| 1 | /* |
| 2 | ------------------------------------------------------------------------------- |
| 3 | lookup3.c, by Bob Jenkins, May 2006, Public Domain. |
| 4 | |
| 5 | These are functions for producing 32-bit hashes for hash table lookup. |
| 6 | hashword(), hashlittle(), hashlittle2(), hashbig(), mix(), and final() |
| 7 | are externally useful functions. Routines to test the hash are included |
| 8 | if SELF_TEST is defined. You can use this free for any purpose. It's in |
| 9 | the public domain. It has no warranty. |
| 10 | |
| 11 | You probably want to use hashlittle(). hashlittle() and hashbig() |
| 12 | hash byte arrays. hashlittle() is is faster than hashbig() on |
| 13 | little-endian machines. Intel and AMD are little-endian machines. |
| 14 | On second thought, you probably want hashlittle2(), which is identical to |
| 15 | hashlittle() except it returns two 32-bit hashes for the price of one. |
| 16 | You could implement hashbig2() if you wanted but I haven't bothered here. |
| 17 | |
| 18 | If you want to find a hash of, say, exactly 7 integers, do |
| 19 | a = i1; b = i2; c = i3; |
| 20 | mix(a,b,c); |
| 21 | a += i4; b += i5; c += i6; |
| 22 | mix(a,b,c); |
| 23 | a += i7; |
| 24 | final(a,b,c); |
| 25 | then use c as the hash value. If you have a variable length array of |
| 26 | 4-byte integers to hash, use hashword(). If you have a byte array (like |
| 27 | a character string), use hashlittle(). If you have several byte arrays, or |
| 28 | a mix of things, see the comments above hashlittle(). |
| 29 | |
| 30 | Why is this so big? I read 12 bytes at a time into 3 4-byte integers, |
| 31 | then mix those integers. This is fast (you can do a lot more thorough |
| 32 | mixing with 12*3 instructions on 3 integers than you can with 3 instructions |
| 33 | on 1 byte), but shoehorning those bytes into integers efficiently is messy. |
| 34 | ------------------------------------------------------------------------------- |
| 35 | */ |
| 36 | |
| 37 | #include <stdlib.h> |
| 38 | |
| 39 | #ifdef HAVE_CONFIG_H |
| 40 | #include <jansson_private_config.h> |
| 41 | #endif |
| 42 | |
| 43 | #ifdef HAVE_STDINT_H |
| 44 | #include <stdint.h> /* defines uint32_t etc */ |
| 45 | #endif |
| 46 | |
| 47 | #ifdef HAVE_SYS_PARAM_H |
| 48 | #include <sys/param.h> /* attempt to define endianness */ |
| 49 | #endif |
| 50 | |
| 51 | #ifdef HAVE_ENDIAN_H |
| 52 | # include <endian.h> /* attempt to define endianness */ |
| 53 | #endif |
| 54 | |
| 55 | /* |
| 56 | * My best guess at if you are big-endian or little-endian. This may |
| 57 | * need adjustment. |
| 58 | */ |
| 59 | #if (defined(__BYTE_ORDER) && defined(__LITTLE_ENDIAN) && \ |
| 60 | __BYTE_ORDER == __LITTLE_ENDIAN) || \ |
| 61 | (defined(i386) || defined(__i386__) || defined(__i486__) || \ |
| 62 | defined(__i586__) || defined(__i686__) || defined(vax) || defined(MIPSEL)) |
| 63 | # define HASH_LITTLE_ENDIAN 1 |
| 64 | # define HASH_BIG_ENDIAN 0 |
| 65 | #elif (defined(__BYTE_ORDER) && defined(__BIG_ENDIAN) && \ |
| 66 | __BYTE_ORDER == __BIG_ENDIAN) || \ |
| 67 | (defined(sparc) || defined(POWERPC) || defined(mc68000) || defined(sel)) |
| 68 | # define HASH_LITTLE_ENDIAN 0 |
| 69 | # define HASH_BIG_ENDIAN 1 |
| 70 | #else |
| 71 | # define HASH_LITTLE_ENDIAN 0 |
| 72 | # define HASH_BIG_ENDIAN 0 |
| 73 | #endif |
| 74 | |
| 75 | #define hashsize(n) ((uint32_t)1<<(n)) |
| 76 | #define hashmask(n) (hashsize(n)-1) |
| 77 | #define rot(x,k) (((x)<<(k)) | ((x)>>(32-(k)))) |
| 78 | |
| 79 | /* |
| 80 | ------------------------------------------------------------------------------- |
| 81 | mix -- mix 3 32-bit values reversibly. |
| 82 | |
| 83 | This is reversible, so any information in (a,b,c) before mix() is |
| 84 | still in (a,b,c) after mix(). |
| 85 | |
| 86 | If four pairs of (a,b,c) inputs are run through mix(), or through |
| 87 | mix() in reverse, there are at least 32 bits of the output that |
| 88 | are sometimes the same for one pair and different for another pair. |
| 89 | This was tested for: |
| 90 | * pairs that differed by one bit, by two bits, in any combination |
| 91 | of top bits of (a,b,c), or in any combination of bottom bits of |
| 92 | (a,b,c). |
| 93 | * "differ" is defined as +, -, ^, or ~^. For + and -, I transformed |
| 94 | the output delta to a Gray code (a^(a>>1)) so a string of 1's (as |
| 95 | is commonly produced by subtraction) look like a single 1-bit |
| 96 | difference. |
| 97 | * the base values were pseudorandom, all zero but one bit set, or |
| 98 | all zero plus a counter that starts at zero. |
| 99 | |
| 100 | Some k values for my "a-=c; a^=rot(c,k); c+=b;" arrangement that |
| 101 | satisfy this are |
| 102 | 4 6 8 16 19 4 |
| 103 | 9 15 3 18 27 15 |
| 104 | 14 9 3 7 17 3 |
| 105 | Well, "9 15 3 18 27 15" didn't quite get 32 bits diffing |
| 106 | for "differ" defined as + with a one-bit base and a two-bit delta. I |
| 107 | used http://burtleburtle.net/bob/hash/avalanche.html to choose |
| 108 | the operations, constants, and arrangements of the variables. |
| 109 | |
| 110 | This does not achieve avalanche. There are input bits of (a,b,c) |
| 111 | that fail to affect some output bits of (a,b,c), especially of a. The |
| 112 | most thoroughly mixed value is c, but it doesn't really even achieve |
| 113 | avalanche in c. |
| 114 | |
| 115 | This allows some parallelism. Read-after-writes are good at doubling |
| 116 | the number of bits affected, so the goal of mixing pulls in the opposite |
| 117 | direction as the goal of parallelism. I did what I could. Rotates |
| 118 | seem to cost as much as shifts on every machine I could lay my hands |
| 119 | on, and rotates are much kinder to the top and bottom bits, so I used |
| 120 | rotates. |
| 121 | ------------------------------------------------------------------------------- |
| 122 | */ |
| 123 | #define mix(a,b,c) \ |
| 124 | { \ |
| 125 | a -= c; a ^= rot(c, 4); c += b; \ |
| 126 | b -= a; b ^= rot(a, 6); a += c; \ |
| 127 | c -= b; c ^= rot(b, 8); b += a; \ |
| 128 | a -= c; a ^= rot(c,16); c += b; \ |
| 129 | b -= a; b ^= rot(a,19); a += c; \ |
| 130 | c -= b; c ^= rot(b, 4); b += a; \ |
| 131 | } |
| 132 | |
| 133 | /* |
| 134 | ------------------------------------------------------------------------------- |
| 135 | final -- final mixing of 3 32-bit values (a,b,c) into c |
| 136 | |
| 137 | Pairs of (a,b,c) values differing in only a few bits will usually |
| 138 | produce values of c that look totally different. This was tested for |
| 139 | * pairs that differed by one bit, by two bits, in any combination |
| 140 | of top bits of (a,b,c), or in any combination of bottom bits of |
| 141 | (a,b,c). |
| 142 | * "differ" is defined as +, -, ^, or ~^. For + and -, I transformed |
| 143 | the output delta to a Gray code (a^(a>>1)) so a string of 1's (as |
| 144 | is commonly produced by subtraction) look like a single 1-bit |
| 145 | difference. |
| 146 | * the base values were pseudorandom, all zero but one bit set, or |
| 147 | all zero plus a counter that starts at zero. |
| 148 | |
| 149 | These constants passed: |
| 150 | 14 11 25 16 4 14 24 |
| 151 | 12 14 25 16 4 14 24 |
| 152 | and these came close: |
| 153 | 4 8 15 26 3 22 24 |
| 154 | 10 8 15 26 3 22 24 |
| 155 | 11 8 15 26 3 22 24 |
| 156 | ------------------------------------------------------------------------------- |
| 157 | */ |
| 158 | #define final(a,b,c) \ |
| 159 | { \ |
| 160 | c ^= b; c -= rot(b,14); \ |
| 161 | a ^= c; a -= rot(c,11); \ |
| 162 | b ^= a; b -= rot(a,25); \ |
| 163 | c ^= b; c -= rot(b,16); \ |
| 164 | a ^= c; a -= rot(c,4); \ |
| 165 | b ^= a; b -= rot(a,14); \ |
| 166 | c ^= b; c -= rot(b,24); \ |
| 167 | } |
| 168 | |
| 169 | /* |
| 170 | ------------------------------------------------------------------------------- |
| 171 | hashlittle() -- hash a variable-length key into a 32-bit value |
| 172 | k : the key (the unaligned variable-length array of bytes) |
| 173 | length : the length of the key, counting by bytes |
| 174 | initval : can be any 4-byte value |
| 175 | Returns a 32-bit value. Every bit of the key affects every bit of |
| 176 | the return value. Two keys differing by one or two bits will have |
| 177 | totally different hash values. |
| 178 | |
| 179 | The best hash table sizes are powers of 2. There is no need to do |
| 180 | mod a prime (mod is sooo slow!). If you need less than 32 bits, |
| 181 | use a bitmask. For example, if you need only 10 bits, do |
| 182 | h = (h & hashmask(10)); |
| 183 | In which case, the hash table should have hashsize(10) elements. |
| 184 | |
| 185 | If you are hashing n strings (uint8_t **)k, do it like this: |
| 186 | for (i=0, h=0; i<n; ++i) h = hashlittle( k[i], len[i], h); |
| 187 | |
| 188 | By Bob Jenkins, 2006. bob_jenkins@burtleburtle.net. You may use this |
| 189 | code any way you wish, private, educational, or commercial. It's free. |
| 190 | |
| 191 | Use for hash table lookup, or anything where one collision in 2^^32 is |
| 192 | acceptable. Do NOT use for cryptographic purposes. |
| 193 | ------------------------------------------------------------------------------- |
| 194 | */ |
| 195 | |
| 196 | static uint32_t hashlittle(const void *key, size_t length, uint32_t initval) |
| 197 | { |
| 198 | uint32_t a,b,c; /* internal state */ |
| 199 | union { const void *ptr; size_t i; } u; /* needed for Mac Powerbook G4 */ |
| 200 | |
| 201 | /* Set up the internal state */ |
| 202 | a = b = c = 0xdeadbeef + ((uint32_t)length) + initval; |
| 203 | |
| 204 | u.ptr = key; |
| 205 | if (HASH_LITTLE_ENDIAN && ((u.i & 0x3) == 0)) { |
| 206 | const uint32_t *k = (const uint32_t *)key; /* read 32-bit chunks */ |
| 207 | |
| 208 | /* Detect Valgrind or AddressSanitizer */ |
| 209 | #ifdef VALGRIND |
| 210 | # define NO_MASKING_TRICK 1 |
| 211 | #else |
| 212 | # if defined(__has_feature) /* Clang */ |
| 213 | # if __has_feature(address_sanitizer) /* is ASAN enabled? */ |
| 214 | # define NO_MASKING_TRICK 1 |
| 215 | # endif |
| 216 | # else |
| 217 | # if defined(__SANITIZE_ADDRESS__) /* GCC 4.8.x, is ASAN enabled? */ |
| 218 | # define NO_MASKING_TRICK 1 |
| 219 | # endif |
| 220 | # endif |
| 221 | #endif |
| 222 | |
| 223 | #ifdef NO_MASKING_TRICK |
| 224 | const uint8_t *k8; |
| 225 | #endif |
| 226 | |
| 227 | /*------ all but last block: aligned reads and affect 32 bits of (a,b,c) */ |
| 228 | while (length > 12) |
| 229 | { |
| 230 | a += k[0]; |
| 231 | b += k[1]; |
| 232 | c += k[2]; |
| 233 | mix(a,b,c); |
| 234 | length -= 12; |
| 235 | k += 3; |
| 236 | } |
| 237 | |
| 238 | /*----------------------------- handle the last (probably partial) block */ |
| 239 | /* |
| 240 | * "k[2]&0xffffff" actually reads beyond the end of the string, but |
| 241 | * then masks off the part it's not allowed to read. Because the |
| 242 | * string is aligned, the masked-off tail is in the same word as the |
| 243 | * rest of the string. Every machine with memory protection I've seen |
| 244 | * does it on word boundaries, so is OK with this. But VALGRIND will |
| 245 | * still catch it and complain. The masking trick does make the hash |
| 246 | * noticably faster for short strings (like English words). |
| 247 | */ |
| 248 | #ifndef NO_MASKING_TRICK |
| 249 | |
| 250 | switch(length) |
| 251 | { |
| 252 | case 12: c+=k[2]; b+=k[1]; a+=k[0]; break; |
| 253 | case 11: c+=k[2]&0xffffff; b+=k[1]; a+=k[0]; break; |
| 254 | case 10: c+=k[2]&0xffff; b+=k[1]; a+=k[0]; break; |
| 255 | case 9 : c+=k[2]&0xff; b+=k[1]; a+=k[0]; break; |
| 256 | case 8 : b+=k[1]; a+=k[0]; break; |
| 257 | case 7 : b+=k[1]&0xffffff; a+=k[0]; break; |
| 258 | case 6 : b+=k[1]&0xffff; a+=k[0]; break; |
| 259 | case 5 : b+=k[1]&0xff; a+=k[0]; break; |
| 260 | case 4 : a+=k[0]; break; |
| 261 | case 3 : a+=k[0]&0xffffff; break; |
| 262 | case 2 : a+=k[0]&0xffff; break; |
| 263 | case 1 : a+=k[0]&0xff; break; |
| 264 | case 0 : return c; /* zero length strings require no mixing */ |
| 265 | } |
| 266 | |
| 267 | #else /* make valgrind happy */ |
| 268 | |
| 269 | k8 = (const uint8_t *)k; |
| 270 | switch(length) |
| 271 | { |
| 272 | case 12: c+=k[2]; b+=k[1]; a+=k[0]; break; |
| 273 | case 11: c+=((uint32_t)k8[10])<<16; /* fall through */ |
| 274 | case 10: c+=((uint32_t)k8[9])<<8; /* fall through */ |
| 275 | case 9 : c+=k8[8]; /* fall through */ |
| 276 | case 8 : b+=k[1]; a+=k[0]; break; |
| 277 | case 7 : b+=((uint32_t)k8[6])<<16; /* fall through */ |
| 278 | case 6 : b+=((uint32_t)k8[5])<<8; /* fall through */ |
| 279 | case 5 : b+=k8[4]; /* fall through */ |
| 280 | case 4 : a+=k[0]; break; |
| 281 | case 3 : a+=((uint32_t)k8[2])<<16; /* fall through */ |
| 282 | case 2 : a+=((uint32_t)k8[1])<<8; /* fall through */ |
| 283 | case 1 : a+=k8[0]; break; |
| 284 | case 0 : return c; |
| 285 | } |
| 286 | |
| 287 | #endif /* !valgrind */ |
| 288 | |
| 289 | } else if (HASH_LITTLE_ENDIAN && ((u.i & 0x1) == 0)) { |
| 290 | const uint16_t *k = (const uint16_t *)key; /* read 16-bit chunks */ |
| 291 | const uint8_t *k8; |
| 292 | |
| 293 | /*--------------- all but last block: aligned reads and different mixing */ |
| 294 | while (length > 12) |
| 295 | { |
| 296 | a += k[0] + (((uint32_t)k[1])<<16); |
| 297 | b += k[2] + (((uint32_t)k[3])<<16); |
| 298 | c += k[4] + (((uint32_t)k[5])<<16); |
| 299 | mix(a,b,c); |
| 300 | length -= 12; |
| 301 | k += 6; |
| 302 | } |
| 303 | |
| 304 | /*----------------------------- handle the last (probably partial) block */ |
| 305 | k8 = (const uint8_t *)k; |
| 306 | switch(length) |
| 307 | { |
| 308 | case 12: c+=k[4]+(((uint32_t)k[5])<<16); |
| 309 | b+=k[2]+(((uint32_t)k[3])<<16); |
| 310 | a+=k[0]+(((uint32_t)k[1])<<16); |
| 311 | break; |
| 312 | case 11: c+=((uint32_t)k8[10])<<16; /* fall through */ |
| 313 | case 10: c+=k[4]; |
| 314 | b+=k[2]+(((uint32_t)k[3])<<16); |
| 315 | a+=k[0]+(((uint32_t)k[1])<<16); |
| 316 | break; |
| 317 | case 9 : c+=k8[8]; /* fall through */ |
| 318 | case 8 : b+=k[2]+(((uint32_t)k[3])<<16); |
| 319 | a+=k[0]+(((uint32_t)k[1])<<16); |
| 320 | break; |
| 321 | case 7 : b+=((uint32_t)k8[6])<<16; /* fall through */ |
| 322 | case 6 : b+=k[2]; |
| 323 | a+=k[0]+(((uint32_t)k[1])<<16); |
| 324 | break; |
| 325 | case 5 : b+=k8[4]; /* fall through */ |
| 326 | case 4 : a+=k[0]+(((uint32_t)k[1])<<16); |
| 327 | break; |
| 328 | case 3 : a+=((uint32_t)k8[2])<<16; /* fall through */ |
| 329 | case 2 : a+=k[0]; |
| 330 | break; |
| 331 | case 1 : a+=k8[0]; |
| 332 | break; |
| 333 | case 0 : return c; /* zero length requires no mixing */ |
| 334 | } |
| 335 | |
| 336 | } else { /* need to read the key one byte at a time */ |
| 337 | const uint8_t *k = (const uint8_t *)key; |
| 338 | |
| 339 | /*--------------- all but the last block: affect some 32 bits of (a,b,c) */ |
| 340 | while (length > 12) |
| 341 | { |
| 342 | a += k[0]; |
| 343 | a += ((uint32_t)k[1])<<8; |
| 344 | a += ((uint32_t)k[2])<<16; |
| 345 | a += ((uint32_t)k[3])<<24; |
| 346 | b += k[4]; |
| 347 | b += ((uint32_t)k[5])<<8; |
| 348 | b += ((uint32_t)k[6])<<16; |
| 349 | b += ((uint32_t)k[7])<<24; |
| 350 | c += k[8]; |
| 351 | c += ((uint32_t)k[9])<<8; |
| 352 | c += ((uint32_t)k[10])<<16; |
| 353 | c += ((uint32_t)k[11])<<24; |
| 354 | mix(a,b,c); |
| 355 | length -= 12; |
| 356 | k += 12; |
| 357 | } |
| 358 | |
| 359 | /*-------------------------------- last block: affect all 32 bits of (c) */ |
| 360 | switch(length) /* all the case statements fall through */ |
| 361 | { |
| 362 | case 12: c+=((uint32_t)k[11])<<24; /* fall through */ |
| 363 | case 11: c+=((uint32_t)k[10])<<16; /* fall through */ |
| 364 | case 10: c+=((uint32_t)k[9])<<8; /* fall through */ |
| 365 | case 9 : c+=k[8]; /* fall through */ |
| 366 | case 8 : b+=((uint32_t)k[7])<<24; /* fall through */ |
| 367 | case 7 : b+=((uint32_t)k[6])<<16; /* fall through */ |
| 368 | case 6 : b+=((uint32_t)k[5])<<8; /* fall through */ |
| 369 | case 5 : b+=k[4]; /* fall through */ |
| 370 | case 4 : a+=((uint32_t)k[3])<<24; /* fall through */ |
| 371 | case 3 : a+=((uint32_t)k[2])<<16; /* fall through */ |
| 372 | case 2 : a+=((uint32_t)k[1])<<8; /* fall through */ |
| 373 | case 1 : a+=k[0]; |
| 374 | break; |
| 375 | case 0 : return c; |
| 376 | } |
| 377 | } |
| 378 | |
| 379 | final(a,b,c); |
| 380 | return c; |
| 381 | } |