liblttng-ust/jhash.h

   1 /*
   2  * Copyright (C) 2011 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
   3  *
   4  * This library is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU Lesser General Public
   6  * License as published by the Free Software Foundation;
   7  * version 2.1 of the License.
   8  *
   9  * This library is distributed in the hope that it will be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  12  * Lesser General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU Lesser General Public
  15  * License along with this library; if not, write to the Free Software
  16  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301 USA
  17  */
  18
  19 #include <stddef.h>
  20 #include <urcu/compiler.h>
  21 #include <lttng/ust-endian.h>
  22
  23 /*
  24  * Hash function
  25  * Source: http://burtleburtle.net/bob/c/lookup3.c
  26  * Originally Public Domain
  27  */
  28
  29 #define rot(x, k) (((x) << (k)) | ((x) >> (32 - (k))))
  30
  31 #define mix(a, b, c) \
  32 do { \
  33         a -= c; a ^= rot(c,  4); c += b; \
  34         b -= a; b ^= rot(a,  6); a += c; \
  35         c -= b; c ^= rot(b,  8); b += a; \
  36         a -= c; a ^= rot(c, 16); c += b; \
  37         b -= a; b ^= rot(a, 19); a += c; \
  38         c -= b; c ^= rot(b,  4); b += a; \
  39 } while (0)
  40
  41 #define final(a, b, c) \
  42 { \
  43         c ^= b; c -= rot(b, 14); \
  44         a ^= c; a -= rot(c, 11); \
  45         b ^= a; b -= rot(a, 25); \
  46         c ^= b; c -= rot(b, 16); \
  47         a ^= c; a -= rot(c,  4);\
  48         b ^= a; b -= rot(a, 14); \
  49         c ^= b; c -= rot(b, 24); \
  50 }
  51
  52 #if (BYTE_ORDER == LITTLE_ENDIAN)
  53 #define HASH_LITTLE_ENDIAN      1
  54 #else
  55 #define HASH_LITTLE_ENDIAN      0
  56 #endif
  57
  58 /*
  59  *
  60  * hashlittle() -- hash a variable-length key into a 32-bit value
  61  *   k       : the key (the unaligned variable-length array of bytes)
  62  *   length  : the length of the key, counting by bytes
  63  *   initval : can be any 4-byte value
  64  * Returns a 32-bit value.  Every bit of the key affects every bit of
  65  * the return value.  Two keys differing by one or two bits will have
  66  * totally different hash values.
  67  *
  68  * The best hash table sizes are powers of 2.  There is no need to do
  69  * mod a prime (mod is sooo slow!).  If you need less than 32 bits,
  70  * use a bitmask.  For example, if you need only 10 bits, do
  71  *   h = (h & hashmask(10));
  72  * In which case, the hash table should have hashsize(10) elements.
  73  *
  74  * If you are hashing n strings (uint8_t **)k, do it like this:
  75  *   for (i = 0, h = 0; i < n; ++i) h = hashlittle(k[i], len[i], h);
  76  *
  77  * By Bob Jenkins, 2006.  bob_jenkins@burtleburtle.net.  You may use this
  78  * code any way you wish, private, educational, or commercial.  It's free.
  79  *
  80  * Use for hash table lookup, or anything where one collision in 2^^32 is
  81  * acceptable.  Do NOT use for cryptographic purposes.
  82  */
  83 static
  84 uint32_t hashlittle(const void *key, size_t length, uint32_t initval)
  85 {
  86         uint32_t a, b, c;       /* internal state */
  87         union {
  88                 const void *ptr;
  89                 size_t i;
  90         } u;
  91
  92         /* Set up the internal state */
  93         a = b = c = 0xdeadbeef + ((uint32_t)length) + initval;
  94
  95         u.ptr = key;
  96         if (HASH_LITTLE_ENDIAN && ((u.i & 0x3) == 0)) {
  97                 const uint32_t *k = (const uint32_t *) key;     /* read 32-bit chunks */
  98
  99                 /*------ all but last block: aligned reads and affect 32 bits of (a,b,c) */
 100                 while (length > 12) {
 101                         a += k[0];
 102                         b += k[1];
 103                         c += k[2];
 104                         mix(a, b, c);
 105                         length -= 12;
 106                         k += 3;
 107                 }
 108
 109                 /*----------------------------- handle the last (probably partial) block */
 110                 /*
 111                  * "k[2]&0xffffff" actually reads beyond the end of the string, but
 112                  * then masks off the part it's not allowed to read.    Because the
 113                  * string is aligned, the masked-off tail is in the same word as the
 114                  * rest of the string.  Every machine with memory protection I've seen
 115                  * does it on word boundaries, so is OK with this.      But VALGRIND will
 116                  * still catch it and complain. The masking trick does make the hash
 117                  * noticably faster for short strings (like English words).
 118                  */
 119 #ifndef VALGRIND
 120
 121                 switch (length) {
 122                 case 12: c+=k[2]; b+=k[1]; a+=k[0]; break;
 123                 case 11: c+=k[2]&0xffffff; b+=k[1]; a+=k[0]; break;
 124                 case 10: c+=k[2]&0xffff; b+=k[1]; a+=k[0]; break;
 125                 case 9 : c+=k[2]&0xff; b+=k[1]; a+=k[0]; break;
 126                 case 8 : b+=k[1]; a+=k[0]; break;
 127                 case 7 : b+=k[1]&0xffffff; a+=k[0]; break;
 128                 case 6 : b+=k[1]&0xffff; a+=k[0]; break;
 129                 case 5 : b+=k[1]&0xff; a+=k[0]; break;
 130                 case 4 : a+=k[0]; break;
 131                 case 3 : a+=k[0]&0xffffff; break;
 132                 case 2 : a+=k[0]&0xffff; break;
 133                 case 1 : a+=k[0]&0xff; break;
 134                 case 0 : return c;              /* zero length strings require no mixing */
 135                 }
 136
 137 #else /* make valgrind happy */
 138                 {
 139                         const uint8_t *k8;
 140
 141                         k8 = (const uint8_t *) k;
 142                         switch (length) {
 143                         case 12: c+=k[2]; b+=k[1]; a+=k[0]; break;
 144                         case 11: c+=((uint32_t) k8[10])<<16;    /* fall through */
 145                         case 10: c+=((uint32_t) k8[9])<<8;      /* fall through */
 146                         case 9 : c+=k8[8];                      /* fall through */
 147                         case 8 : b+=k[1]; a+=k[0]; break;
 148                         case 7 : b+=((uint32_t) k8[6])<<16;     /* fall through */
 149                         case 6 : b+=((uint32_t) k8[5])<<8;      /* fall through */
 150                         case 5 : b+=k8[4];                      /* fall through */
 151                         case 4 : a+=k[0]; break;
 152                         case 3 : a+=((uint32_t) k8[2])<<16;     /* fall through */
 153                         case 2 : a+=((uint32_t) k8[1])<<8;      /* fall through */
 154                         case 1 : a+=k8[0]; break;
 155                         case 0 : return c;
 156                         }
 157                 }
 158 #endif /* !valgrind */
 159
 160         } else if (HASH_LITTLE_ENDIAN && ((u.i & 0x1) == 0)) {
 161                 const uint16_t *k = (const uint16_t *) key;     /* read 16-bit chunks */
 162                 const uint8_t *k8;
 163
 164                 /*--------------- all but last block: aligned reads and different mixing */
 165                 while (length > 12)
 166                 {
 167                         a += k[0] + (((uint32_t) k[1])<<16);
 168                         b += k[2] + (((uint32_t) k[3])<<16);
 169                         c += k[4] + (((uint32_t) k[5])<<16);
 170                         mix(a, b, c);
 171                         length -= 12;
 172                         k += 6;
 173                 }
 174
 175                 /*----------------------------- handle the last (probably partial) block */
 176                 k8 = (const uint8_t *) k;
 177                 switch(length)
 178                 {
 179                 case 12: c+=k[4]+(((uint32_t) k[5])<<16);
 180                          b+=k[2]+(((uint32_t) k[3])<<16);
 181                          a+=k[0]+(((uint32_t) k[1])<<16);
 182                          break;
 183                 case 11: c+=((uint32_t) k8[10])<<16;    /* fall through */
 184                 case 10: c+=k[4];
 185                          b+=k[2]+(((uint32_t) k[3])<<16);
 186                          a+=k[0]+(((uint32_t) k[1])<<16);
 187                          break;
 188                 case 9 : c+=k8[8];                      /* fall through */
 189                 case 8 : b+=k[2]+(((uint32_t) k[3])<<16);
 190                          a+=k[0]+(((uint32_t) k[1])<<16);
 191                          break;
 192                 case 7 : b+=((uint32_t) k8[6])<<16;     /* fall through */
 193                 case 6 : b+=k[2];
 194                          a+=k[0]+(((uint32_t) k[1])<<16);
 195                          break;
 196                 case 5 : b+=k8[4];                      /* fall through */
 197                 case 4 : a+=k[0]+(((uint32_t) k[1])<<16);
 198                          break;
 199                 case 3 : a+=((uint32_t) k8[2])<<16;     /* fall through */
 200                 case 2 : a+=k[0];
 201                          break;
 202                 case 1 : a+=k8[0];
 203                          break;
 204                 case 0 : return c;                      /* zero length requires no mixing */
 205                 }
 206
 207         } else {                                        /* need to read the key one byte at a time */
 208                 const uint8_t *k = (const uint8_t *)key;
 209
 210                 /*--------------- all but the last block: affect some 32 bits of (a, b, c) */
 211                 while (length > 12) {
 212                         a += k[0];
 213                         a += ((uint32_t) k[1])<<8;
 214                         a += ((uint32_t) k[2])<<16;
 215                         a += ((uint32_t) k[3])<<24;
 216                         b += k[4];
 217                         b += ((uint32_t) k[5])<<8;
 218                         b += ((uint32_t) k[6])<<16;
 219                         b += ((uint32_t) k[7])<<24;
 220                         c += k[8];
 221                         c += ((uint32_t) k[9])<<8;
 222                         c += ((uint32_t) k[10])<<16;
 223                         c += ((uint32_t) k[11])<<24;
 224                         mix(a,b,c);
 225                         length -= 12;
 226                         k += 12;
 227                 }
 228
 229                 /*-------------------------------- last block: affect all 32 bits of (c) */
 230                 switch (length) {                /* all the case statements fall through */
 231                 case 12: c+=((uint32_t) k[11])<<24;
 232                 case 11: c+=((uint32_t) k[10])<<16;
 233                 case 10: c+=((uint32_t) k[9])<<8;
 234                 case 9 : c+=k[8];
 235                 case 8 : b+=((uint32_t) k[7])<<24;
 236                 case 7 : b+=((uint32_t) k[6])<<16;
 237                 case 6 : b+=((uint32_t) k[5])<<8;
 238                 case 5 : b+=k[4];
 239                 case 4 : a+=((uint32_t) k[3])<<24;
 240                 case 3 : a+=((uint32_t) k[2])<<16;
 241                 case 2 : a+=((uint32_t) k[1])<<8;
 242                 case 1 : a+=k[0];
 243                          break;
 244                 case 0 : return c;
 245                 }
 246         }
 247
 248         final(a, b, c);
 249         return c;
 250 }
 251
 252 static inline
 253 uint32_t jhash(const void *key, size_t length, uint32_t seed)
 254 {
 255         return hashlittle(key, length, seed);
 256 }