((*(WP(p)+1))&0x00ffffff) << 8)) 549*b1d06847SJean-Baptiste Boric #endif 550*b1d06847SJean-Baptiste Boric #define MUR_GETBLOCK(p,i) (MUR_PLUS0_ALIGNED(p) ? ((p)[i]) : \ 551*b1d06847SJean-Baptiste Boric (MUR_PLUS1_ALIGNED(p) ? MUR_THREE_ONE(p) : \ 552*b1d06847SJean-Baptiste Boric (MUR_PLUS2_ALIGNED(p) ? MUR_TWO_TWO(p) : \ 553*b1d06847SJean-Baptiste Boric MUR_ONE_THREE(p)))) 554*b1d06847SJean-Baptiste Boric #endif 555*b1d06847SJean-Baptiste Boric #define MUR_ROTL32(x,r) (((x) << (r)) | ((x) >> (32 - (r)))) 556*b1d06847SJean-Baptiste Boric #define MUR_FMIX(_h) \ 557*b1d06847SJean-Baptiste Boric do { \ 558*b1d06847SJean-Baptiste Boric _h ^= _h >> 16; \ 559*b1d06847SJean-Baptiste Boric _h *= 0x85ebca6b; \ 560*b1d06847SJean-Baptiste Boric _h ^= _h >> 13; \ 561*b1d06847SJean-Baptiste Boric _h *= 0xc2b2ae35l; \ 562*b1d06847SJean-Baptiste Boric _h ^= _h >> 16; \ 563*b1d06847SJean-Baptiste Boric } while(0) 564*b1d06847SJean-Baptiste Boric 565*b1d06847SJean-Baptiste Boric #define HASH_MUR(key,keylen,num_bkts,hashv,bkt) \ 566*b1d06847SJean-Baptiste Boric do { \ 567*b1d06847SJean-Baptiste Boric const uint8_t *_mur_data = (const uint8_t*)(key); \ 568*b1d06847SJean-Baptiste Boric const int _mur_nblocks = (keylen) / 4; \ 569*b1d06847SJean-Baptiste Boric uint32_t _mur_h1 = 0xf88D5353; \ 570*b1d06847SJean-Baptiste Boric uint32_t _mur_c1 = 0xcc9e2d51; \ 571*b1d06847SJean-Baptiste Boric uint32_t _mur_c2 = 0x1b873593; \ 572*b1d06847SJean-Baptiste Boric uint32_t _mur_k1 = 0; \ 573*b1d06847SJean-Baptiste Boric const uint8_t *_mur_tail; \ 574*b1d06847SJean-Baptiste Boric const uint32_t *_mur_blocks = (const uint32_t*)(_mur_data+_mur_nblocks*4); \ 575*b1d06847SJean-Baptiste Boric int _mur_i; \ 576*b1d06847SJean-Baptiste Boric for(_mur_i = -_mur_nblocks; _mur_i; _mur_i++) { \ 577*b1d06847SJean-Baptiste Boric _mur_k1 = MUR_GETBLOCK(_mur_blocks,_mur_i); \ 578*b1d06847SJean-Baptiste Boric _mur_k1 *= _mur_c1; \ 579*b1d06847SJean-Baptiste Boric _mur_k1 = MUR_ROTL32(_mur_k1,15); \ 580*b1d06847SJean-Baptiste Boric _mur_k1 *= _mur_c2; \ 581*b1d06847SJean-Baptiste Boric \ 582*b1d06847SJean-Baptiste Boric _mur_h1 ^= _mur_k1; \ 583*b1d06847SJean-Baptiste Boric _mur_h1 = MUR_ROTL32(_mur_h1,13); \ 584*b1d06847SJean-Baptiste Boric _mur_h1 = _mur_h1*5+0xe6546b64; \ 585*b1d06847SJean-Baptiste Boric } \ 586*b1d06847SJean-Baptiste Boric _mur_tail = (const uint8_t*)(_mur_data + _mur_nblocks*4); \ 587*b1d06847SJean-Baptiste Boric _mur_k1=0; \ 588*b1d06847SJean-Baptiste Boric switch((keylen) & 3) { \ 589*b1d06847SJean-Baptiste Boric case 3: _mur_k1 ^= _mur_tail[2] << 16; \ 590*b1d06847SJean-Baptiste Boric case 2: _mur_k1 ^= _mur_tail[1] << 8; \ 591*b1d06847SJean-Baptiste Boric case 1: _mur_k1 ^= _mur_tail[0]; \ 592*b1d06847SJean-Baptiste Boric _mur_k1 *= _mur_c1; \ 593*b1d06847SJean-Baptiste Boric _mur_k1 = MUR_ROTL32(_mur_k1,15); \ 594*b1d06847SJean-Baptiste Boric _mur_k1 *= _mur_c2; \ 595*b1d06847SJean-Baptiste Boric _mur_h1 ^= _mur_k1; \ 596*b1d06847SJean-Baptiste Boric } \ 597*b1d06847SJean-Baptiste Boric _mur_h1 ^= (keylen); \ 598*b1d06847SJean-Baptiste Boric MUR_FMIX(_mur_h1); \ 599*b1d06847SJean-Baptiste Boric hashv = _mur_h1; \ 600*b1d06847SJean-Baptiste Boric bkt = hashv & (num_bkts-1); \ 601*b1d06847SJean-Baptiste Boric } while(0) 602*b1d06847SJean-Baptiste Boric #endif /* HASH_USING_NO_STRICT_ALIASING */ 603*b1d06847SJean-Baptiste Boric 604*b1d06847SJean-Baptiste Boric /* key comparison function; return 0 if keys equal */ 605*b1d06847SJean-Baptiste Boric #define HASH_KEYCMP(a,b,len) memcmp(a,b,len) 606*b1d06847SJean-Baptiste Boric 607*b1d06847SJean-Baptiste Boric /* iterate over items in a known bucket to find desired item */ 608*b1d06847SJean-Baptiste Boric #define HASH_FIND_IN_BKT(tbl,hh,head,keyptr,keylen_in,out) \ 609*b1d06847SJean-Baptiste Boric do { \ 610*b1d06847SJean-Baptiste Boric if (head.hh_head) DECLTYPE_ASSIGN(out,ELMT_FROM_HH(tbl,head.hh_head)); \ 611*b1d06847SJean-Baptiste Boric else out=NULL; \ 612*b1d06847SJean-Baptiste Boric while (out) { \ 613*b1d06847SJean-Baptiste Boric if ((out)->hh.keylen == keylen_in) { \ 614*b1d06847SJean-Baptiste Boric if ((HASH_KEYCMP((out)->hh.key,keyptr,keylen_in)) == 0) break; \ 615*b1d06847SJean-Baptiste Boric } \ 616*b1d06847SJean-Baptiste Boric if ((out)->hh.hh_next) DECLTYPE_ASSIGN(out,ELMT_FROM_HH(tbl,(out)->hh.hh_next)); \ 617*b1d06847SJean-Baptiste Boric else out = NULL; \ 618*b1d06847SJean-Baptiste Boric } \ 619*b1d06847SJean-Baptiste Boric } while(0) 620*b1d06847SJean-Baptiste Boric 621*b1d06847SJean-Baptiste Boric /* add an item to a bucket */ 622*b1d06847SJean-Baptiste Boric #define HASH_ADD_TO_BKT(head,addhh) \ 623*b1d06847SJean-Baptiste Boric do { \ 624*b1d06847SJean-Baptiste Boric head.count++; \ 625*b1d06847SJean-Baptiste Boric (addhh)->hh_next = head.hh_head; \ 626*b1d06847SJean-Baptiste Boric (addhh)->hh_prev = NULL; \ 627*b1d06847SJean-Baptiste Boric if (head.hh_head) { (head).hh_head->hh_prev = (addhh); } \ 628*b1d06847SJean-Baptiste Boric (head).hh_head=addhh; \ 629*b1d06847SJean-Baptiste Boric if (head.count >= ((head.expand_mult+1) * HASH_BKT_CAPACITY_THRESH) \ 630*b1d06847SJean-Baptiste Boric && (addhh)->tbl->noexpand != 1) { \ 631*b1d06847SJean-Baptiste Boric HASH_EXPAND_BUCKETS((addhh)->tbl); \ 632*b1d06847SJean-Baptiste Boric } \ 633*b1d06847SJean-Baptiste Boric } while(0) 634*b1d06847SJean-Baptiste Boric 635*b1d06847SJean-Baptiste Boric /* remove an item from a given bucket */ 636*b1d06847SJean-Baptiste Boric #define HASH_DEL_IN_BKT(hh,head,hh_del) \ 637*b1d06847SJean-Baptiste Boric (head).count--; \ 638*b1d06847SJean-Baptiste Boric if ((head).hh_head == hh_del) { \ 639*b1d06847SJean-Baptiste Boric (head).hh_head = hh_del->hh_next; \ 640*b1d06847SJean-Baptiste Boric } \ 641*b1d06847SJean-Baptiste Boric if (hh_del->hh_prev) { \ 642*b1d06847SJean-Baptiste Boric hh_del->hh_prev->hh_next = hh_del->hh_next; \ 643*b1d06847SJean-Baptiste Boric } \ 644*b1d06847SJean-Baptiste Boric if (hh_del->hh_next) { \ 645*b1d06847SJean-Baptiste Boric hh_del->hh_next->hh_prev = hh_del->hh_prev; \ 646*b1d06847SJean-Baptiste Boric } 647*b1d06847SJean-Baptiste Boric 648*b1d06847SJean-Baptiste Boric /* Bucket expansion has the effect of doubling the number of buckets 649*b1d06847SJean-Baptiste Boric * and redistributing the items into the new buckets. Ideally the 650*b1d06847SJean-Baptiste Boric * items will distribute more or less evenly into the new buckets 651*b1d06847SJean-Baptiste Boric * (the extent to which this is true is a measure of the quality of 652*b1d06847SJean-Baptiste Boric * the hash function as it applies to the key domain). 653*b1d06847SJean-Baptiste Boric * 654*b1d06847SJean-Baptiste Boric * With the items distributed into more buckets, the chain length 655*b1d06847SJean-Baptiste Boric * (item count) in each bucket is reduced. Thus by expanding buckets 656*b1d06847SJean-Baptiste Boric * the hash keeps a bound on the chain length. This bounded chain 657*b1d06847SJean-Baptiste Boric * length is the essence of how a hash provides constant time lookup. 658*b1d06847SJean-Baptiste Boric * 659*b1d06847SJean-Baptiste Boric * The calculation of tbl->ideal_chain_maxlen below deserves some 660*b1d06847SJean-Baptiste Boric * explanation. First, keep in mind that we're calculating the ideal 661*b1d06847SJean-Baptiste Boric * maximum chain length based on the *new* (doubled) bucket count. 662*b1d06847SJean-Baptiste Boric * In fractions this is just n/b (n=number of items,b=new num buckets). 663*b1d06847SJean-Baptiste Boric * Since the ideal chain length is an integer, we want to calculate 664*b1d06847SJean-Baptiste Boric * ceil(n/b). We don't depend on floating point arithmetic in this 665*b1d06847SJean-Baptiste Boric * hash, so to calculate ceil(n/b) with integers we could write 666*b1d06847SJean-Baptiste Boric * 667*b1d06847SJean-Baptiste Boric * ceil(n/b) = (n/b) + ((n%b)?1:0) 668*b1d06847SJean-Baptiste Boric * 669*b1d06847SJean-Baptiste Boric * and in fact a previous version of this hash did just that. 670*b1d06847SJean-Baptiste Boric * But now we have improved things a bit by recognizing that b is 671*b1d06847SJean-Baptiste Boric * always a power of two. We keep its base 2 log handy (call it lb), 672*b1d06847SJean-Baptiste Boric * so now we can write this with a bit shift and logical AND: 673*b1d06847SJean-Baptiste Boric * 674*b1d06847SJean-Baptiste Boric * ceil(n/b) = (n>>lb) + ( (n & (b-1)) ? 1:0) 675*b1d06847SJean-Baptiste Boric * 676*b1d06847SJean-Baptiste Boric */ 677*b1d06847SJean-Baptiste Boric #define HASH_EXPAND_BUCKETS(tbl) \ 678*b1d06847SJean-Baptiste Boric do { \ 679*b1d06847SJean-Baptiste Boric unsigned _he_bkt; \ 680*b1d06847SJean-Baptiste Boric unsigned _he_bkt_i; \ 681*b1d06847SJean-Baptiste Boric struct UT_hash_handle *_he_thh, *_he_hh_nxt; \ 682*b1d06847SJean-Baptiste Boric UT_hash_bucket *_he_new_buckets, *_he_newbkt; \ 683*b1d06847SJean-Baptiste Boric _he_new_buckets = (UT_hash_bucket*)uthash_malloc( \ 684*b1d06847SJean-Baptiste Boric 2 * tbl->num_buckets * sizeof(struct UT_hash_bucket)); \ 685*b1d06847SJean-Baptiste Boric if (!_he_new_buckets) { uthash_fatal( "out of memory"); } \ 686*b1d06847SJean-Baptiste Boric memset(_he_new_buckets, 0, \ 687*b1d06847SJean-Baptiste Boric 2 * tbl->num_buckets * sizeof(struct UT_hash_bucket)); \ 688*b1d06847SJean-Baptiste Boric tbl->ideal_chain_maxlen = \ 689*b1d06847SJean-Baptiste Boric (tbl->num_items >> (tbl->log2_num_buckets+1)) + \ 690*b1d06847SJean-Baptiste Boric ((tbl->num_items & ((tbl->num_buckets*2)-1)) ? 1 : 0); \ 691*b1d06847SJean-Baptiste Boric tbl->nonideal_items = 0; \ 692*b1d06847SJean-Baptiste Boric for(_he_bkt_i = 0; _he_bkt_i < tbl->num_buckets; _he_bkt_i++) \ 693*b1d06847SJean-Baptiste Boric { \ 694*b1d06847SJean-Baptiste Boric _he_thh = tbl->buckets[ _he_bkt_i ].hh_head; \ 695*b1d06847SJean-Baptiste Boric while (_he_thh) { \ 696*b1d06847SJean-Baptiste Boric _he_hh_nxt = _he_thh->hh_next; \ 697*b1d06847SJean-Baptiste Boric HASH_TO_BKT( _he_thh->hashv, tbl->num_buckets*2, _he_bkt); \ 698*b1d06847SJean-Baptiste Boric _he_newbkt = &(_he_new_buckets[ _he_bkt ]); \ 699*b1d06847SJean-Baptiste Boric if (++(_he_newbkt->count) > tbl->ideal_chain_maxlen) { \ 700*b1d06847SJean-Baptiste Boric tbl->nonideal_items++; \ 701*b1d06847SJean-Baptiste Boric _he_newbkt->expand_mult = _he_newbkt->count / \ 702*b1d06847SJean-Baptiste Boric tbl->ideal_chain_maxlen; \ 703*b1d06847SJean-Baptiste Boric } \ 704*b1d06847SJean-Baptiste Boric _he_thh->hh_prev = NULL; \ 705*b1d06847SJean-Baptiste Boric _he_thh->hh_next = _he_newbkt->hh_head; \ 706*b1d06847SJean-Baptiste Boric if (_he_newbkt->hh_head) _he_newbkt->hh_head->hh_prev = \ 707*b1d06847SJean-Baptiste Boric _he_thh; \ 708*b1d06847SJean-Baptiste Boric _he_newbkt->hh_head = _he_thh; \ 709*b1d06847SJean-Baptiste Boric _he_thh = _he_hh_nxt; \ 710*b1d06847SJean-Baptiste Boric } \ 711*b1d06847SJean-Baptiste Boric } \ 712*b1d06847SJean-Baptiste Boric uthash_free( tbl->buckets, tbl->num_buckets*sizeof(struct UT_hash_bucket) ); \ 713*b1d06847SJean-Baptiste Boric tbl->num_buckets *= 2; \ 714*b1d06847SJean-Baptiste Boric tbl->log2_num_buckets++; \ 715*b1d06847SJean-Baptiste Boric tbl->buckets = _he_new_buckets; \ 716*b1d06847SJean-Baptiste Boric tbl->ineff_expands = (tbl->nonideal_items > (tbl->num_items >> 1)) ? \ 717*b1d06847SJean-Baptiste Boric (tbl->ineff_expands+1) : 0; \ 718*b1d06847SJean-Baptiste Boric if (tbl->ineff_expands > 1) { \ 719*b1d06847SJean-Baptiste Boric tbl->noexpand=1; \ 720*b1d06847SJean-Baptiste Boric uthash_noexpand_fyi(tbl); \ 721*b1d06847SJean-Baptiste Boric } \ 722*b1d06847SJean-Baptiste Boric uthash_expand_fyi(tbl); \ 723*b1d06847SJean-Baptiste Boric } while(0) 724*b1d06847SJean-Baptiste Boric 725*b1d06847SJean-Baptiste Boric 726*b1d06847SJean-Baptiste Boric /* This is an adaptation of Simon Tatham's O(n log(n)) mergesort */ 727*b1d06847SJean-Baptiste Boric /* Note that HASH_SORT assumes the hash handle name to be hh. 728*b1d06847SJean-Baptiste Boric * HASH_SRT was added to allow the hash handle name to be passed in. */ 729*b1d06847SJean-Baptiste Boric #define HASH_SORT(head,cmpfcn) HASH_SRT(hh,head,cmpfcn) 730*b1d06847SJean-Baptiste Boric #define HASH_SRT(hh,head,cmpfcn) \ 731*b1d06847SJean-Baptiste Boric do { \ 732*b1d06847SJean-Baptiste Boric unsigned _hs_i; \ 733*b1d06847SJean-Baptiste Boric unsigned _hs_looping,_hs_nmerges,_hs_insize,_hs_psize,_hs_qsize; \ 734*b1d06847SJean-Baptiste Boric struct UT_hash_handle *_hs_p, *_hs_q, *_hs_e, *_hs_list, *_hs_tail; \ 735*b1d06847SJean-Baptiste Boric if (head) { \ 736*b1d06847SJean-Baptiste Boric _hs_insize = 1; \ 737*b1d06847SJean-Baptiste Boric _hs_looping = 1; \ 738*b1d06847SJean-Baptiste Boric _hs_list = &((head)->hh); \ 739*b1d06847SJean-Baptiste Boric while (_hs_looping) { \ 740*b1d06847SJean-Baptiste Boric _hs_p = _hs_list; \ 741*b1d06847SJean-Baptiste Boric _hs_list = NULL; \ 742*b1d06847SJean-Baptiste Boric _hs_tail = NULL; \ 743*b1d06847SJean-Baptiste Boric _hs_nmerges = 0; \ 744*b1d06847SJean-Baptiste Boric while (_hs_p) { \ 745*b1d06847SJean-Baptiste Boric _hs_nmerges++; \ 746*b1d06847SJean-Baptiste Boric _hs_q = _hs_p; \ 747*b1d06847SJean-Baptiste Boric _hs_psize = 0; \ 748*b1d06847SJean-Baptiste Boric for ( _hs_i = 0; _hs_i < _hs_insize; _hs_i++ ) { \ 749*b1d06847SJean-Baptiste Boric _hs_psize++; \ 750*b1d06847SJean-Baptiste Boric _hs_q = (UT_hash_handle*)((_hs_q->next) ? \ 751*b1d06847SJean-Baptiste Boric ((void*)((char*)(_hs_q->next) + \ 752*b1d06847SJean-Baptiste Boric (head)->hh.tbl->hho)) : NULL); \ 753*b1d06847SJean-Baptiste Boric if (! (_hs_q) ) break; \ 754*b1d06847SJean-Baptiste Boric } \ 755*b1d06847SJean-Baptiste Boric _hs_qsize = _hs_insize; \ 756*b1d06847SJean-Baptiste Boric while ((_hs_psize > 0) || ((_hs_qsize > 0) && _hs_q )) { \ 757*b1d06847SJean-Baptiste Boric if (_hs_psize == 0) { \ 758*b1d06847SJean-Baptiste Boric _hs_e = _hs_q; \ 759*b1d06847SJean-Baptiste Boric _hs_q = (UT_hash_handle*)((_hs_q->next) ? \ 760*b1d06847SJean-Baptiste Boric ((void*)((char*)(_hs_q->next) + \ 761*b1d06847SJean-Baptiste Boric (head)->hh.tbl->hho)) : NULL); \ 762*b1d06847SJean-Baptiste Boric _hs_qsize--; \ 763*b1d06847SJean-Baptiste Boric } else if ( (_hs_qsize == 0) || !(_hs_q) ) { \ 764*b1d06847SJean-Baptiste Boric _hs_e = _hs_p; \ 765*b1d06847SJean-Baptiste Boric if (_hs_p){ \ 766*b1d06847SJean-Baptiste Boric _hs_p = (UT_hash_handle*)((_hs_p->next) ? \ 767*b1d06847SJean-Baptiste Boric ((void*)((char*)(_hs_p->next) + \ 768*b1d06847SJean-Baptiste Boric (head)->hh.tbl->hho)) : NULL); \ 769*b1d06847SJean-Baptiste Boric } \ 770*b1d06847SJean-Baptiste Boric _hs_psize--; \ 771*b1d06847SJean-Baptiste Boric } else if (( \ 772*b1d06847SJean-Baptiste Boric cmpfcn(DECLTYPE(head)(ELMT_FROM_HH((head)->hh.tbl,_hs_p)), \ 773*b1d06847SJean-Baptiste Boric DECLTYPE(head)(ELMT_FROM_HH((head)->hh.tbl,_hs_q))) \ 774*b1d06847SJean-Baptiste Boric ) <= 0) { \ 775*b1d06847SJean-Baptiste Boric _hs_e = _hs_p; \ 776*b1d06847SJean-Baptiste Boric if (_hs_p){ \ 777*b1d06847SJean-Baptiste Boric _hs_p = (UT_hash_handle*)((_hs_p->next) ? \ 778*b1d06847SJean-Baptiste Boric ((void*)((char*)(_hs_p->next) + \ 779*b1d06847SJean-Baptiste Boric (head)->hh.tbl->hho)) : NULL); \ 780*b1d06847SJean-Baptiste Boric } \ 781*b1d06847SJean-Baptiste Boric _hs_psize--; \ 782*b1d06847SJean-Baptiste Boric } else { \ 783*b1d06847SJean-Baptiste Boric _hs_e = _hs_q; \ 784*b1d06847SJean-Baptiste Boric _hs_q = (UT_hash_handle*)((_hs_q->next) ? \ 785*b1d06847SJean-Baptiste Boric ((void*)((char*)(_hs_q->next) + \ 786*b1d06847SJean-Baptiste Boric (head)->hh.tbl->hho)) : NULL); \ 787*b1d06847SJean-Baptiste Boric _hs_qsize--; \ 788*b1d06847SJean-Baptiste Boric } \ 789*b1d06847SJean-Baptiste Boric if ( _hs_tail ) { \ 790*b1d06847SJean-Baptiste Boric _hs_tail->next = ((_hs_e) ? \ 791*b1d06847SJean-Baptiste Boric ELMT_FROM_HH((head)->hh.tbl,_hs_e) : NULL); \ 792*b1d06847SJean-Baptiste Boric } else { \ 793*b1d06847SJean-Baptiste Boric _hs_list = _hs_e; \ 794*b1d06847SJean-Baptiste Boric } \ 795*b1d06847SJean-Baptiste Boric if (_hs_e) { \ 796*b1d06847SJean-Baptiste Boric _hs_e->prev = ((_hs_tail) ? \ 797*b1d06847SJean-Baptiste Boric ELMT_FROM_HH((head)->hh.tbl,_hs_tail) : NULL); \ 798*b1d06847SJean-Baptiste Boric } \ 799*b1d06847SJean-Baptiste Boric _hs_tail = _hs_e; \ 800*b1d06847SJean-Baptiste Boric } \ 801*b1d06847SJean-Baptiste Boric _hs_p = _hs_q; \ 802*b1d06847SJean-Baptiste Boric } \ 803*b1d06847SJean-Baptiste Boric if (_hs_tail){ \ 804*b1d06847SJean-Baptiste Boric _hs_tail->next = NULL; \ 805*b1d06847SJean-Baptiste Boric } \ 806*b1d06847SJean-Baptiste Boric if ( _hs_nmerges <= 1 ) { \ 807*b1d06847SJean-Baptiste Boric _hs_looping=0; \ 808*b1d06847SJean-Baptiste Boric (head)->hh.tbl->tail = _hs_tail; \ 809*b1d06847SJean-Baptiste Boric DECLTYPE_ASSIGN(head,ELMT_FROM_HH((head)->hh.tbl, _hs_list)); \ 810*b1d06847SJean-Baptiste Boric } \ 811*b1d06847SJean-Baptiste Boric _hs_insize *= 2; \ 812*b1d06847SJean-Baptiste Boric } \ 813*b1d06847SJean-Baptiste Boric HASH_FSCK(hh,head); \ 814*b1d06847SJean-Baptiste Boric } \ 815*b1d06847SJean-Baptiste Boric } while (0) 816*b1d06847SJean-Baptiste Boric 817*b1d06847SJean-Baptiste Boric /* This function selects items from one hash into another hash. 818*b1d06847SJean-Baptiste Boric * The end result is that the selected items have dual presence 819*b1d06847SJean-Baptiste Boric * in both hashes. There is no copy of the items made; rather 820*b1d06847SJean-Baptiste Boric * they are added into the new hash through a secondary hash 821*b1d06847SJean-Baptiste Boric * hash handle that must be present in the structure. */ 822*b1d06847SJean-Baptiste Boric #define HASH_SELECT(hh_dst, dst, hh_src, src, cond) \ 823*b1d06847SJean-Baptiste Boric do { \ 824*b1d06847SJean-Baptiste Boric unsigned _src_bkt, _dst_bkt; \ 825*b1d06847SJean-Baptiste Boric void *_last_elt=NULL, *_elt; \ 826*b1d06847SJean-Baptiste Boric UT_hash_handle *_src_hh, *_dst_hh, *_last_elt_hh=NULL; \ 827*b1d06847SJean-Baptiste Boric ptrdiff_t _dst_hho = ((char*)(&(dst)->hh_dst) - (char*)(dst)); \ 828*b1d06847SJean-Baptiste Boric if (src) { \ 829*b1d06847SJean-Baptiste Boric for(_src_bkt=0; _src_bkt < (src)->hh_src.tbl->num_buckets; _src_bkt++) { \ 830*b1d06847SJean-Baptiste Boric for(_src_hh = (src)->hh_src.tbl->buckets[_src_bkt].hh_head; \ 831*b1d06847SJean-Baptiste Boric _src_hh; \ 832*b1d06847SJean-Baptiste Boric _src_hh = _src_hh->hh_next) { \ 833*b1d06847SJean-Baptiste Boric _elt = ELMT_FROM_HH((src)->hh_src.tbl, _src_hh); \ 834*b1d06847SJean-Baptiste Boric if (cond(_elt)) { \ 835*b1d06847SJean-Baptiste Boric _dst_hh = (UT_hash_handle*)(((char*)_elt) + _dst_hho); \ 836*b1d06847SJean-Baptiste Boric _dst_hh->key = _src_hh->key; \ 837*b1d06847SJean-Baptiste Boric _dst_hh->keylen = _src_hh->keylen; \ 838*b1d06847SJean-Baptiste Boric _dst_hh->hashv = _src_hh->hashv; \ 839*b1d06847SJean-Baptiste Boric _dst_hh->prev = _last_elt; \ 840*b1d06847SJean-Baptiste Boric _dst_hh->next = NULL; \ 841*b1d06847SJean-Baptiste Boric if (_last_elt_hh) { _last_elt_hh->next = _elt; } \ 842*b1d06847SJean-Baptiste Boric if (!dst) { \ 843*b1d06847SJean-Baptiste Boric DECLTYPE_ASSIGN(dst,_elt); \ 844*b1d06847SJean-Baptiste Boric HASH_MAKE_TABLE(hh_dst,dst); \ 845*b1d06847SJean-Baptiste Boric } else { \ 846*b1d06847SJean-Baptiste Boric _dst_hh->tbl = (dst)->hh_dst.tbl; \ 847*b1d06847SJean-Baptiste Boric } \ 848*b1d06847SJean-Baptiste Boric HASH_TO_BKT(_dst_hh->hashv, _dst_hh->tbl->num_buckets, _dst_bkt); \ 849*b1d06847SJean-Baptiste Boric HASH_ADD_TO_BKT(_dst_hh->tbl->buckets[_dst_bkt],_dst_hh); \ 850*b1d06847SJean-Baptiste Boric (dst)->hh_dst.tbl->num_items++; \ 851*b1d06847SJean-Baptiste Boric _last_elt = _elt; \ 852*b1d06847SJean-Baptiste Boric _last_elt_hh = _dst_hh; \ 853*b1d06847SJean-Baptiste Boric } \ 854*b1d06847SJean-Baptiste Boric } \ 855*b1d06847SJean-Baptiste Boric } \ 856*b1d06847SJean-Baptiste Boric } \ 857*b1d06847SJean-Baptiste Boric HASH_FSCK(hh_dst,dst); \ 858*b1d06847SJean-Baptiste Boric } while (0) 859*b1d06847SJean-Baptiste Boric 860*b1d06847SJean-Baptiste Boric #define HASH_CLEAR(hh,head) \ 861*b1d06847SJean-Baptiste Boric do { \ 862*b1d06847SJean-Baptiste Boric if (head) { \ 863*b1d06847SJean-Baptiste Boric uthash_free((head)->hh.tbl->buckets, \ 864*b1d06847SJean-Baptiste Boric (head)->hh.tbl->num_buckets*sizeof(struct UT_hash_bucket)); \ 865*b1d06847SJean-Baptiste Boric HASH_BLOOM_FREE((head)->hh.tbl); \ 866*b1d06847SJean-Baptiste Boric uthash_free((head)->hh.tbl, sizeof(UT_hash_table)); \ 867*b1d06847SJean-Baptiste Boric (head)=NULL; \ 868*b1d06847SJean-Baptiste Boric } \ 869*b1d06847SJean-Baptiste Boric } while(0) 870*b1d06847SJean-Baptiste Boric 871*b1d06847SJean-Baptiste Boric #define HASH_OVERHEAD(hh,head) \ 872*b1d06847SJean-Baptiste Boric ((head) ? ( \ 873*b1d06847SJean-Baptiste Boric (size_t)((((head)->hh.tbl->num_items * sizeof(UT_hash_handle)) + \ 874*b1d06847SJean-Baptiste Boric ((head)->hh.tbl->num_buckets * sizeof(UT_hash_bucket)) + \ 875*b1d06847SJean-Baptiste Boric (sizeof(UT_hash_table)) + \ 876*b1d06847SJean-Baptiste Boric (HASH_BLOOM_BYTELEN)))) : 0) 877*b1d06847SJean-Baptiste Boric 878*b1d06847SJean-Baptiste Boric #ifdef NO_DECLTYPE 879*b1d06847SJean-Baptiste Boric #define HASH_ITER(hh,head,el,tmp) \ 880*b1d06847SJean-Baptiste Boric for((el)=(head), (*(char**)(&(tmp)))=(char*)((head)?(head)->hh.next:NULL); \ 881*b1d06847SJean-Baptiste Boric el; (el)=(tmp),(*(char**)(&(tmp)))=(char*)((tmp)?(tmp)->hh.next:NULL)) 882*b1d06847SJean-Baptiste Boric #else 883*b1d06847SJean-Baptiste Boric #define HASH_ITER(hh,head,el,tmp) \ 884*b1d06847SJean-Baptiste Boric for((el)=(head),(tmp)=DECLTYPE(el)((head)?(head)->hh.next:NULL); \ 885*b1d06847SJean-Baptiste Boric el; (el)=(tmp),(tmp)=DECLTYPE(el)((tmp)?(tmp)->hh.next:NULL)) 886*b1d06847SJean-Baptiste Boric #endif 887*b1d06847SJean-Baptiste Boric 888*b1d06847SJean-Baptiste Boric /* obtain a count of items in the hash */ 889*b1d06847SJean-Baptiste Boric #define HASH_COUNT(head) HASH_CNT(hh,head) 890*b1d06847SJean-Baptiste Boric #define HASH_CNT(hh,head) ((head)?((head)->hh.tbl->num_items):0) 891*b1d06847SJean-Baptiste Boric 892*b1d06847SJean-Baptiste Boric typedef struct UT_hash_bucket { 893*b1d06847SJean-Baptiste Boric struct UT_hash_handle *hh_head; 894*b1d06847SJean-Baptiste Boric unsigned count; 895*b1d06847SJean-Baptiste Boric 896*b1d06847SJean-Baptiste Boric /* expand_mult is normally set to 0. In this situation, the max chain length 897*b1d06847SJean-Baptiste Boric * threshold is enforced at its default value, HASH_BKT_CAPACITY_THRESH. (If 898*b1d06847SJean-Baptiste Boric * the bucket's chain exceeds this length, bucket expansion is triggered). 899*b1d06847SJean-Baptiste Boric * However, setting expand_mult to a non-zero value delays bucket expansion 900*b1d06847SJean-Baptiste Boric * (that would be triggered by additions to this particular bucket) 901*b1d06847SJean-Baptiste Boric * until its chain length reaches a *multiple* of HASH_BKT_CAPACITY_THRESH. 902*b1d06847SJean-Baptiste Boric * (The multiplier is simply expand_mult+1). The whole idea of this 903*b1d06847SJean-Baptiste Boric * multiplier is to reduce bucket expansions, since they are expensive, in 904*b1d06847SJean-Baptiste Boric * situations where we know that a particular bucket tends to be overused. 905*b1d06847SJean-Baptiste Boric * It is better to let its chain length grow to a longer yet-still-bounded 906*b1d06847SJean-Baptiste Boric * value, than to do an O(n) bucket expansion too often. 907*b1d06847SJean-Baptiste Boric */ 908*b1d06847SJean-Baptiste Boric unsigned expand_mult; 909*b1d06847SJean-Baptiste Boric 910*b1d06847SJean-Baptiste Boric } UT_hash_bucket; 911*b1d06847SJean-Baptiste Boric 912*b1d06847SJean-Baptiste Boric /* random signature used only to find hash tables in external analysis */ 913*b1d06847SJean-Baptiste Boric #define HASH_SIGNATURE 0xa0111fe1 914*b1d06847SJean-Baptiste Boric #define HASH_BLOOM_SIGNATURE 0xb12220f2 915*b1d06847SJean-Baptiste Boric 916*b1d06847SJean-Baptiste Boric typedef struct UT_hash_table { 917*b1d06847SJean-Baptiste Boric UT_hash_bucket *buckets; 918*b1d06847SJean-Baptiste Boric unsigned num_buckets, log2_num_buckets; 919*b1d06847SJean-Baptiste Boric unsigned num_items; 920*b1d06847SJean-Baptiste Boric struct UT_hash_handle *tail; /* tail hh in app order, for fast append */ 921*b1d06847SJean-Baptiste Boric ptrdiff_t hho; /* hash handle offset (byte pos of hash handle in element */ 922*b1d06847SJean-Baptiste Boric 923*b1d06847SJean-Baptiste Boric /* in an ideal situation (all buckets used equally), no bucket would have 924*b1d06847SJean-Baptiste Boric * more than ceil(#items/#buckets) items. that's the ideal chain length. */ 925*b1d06847SJean-Baptiste Boric unsigned ideal_chain_maxlen; 926*b1d06847SJean-Baptiste Boric 927*b1d06847SJean-Baptiste Boric /* nonideal_items is the number of items in the hash whose chain position 928*b1d06847SJean-Baptiste Boric * exceeds the ideal chain maxlen. these items pay the penalty for an uneven 929*b1d06847SJean-Baptiste Boric * hash distribution; reaching them in a chain traversal takes >ideal steps */ 930*b1d06847SJean-Baptiste Boric unsigned nonideal_items; 931*b1d06847SJean-Baptiste Boric 932*b1d06847SJean-Baptiste Boric /* ineffective expands occur when a bucket doubling was performed, but 933*b1d06847SJean-Baptiste Boric * afterward, more than half the items in the hash had nonideal chain 934*b1d06847SJean-Baptiste Boric * positions. If this happens on two consecutive expansions we inhibit any 935*b1d06847SJean-Baptiste Boric * further expansion, as it's not helping; this happens when the hash 936*b1d06847SJean-Baptiste Boric * function isn't a good fit for the key domain. When expansion is inhibited 937*b1d06847SJean-Baptiste Boric * the hash will still work, albeit no longer in constant time. */ 938*b1d06847SJean-Baptiste Boric unsigned ineff_expands, noexpand; 939*b1d06847SJean-Baptiste Boric 940*b1d06847SJean-Baptiste Boric uint32_t signature; /* used only to find hash tables in external analysis */ 941*b1d06847SJean-Baptiste Boric #ifdef HASH_BLOOM 942*b1d06847SJean-Baptiste Boric uint32_t bloom_sig; /* used only to test bloom exists in external analysis */ 943*b1d06847SJean-Baptiste Boric uint8_t *bloom_bv; 944*b1d06847SJean-Baptiste Boric char bloom_nbits; 945*b1d06847SJean-Baptiste Boric #endif 946*b1d06847SJean-Baptiste Boric 947*b1d06847SJean-Baptiste Boric } UT_hash_table; 948*b1d06847SJean-Baptiste Boric 949*b1d06847SJean-Baptiste Boric typedef struct UT_hash_handle { 950*b1d06847SJean-Baptiste Boric struct UT_hash_table *tbl; 951*b1d06847SJean-Baptiste Boric void *prev; /* prev element in app order */ 952*b1d06847SJean-Baptiste Boric void *next; /* next element in app order */ 953*b1d06847SJean-Baptiste Boric struct UT_hash_handle *hh_prev; /* previous hh in bucket order */ 954*b1d06847SJean-Baptiste Boric struct UT_hash_handle *hh_next; /* next hh in bucket order */ 955*b1d06847SJean-Baptiste Boric void *key; /* ptr to enclosing struct's key */ 956*b1d06847SJean-Baptiste Boric unsigned keylen; /* enclosing struct's key len */ 957*b1d06847SJean-Baptiste Boric unsigned hashv; /* result of hash-fcn(key) */ 958*b1d06847SJean-Baptiste Boric } UT_hash_handle; 959*b1d06847SJean-Baptiste Boric 960*b1d06847SJean-Baptiste Boric #endif /* UTHASH_H */ 961