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 <stdint.h>
  21 #include <urcu/compiler.h>
  22 #include <lttng/ust-endian.h>
  23
  24 /*
  25  * Hash function
  26  * Source: http://burtleburtle.net/bob/c/lookup3.c
  27  * Originally Public Domain
  28  */
  29
  30 #define rot(x, k) (((x) << (k)) | ((x) >> (32 - (k))))
  31
  32 #define mix(a, b, c) \
  33 do { \
  34         a -= c; a ^= rot(c,  4); c += b; \
  35         b -= a; b ^= rot(a,  6); a += c; \
  36         c -= b; c ^= rot(b,  8); b += a; \
  37         a -= c; a ^= rot(c, 16); c += b; \
  38         b -= a; b ^= rot(a, 19); a += c; \
  39         c -= b; c ^= rot(b,  4); b += a; \
  40 } while (0)
  41
  42 #define final(a, b, c) \
  43 { \
  44         c ^= b; c -= rot(b, 14); \
  45         a ^= c; a -= rot(c, 11); \
  46         b ^= a; b -= rot(a, 25); \
  47         c ^= b; c -= rot(b, 16); \
  48         a ^= c; a -= rot(c,  4);\
  49         b ^= a; b -= rot(a, 14); \
  50         c ^= b; c -= rot(b, 24); \
  51 }
  52
  53 #if (BYTE_ORDER == LITTLE_ENDIAN)
  54 #define HASH_LITTLE_ENDIAN      1
  55 #else
  56 #define HASH_LITTLE_ENDIAN      0
  57 #endif
  58
  59 /*
  60  *
  61  * hashlittle() -- hash a variable-length key into a 32-bit value
  62  *   k       : the key (the unaligned variable-length array of bytes)
  63  *   length  : the length of the key, counting by bytes
  64  *   initval : can be any 4-byte value
  65  * Returns a 32-bit value.  Every bit of the key affects every bit of
  66  * the return value.  Two keys differing by one or two bits will have
  67  * totally different hash values.
  68  *
  69  * The best hash table sizes are powers of 2.  There is no need to do
  70  * mod a prime (mod is sooo slow!).  If you need less than 32 bits,
  71  * use a bitmask.  For example, if you need only 10 bits, do
  72  *   h = (h & hashmask(10));
  73  * In which case, the hash table should have hashsize(10) elements.
  74  *
  75  * If you are hashing n strings (uint8_t **)k, do it like this:
  76  *   for (i = 0, h = 0; i < n; ++i) h = hashlittle(k[i], len[i], h);
  77  *
  78  * By Bob Jenkins, 2006.  bob_jenkins@burtleburtle.net.  You may use this
  79  * code any way you wish, private, educational, or commercial.  It's free.
  80  *
  81  * Use for hash table lookup, or anything where one collision in 2^^32 is
  82  * acceptable.  Do NOT use for cryptographic purposes.
  83  */
  84 static
  85 uint32_t hashlittle(const void *key, size_t length, uint32_t initval)
  86 {
  87         uint32_t a, b, c;       /* internal state */
  88         union {
  89                 const void *ptr;
  90                 size_t i;
  91         } u;
  92
  93         /* Set up the internal state */
  94         a = b = c = 0xdeadbeef + ((uint32_t)length) + initval;
  95
  96         u.ptr = key;
  97         if (HASH_LITTLE_ENDIAN && ((u.i & 0x3) == 0)) {
  98                 const uint32_t *k = (const uint32_t *) key;     /* read 32-bit chunks */
  99
 100                 /*------ all but last block: aligned reads and affect 32 bits of (a,b,c) */
 101                 while (length > 12) {
 102                         a += k[0];
 103                         b += k[1];
 104                         c += k[2];
 105                         mix(a, b, c);
 106                         length -= 12;
 107                         k += 3;
 108                 }
 109
 110                 /*----------------------------- handle the last (probably partial) block */
 111                 /*
 112                  * "k[2]&0xffffff" actually reads beyond the end of the string, but
 113                  * then masks off the part it's not allowed to read.    Because the
 114                  * string is aligned, the masked-off tail is in the same word as the
 115                  * rest of the string.  Every machine with memory protection I've seen
 116                  * does it on word boundaries, so is OK with this.      But VALGRIND will
 117                  * still catch it and complain. The masking trick does make the hash
 118                  * noticably faster for short strings (like English words).
 119                  */
 120 #ifndef VALGRIND
 121
 122                 switch (length) {
 123                 case 12: c+=k[2]; b+=k[1]; a+=k[0]; break;
 124                 case 11: c+=k[2]&0xffffff; b+=k[1]; a+=k[0]; break;
 125                 case 10: c+=k[2]&0xffff; b+=k[1]; a+=k[0]; break;
 126                 case 9 : c+=k[2]&0xff; b+=k[1]; a+=k[0]; break;
 127                 case 8 : b+=k[1]; a+=k[0]; break;
 128                 case 7 : b+=k[1]&0xffffff; a+=k[0]; break;
 129                 case 6 : b+=k[1]&0xffff; a+=k[0]; break;
 130                 case 5 : b+=k[1]&0xff; a+=k[0]; break;
 131                 case 4 : a+=k[0]; break;
 132                 case 3 : a+=k[0]&0xffffff; break;
 133                 case 2 : a+=k[0]&0xffff; break;
 134                 case 1 : a+=k[0]&0xff; break;
 135                 case 0 : return c;              /* zero length strings require no mixing */
 136                 }
 137
 138 #else /* make valgrind happy */
 139                 {
 140                         const uint8_t *k8;
 141
 142                         k8 = (const uint8_t *) k;
 143                         switch (length) {
 144                         case 12: c+=k[2]; b+=k[1]; a+=k[0]; break;
 145                         case 11: c+=((uint32_t) k8[10])<<16;    /* fall through */
 146                         case 10: c+=((uint32_t) k8[9])<<8;      /* fall through */
 147                         case 9 : c+=k8[8];                      /* fall through */
 148                         case 8 : b+=k[1]; a+=k[0]; break;
 149                         case 7 : b+=((uint32_t) k8[6])<<16;     /* fall through */
 150                         case 6 : b+=((uint32_t) k8[5])<<8;      /* fall through */
 151                         case 5 : b+=k8[4];                      /* fall through */
 152                         case 4 : a+=k[0]; break;
 153                         case 3 : a+=((uint32_t) k8[2])<<16;     /* fall through */
 154                         case 2 : a+=((uint32_t) k8[1])<<8;      /* fall through */
 155                         case 1 : a+=k8[0]; break;
 156                         case 0 : return c;
 157                         }
 158                 }
 159 #endif /* !valgrind */
 160
 161         } else if (HASH_LITTLE_ENDIAN && ((u.i & 0x1) == 0)) {
 162                 const uint16_t *k = (const uint16_t *) key;     /* read 16-bit chunks */
 163                 const uint8_t *k8;
 164
 165                 /*--------------- all but last block: aligned reads and different mixing */
 166                 while (length > 12)
 167                 {
 168                         a += k[0] + (((uint32_t) k[1])<<16);
 169                         b += k[2] + (((uint32_t) k[3])<<16);
 170                         c += k[4] + (((uint32_t) k[5])<<16);
 171                         mix(a, b, c);
 172                         length -= 12;
 173                         k += 6;
 174                 }
 175
 176                 /*----------------------------- handle the last (probably partial) block */
 177                 k8 = (const uint8_t *) k;
 178                 switch(length)
 179                 {
 180                 case 12: c+=k[4]+(((uint32_t) k[5])<<16);
 181                          b+=k[2]+(((uint32_t) k[3])<<16);
 182                          a+=k[0]+(((uint32_t) k[1])<<16);
 183                          break;
 184                 case 11: c+=((uint32_t) k8[10])<<16;    /* fall through */
 185                 case 10: c+=k[4];
 186                          b+=k[2]+(((uint32_t) k[3])<<16);
 187                          a+=k[0]+(((uint32_t) k[1])<<16);
 188                          break;
 189                 case 9 : c+=k8[8];                      /* fall through */
 190                 case 8 : b+=k[2]+(((uint32_t) k[3])<<16);
 191                          a+=k[0]+(((uint32_t) k[1])<<16);
 192                          break;
 193                 case 7 : b+=((uint32_t) k8[6])<<16;     /* fall through */
 194                 case 6 : b+=k[2];
 195                          a+=k[0]+(((uint32_t) k[1])<<16);
 196                          break;
 197                 case 5 : b+=k8[4];                      /* fall through */
 198                 case 4 : a+=k[0]+(((uint32_t) k[1])<<16);
 199                          break;
 200                 case 3 : a+=((uint32_t) k8[2])<<16;     /* fall through */
 201                 case 2 : a+=k[0];
 202                          break;
 203                 case 1 : a+=k8[0];
 204                          break;
 205                 case 0 : return c;                      /* zero length requires no mixing */
 206                 }
 207
 208         } else {                                        /* need to read the key one byte at a time */
 209                 const uint8_t *k = (const uint8_t *)key;
 210
 211                 /*--------------- all but the last block: affect some 32 bits of (a, b, c) */
 212                 while (length > 12) {
 213                         a += k[0];
 214                         a += ((uint32_t) k[1])<<8;
 215                         a += ((uint32_t) k[2])<<16;
 216                         a += ((uint32_t) k[3])<<24;
 217                         b += k[4];
 218                         b += ((uint32_t) k[5])<<8;
 219                         b += ((uint32_t) k[6])<<16;
 220                         b += ((uint32_t) k[7])<<24;
 221                         c += k[8];
 222                         c += ((uint32_t) k[9])<<8;
 223                         c += ((uint32_t) k[10])<<16;
 224                         c += ((uint32_t) k[11])<<24;
 225                         mix(a,b,c);
 226                         length -= 12;
 227                         k += 12;
 228                 }
 229
 230                 /*-------------------------------- last block: affect all 32 bits of (c) */
 231                 switch (length) {                /* all the case statements fall through */
 232                 case 12: c+=((uint32_t) k[11])<<24;
 233                 case 11: c+=((uint32_t) k[10])<<16;
 234                 case 10: c+=((uint32_t) k[9])<<8;
 235                 case 9 : c+=k[8];
 236                 case 8 : b+=((uint32_t) k[7])<<24;
 237                 case 7 : b+=((uint32_t) k[6])<<16;
 238                 case 6 : b+=((uint32_t) k[5])<<8;
 239                 case 5 : b+=k[4];
 240                 case 4 : a+=((uint32_t) k[3])<<24;
 241                 case 3 : a+=((uint32_t) k[2])<<16;
 242                 case 2 : a+=((uint32_t) k[1])<<8;
 243                 case 1 : a+=k[0];
 244                          break;
 245                 case 0 : return c;
 246                 }
 247         }
 248
 249         final(a, b, c);
 250         return c;
 251 }
 252
 253 static inline
 254 uint32_t jhash(const void *key, size_t length, uint32_t seed)
 255 {
 256         return hashlittle(key, length, seed);
 257 }