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 #define SELF_TEST 1 37 #undef SELF_TEST 38 39 #include <stdio.h> /* defines printf for tests */ 40 #include <time.h> /* defines time_t for timings in the test */ 41 #include <stdint.h> /* defines uint32_t etc */ 42 #include <sys/param.h> /* attempt to define endianness */ 43 #ifdef linux 44 # include <endian.h> /* attempt to define endianness */ 45 #endif 46 47 /* 48 * My best guess at if you are big-endian or little-endian. This may 49 * need adjustment. 50 */ 51 #if (defined(__BYTE_ORDER) && defined(__LITTLE_ENDIAN) && \ 52 __BYTE_ORDER == __LITTLE_ENDIAN) || \ 53 (defined(i386) || defined(__i386__) || defined(__i486__) || \ 54 defined(__i586__) || defined(__i686__) || defined(vax) || defined(MIPSEL)) 55 # define HASH_LITTLE_ENDIAN 1 56 # define HASH_BIG_ENDIAN 0 57 #elif (defined(__BYTE_ORDER) && defined(__BIG_ENDIAN) && \ 58 __BYTE_ORDER == __BIG_ENDIAN) || \ 59 (defined(sparc) || defined(POWERPC) || defined(mc68000) || defined(sel)) 60 # define HASH_LITTLE_ENDIAN 0 61 # define HASH_BIG_ENDIAN 1 62 #else 63 # define HASH_LITTLE_ENDIAN 0 64 # define HASH_BIG_ENDIAN 0 65 #endif 66 67 #define hashsize(n) ((uint32_t)1<<(n)) 68 #define hashmask(n) (hashsize(n)-1) 69 #define rot(x,k) (((x)<<(k)) | ((x)>>(32-(k)))) 70 71 /* 72 ------------------------------------------------------------------------------- 73 mix -- mix 3 32-bit values reversibly. 74 75 This is reversible, so any information in (a,b,c) before mix() is 76 still in (a,b,c) after mix(). 77 78 If four pairs of (a,b,c) inputs are run through mix(), or through 79 mix() in reverse, there are at least 32 bits of the output that 80 are sometimes the same for one pair and different for another pair. 81 This was tested for: 82 * pairs that differed by one bit, by two bits, in any combination 83 of top bits of (a,b,c), or in any combination of bottom bits of 84 (a,b,c). 85 * "differ" is defined as +, -, ^, or ~^. For + and -, I transformed 86 the output delta to a Gray code (a^(a>>1)) so a string of 1's (as 87 is commonly produced by subtraction) look like a single 1-bit 88 difference. 89 * the base values were pseudorandom, all zero but one bit set, or 90 all zero plus a counter that starts at zero. 91 92 Some k values for my "a-=c; a^=rot(c,k); c+=b;" arrangement that 93 satisfy this are 94 4 6 8 16 19 4 95 9 15 3 18 27 15 96 14 9 3 7 17 3 97 Well, "9 15 3 18 27 15" didn't quite get 32 bits diffing 98 for "differ" defined as + with a one-bit base and a two-bit delta. I 99 used http://burtleburtle.net/bob/hash/avalanche.html to choose 100 the operations, constants, and arrangements of the variables. 101 102 This does not achieve avalanche. There are input bits of (a,b,c) 103 that fail to affect some output bits of (a,b,c), especially of a. The 104 most thoroughly mixed value is c, but it doesn't really even achieve 105 avalanche in c. 106 107 This allows some parallelism. Read-after-writes are good at doubling 108 the number of bits affected, so the goal of mixing pulls in the opposite 109 direction as the goal of parallelism. I did what I could. Rotates 110 seem to cost as much as shifts on every machine I could lay my hands 111 on, and rotates are much kinder to the top and bottom bits, so I used 112 rotates. 113 ------------------------------------------------------------------------------- 114 */ 115 #define mix(a,b,c) \ 116 { \ 117 (a) -= (c); (a) ^= rot(c, 4); (c) += (b); \ 118 (b) -= (a); (b) ^= rot(a, 6); (a) += (c); \ 119 (c) -= (b); (c) ^= rot(b, 8); (b) += (a); \ 120 (a) -= (c); (a) ^= rot(c,16); (c) += (b); \ 121 (b) -= (a); (b) ^= rot(a,19); (a) += (c); \ 122 (c) -= (b); (c) ^= rot(b, 4); (b) += (a); \ 123 } 124 125 /* 126 ------------------------------------------------------------------------------- 127 final -- final mixing of 3 32-bit values (a,b,c) into c 128 129 Pairs of (a,b,c) values differing in only a few bits will usually 130 produce values of c that look totally different. This was tested for 131 * pairs that differed by one bit, by two bits, in any combination 132 of top bits of (a,b,c), or in any combination of bottom bits of 133 (a,b,c). 134 * "differ" is defined as +, -, ^, or ~^. For + and -, I transformed 135 the output delta to a Gray code (a^(a>>1)) so a string of 1's (as 136 is commonly produced by subtraction) look like a single 1-bit 137 difference. 138 * the base values were pseudorandom, all zero but one bit set, or 139 all zero plus a counter that starts at zero. 140 141 These constants passed: 142 14 11 25 16 4 14 24 143 12 14 25 16 4 14 24 144 and these came close: 145 4 8 15 26 3 22 24 146 10 8 15 26 3 22 24 147 11 8 15 26 3 22 24 148 ------------------------------------------------------------------------------- 149 */ 150 #define final(a,b,c) \ 151 { \ 152 (c) ^= (b); (c) -= rot(b,14); \ 153 (a) ^= (c); (a) -= rot(c,11); \ 154 (b) ^= (a); (b) -= rot(a,25); \ 155 (c) ^= (b); (c) -= rot(b,16); \ 156 (a) ^= (c); (a) -= rot(c,4); \ 157 (b) ^= (a); (b) -= rot(a,14); \ 158 (c) ^= (b); (c) -= rot(b,24); \ 159 } 160 161 /* 162 -------------------------------------------------------------------- 163 This works on all machines. To be useful, it requires 164 -- that the key be an array of uint32_t's, and 165 -- that the length be the number of uint32_t's in the key 166 167 The function hashword() is identical to hashlittle() on little-endian 168 machines, and identical to hashbig() on big-endian machines, 169 except that the length has to be measured in uint32_ts rather than in 170 bytes. hashlittle() is more complicated than hashword() only because 171 hashlittle() has to dance around fitting the key bytes into registers. 172 -------------------------------------------------------------------- 173 */ 174 uint32_t hashword( 175 const uint32_t *k, /* the key, an array of uint32_t values */ 176 size_t length, /* the length of the key, in uint32_ts */ 177 uint32_t initval) /* the previous hash, or an arbitrary value */ 178 { 179 uint32_t a,b,c; 180 181 /* Set up the internal state */ 182 a = b = c = 0xdeadbeef + (((uint32_t)length)<<2) + initval; 183 184 /*------------------------------------------------- handle most of the key */ 185 while (length > 3) 186 { 187 a += k[0]; 188 b += k[1]; 189 c += k[2]; 190 mix(a,b,c); 191 length -= 3; 192 k += 3; 193 } 194 195 /*------------------------------------------- handle the last 3 uint32_t's */ 196 switch(length) /* all the case statements fall through */ 197 { 198 case 3 : c+=k[2]; 199 case 2 : b+=k[1]; 200 case 1 : a+=k[0]; 201 final(a,b,c); 202 case 0: /* case 0: nothing left to add */ 203 break; 204 } 205 /*------------------------------------------------------ report the result */ 206 return c; 207 } 208 209 210 /* 211 -------------------------------------------------------------------- 212 hashword2() -- same as hashword(), but take two seeds and return two 213 32-bit values. pc and pb must both be nonnull, and *pc and *pb must 214 both be initialized with seeds. If you pass in (*pb)==0, the output 215 (*pc) will be the same as the return value from hashword(). 216 -------------------------------------------------------------------- 217 */ 218 void hashword2 ( 219 const uint32_t *k, /* the key, an array of uint32_t values */ 220 size_t length, /* the length of the key, in uint32_ts */ 221 uint32_t *pc, /* IN: seed OUT: primary hash value */ 222 uint32_t *pb) /* IN: more seed OUT: secondary hash value */ 223 { 224 uint32_t a,b,c; 225 226 /* Set up the internal state */ 227 a = b = c = 0xdeadbeef + ((uint32_t)(length<<2)) + *pc; 228 c += *pb; 229 230 /*------------------------------------------------- handle most of the key */ 231 while (length > 3) 232 { 233 a += k[0]; 234 b += k[1]; 235 c += k[2]; 236 mix(a,b,c); 237 length -= 3; 238 k += 3; 239 } 240 241 /*------------------------------------------- handle the last 3 uint32_t's */ 242 switch(length) /* all the case statements fall through */ 243 { 244 case 3 : c+=k[2]; 245 case 2 : b+=k[1]; 246 case 1 : a+=k[0]; 247 final(a,b,c); 248 case 0: /* case 0: nothing left to add */ 249 break; 250 } 251 /*------------------------------------------------------ report the result */ 252 *pc=c; *pb=b; 253 } 254 255 256 /* 257 ------------------------------------------------------------------------------- 258 hashlittle() -- hash a variable-length key into a 32-bit value 259 k : the key (the unaligned variable-length array of bytes) 260 length : the length of the key, counting by bytes 261 initval : can be any 4-byte value 262 Returns a 32-bit value. Every bit of the key affects every bit of 263 the return value. Two keys differing by one or two bits will have 264 totally different hash values. 265 266 The best hash table sizes are powers of 2. There is no need to do 267 mod a prime (mod is sooo slow!). If you need less than 32 bits, 268 use a bitmask. For example, if you need only 10 bits, do 269 h = (h & hashmask(10)); 270 In which case, the hash table should have hashsize(10) elements. 271 272 If you are hashing n strings (uint8_t **)k, do it like this: 273 for (i=0, h=0; i<n; ++i) h = hashlittle( k[i], len[i], h); 274 275 By Bob Jenkins, 2006. bob_jenkins (at) burtleburtle.net. You may use this 276 code any way you wish, private, educational, or commercial. It's free. 277 278 Use for hash table lookup, or anything where one collision in 2^^32 is 279 acceptable. Do NOT use for cryptographic purposes. 280 ------------------------------------------------------------------------------- 281 */ 282 283 uint32_t hashlittle( const void *key, size_t length, uint32_t initval) 284 { 285 uint32_t a,b,c; /* internal state */ 286 union { const void *ptr; size_t i; } u; /* needed for Mac Powerbook G4 */ 287 288 /* Set up the internal state */ 289 a = b = c = 0xdeadbeef + ((uint32_t)length) + initval; 290 291 u.ptr = key; 292 if (HASH_LITTLE_ENDIAN && ((u.i & 0x3) == 0)) { 293 const uint32_t *k = (const uint32_t *)key; /* read 32-bit chunks */ 294 const uint8_t *k8; 295 296 /*------ all but last block: aligned reads and affect 32 bits of (a,b,c) */ 297 while (length > 12) 298 { 299 a += k[0]; 300 b += k[1]; 301 c += k[2]; 302 mix(a,b,c); 303 length -= 12; 304 k += 3; 305 } 306 307 /*----------------------------- handle the last (probably partial) block */ 308 /* 309 * "k[2]&0xffffff" actually reads beyond the end of the string, but 310 * then masks off the part it's not allowed to read. Because the 311 * string is aligned, the masked-off tail is in the same word as the 312 * rest of the string. Every machine with memory protection I've seen 313 * does it on word boundaries, so is OK with this. But VALGRIND will 314 * still catch it and complain. The masking trick does make the hash 315 * noticably faster for short strings (like English words). 316 */ 317 #ifndef VALGRIND 318 319 (void) k8; // unused 320 switch(length) 321 { 322 case 12: c+=k[2]; b+=k[1]; a+=k[0]; break; 323 case 11: c+=k[2]&0xffffff; b+=k[1]; a+=k[0]; break; 324 case 10: c+=k[2]&0xffff; b+=k[1]; a+=k[0]; break; 325 case 9 : c+=k[2]&0xff; b+=k[1]; a+=k[0]; break; 326 case 8 : b+=k[1]; a+=k[0]; break; 327 case 7 : b+=k[1]&0xffffff; a+=k[0]; break; 328 case 6 : b+=k[1]&0xffff; a+=k[0]; break; 329 case 5 : b+=k[1]&0xff; a+=k[0]; break; 330 case 4 : a+=k[0]; break; 331 case 3 : a+=k[0]&0xffffff; break; 332 case 2 : a+=k[0]&0xffff; break; 333 case 1 : a+=k[0]&0xff; break; 334 case 0 : return c; /* zero length strings require no mixing */ 335 } 336 337 #else /* make valgrind happy */ 338 339 k8 = (const uint8_t *)k; 340 switch(length) 341 { 342 case 12: c+=k[2]; b+=k[1]; a+=k[0]; break; 343 case 11: c+=((uint32_t)k8[10])<<16; /* fall through */ 344 case 10: c+=((uint32_t)k8[9])<<8; /* fall through */ 345 case 9 : c+=k8[8]; /* fall through */ 346 case 8 : b+=k[1]; a+=k[0]; break; 347 case 7 : b+=((uint32_t)k8[6])<<16; /* fall through */ 348 case 6 : b+=((uint32_t)k8[5])<<8; /* fall through */ 349 case 5 : b+=k8[4]; /* fall through */ 350 case 4 : a+=k[0]; break; 351 case 3 : a+=((uint32_t)k8[2])<<16; /* fall through */ 352 case 2 : a+=((uint32_t)k8[1])<<8; /* fall through */ 353 case 1 : a+=k8[0]; break; 354 case 0 : return c; 355 } 356 357 #endif /* !valgrind */ 358 359 } else if (HASH_LITTLE_ENDIAN && ((u.i & 0x1) == 0)) { 360 const uint16_t *k = (const uint16_t *)key; /* read 16-bit chunks */ 361 const uint8_t *k8; 362 363 /*--------------- all but last block: aligned reads and different mixing */ 364 while (length > 12) 365 { 366 a += k[0] + (((uint32_t)k[1])<<16); 367 b += k[2] + (((uint32_t)k[3])<<16); 368 c += k[4] + (((uint32_t)k[5])<<16); 369 mix(a,b,c); 370 length -= 12; 371 k += 6; 372 } 373 374 /*----------------------------- handle the last (probably partial) block */ 375 k8 = (const uint8_t *)k; 376 switch(length) 377 { 378 case 12: c+=k[4]+(((uint32_t)k[5])<<16); 379 b+=k[2]+(((uint32_t)k[3])<<16); 380 a+=k[0]+(((uint32_t)k[1])<<16); 381 break; 382 case 11: c+=((uint32_t)k8[10])<<16; /* fall through */ 383 case 10: c+=k[4]; 384 b+=k[2]+(((uint32_t)k[3])<<16); 385 a+=k[0]+(((uint32_t)k[1])<<16); 386 break; 387 case 9 : c+=k8[8]; /* fall through */ 388 case 8 : b+=k[2]+(((uint32_t)k[3])<<16); 389 a+=k[0]+(((uint32_t)k[1])<<16); 390 break; 391 case 7 : b+=((uint32_t)k8[6])<<16; /* fall through */ 392 case 6 : b+=k[2]; 393 a+=k[0]+(((uint32_t)k[1])<<16); 394 break; 395 case 5 : b+=k8[4]; /* fall through */ 396 case 4 : a+=k[0]+(((uint32_t)k[1])<<16); 397 break; 398 case 3 : a+=((uint32_t)k8[2])<<16; /* fall through */ 399 case 2 : a+=k[0]; 400 break; 401 case 1 : a+=k8[0]; 402 break; 403 case 0 : return c; /* zero length requires no mixing */ 404 } 405 406 } else { /* need to read the key one byte at a time */ 407 const uint8_t *k = (const uint8_t *)key; 408 409 /*--------------- all but the last block: affect some 32 bits of (a,b,c) */ 410 while (length > 12) 411 { 412 a += k[0]; 413 a += ((uint32_t)k[1])<<8; 414 a += ((uint32_t)k[2])<<16; 415 a += ((uint32_t)k[3])<<24; 416 b += k[4]; 417 b += ((uint32_t)k[5])<<8; 418 b += ((uint32_t)k[6])<<16; 419 b += ((uint32_t)k[7])<<24; 420 c += k[8]; 421 c += ((uint32_t)k[9])<<8; 422 c += ((uint32_t)k[10])<<16; 423 c += ((uint32_t)k[11])<<24; 424 mix(a,b,c); 425 length -= 12; 426 k += 12; 427 } 428 429 /*-------------------------------- last block: affect all 32 bits of (c) */ 430 switch(length) /* all the case statements fall through */ 431 { 432 case 12: c+=((uint32_t)k[11])<<24; 433 case 11: c+=((uint32_t)k[10])<<16; 434 case 10: c+=((uint32_t)k[9])<<8; 435 case 9 : c+=k[8]; 436 case 8 : b+=((uint32_t)k[7])<<24; 437 case 7 : b+=((uint32_t)k[6])<<16; 438 case 6 : b+=((uint32_t)k[5])<<8; 439 case 5 : b+=k[4]; 440 case 4 : a+=((uint32_t)k[3])<<24; 441 case 3 : a+=((uint32_t)k[2])<<16; 442 case 2 : a+=((uint32_t)k[1])<<8; 443 case 1 : a+=k[0]; 444 break; 445 case 0 : return c; 446 } 447 } 448 449 final(a,b,c); 450 return c; 451 } 452 453 454 /* 455 * hashlittle2: return 2 32-bit hash values 456 * 457 * This is identical to hashlittle(), except it returns two 32-bit hash 458 * values instead of just one. This is good enough for hash table 459 * lookup with 2^^64 buckets, or if you want a second hash if you're not 460 * happy with the first, or if you want a probably-unique 64-bit ID for 461 * the key. *pc is better mixed than *pb, so use *pc first. If you want 462 * a 64-bit value do something like "*pc + (((uint64_t)*pb)<<32)". 463 */ 464 void hashlittle2( 465 const void *key, /* the key to hash */ 466 size_t length, /* length of the key */ 467 uint32_t *pc, /* IN: primary initval, OUT: primary hash */ 468 uint32_t *pb) /* IN: secondary initval, OUT: secondary hash */ 469 { 470 uint32_t a,b,c; /* internal state */ 471 union { const void *ptr; size_t i; } u; /* needed for Mac Powerbook G4 */ 472 473 /* Set up the internal state */ 474 a = b = c = 0xdeadbeef + ((uint32_t)length) + *pc; 475 c += *pb; 476 477 u.ptr = key; 478 if (HASH_LITTLE_ENDIAN && ((u.i & 0x3) == 0)) { 479 const uint32_t *k = (const uint32_t *)key; /* read 32-bit chunks */ 480 const uint8_t *k8; 481 482 /*------ all but last block: aligned reads and affect 32 bits of (a,b,c) */ 483 while (length > 12) 484 { 485 a += k[0]; 486 b += k[1]; 487 c += k[2]; 488 mix(a,b,c); 489 length -= 12; 490 k += 3; 491 } 492 493 /*----------------------------- handle the last (probably partial) block */ 494 /* 495 * "k[2]&0xffffff" actually reads beyond the end of the string, but 496 * then masks off the part it's not allowed to read. Because the 497 * string is aligned, the masked-off tail is in the same word as the 498 * rest of the string. Every machine with memory protection I've seen 499 * does it on word boundaries, so is OK with this. But VALGRIND will 500 * still catch it and complain. The masking trick does make the hash 501 * noticably faster for short strings (like English words). 502 */ 503 #ifndef VALGRIND 504 505 (void) k8; // unused 506 switch(length) 507 { 508 case 12: c+=k[2]; b+=k[1]; a+=k[0]; break; 509 case 11: c+=k[2]&0xffffff; b+=k[1]; a+=k[0]; break; 510 case 10: c+=k[2]&0xffff; b+=k[1]; a+=k[0]; break; 511 case 9 : c+=k[2]&0xff; b+=k[1]; a+=k[0]; break; 512 case 8 : b+=k[1]; a+=k[0]; break; 513 case 7 : b+=k[1]&0xffffff; a+=k[0]; break; 514 case 6 : b+=k[1]&0xffff; a+=k[0]; break; 515 case 5 : b+=k[1]&0xff; a+=k[0]; break; 516 case 4 : a+=k[0]; break; 517 case 3 : a+=k[0]&0xffffff; break; 518 case 2 : a+=k[0]&0xffff; break; 519 case 1 : a+=k[0]&0xff; break; 520 case 0 : *pc=c; *pb=b; return; /* zero length strings require no mixing */ 521 } 522 523 #else /* make valgrind happy */ 524 525 k8 = (const uint8_t *)k; 526 switch(length) 527 { 528 case 12: c+=k[2]; b+=k[1]; a+=k[0]; break; 529 case 11: c+=((uint32_t)k8[10])<<16; /* fall through */ 530 case 10: c+=((uint32_t)k8[9])<<8; /* fall through */ 531 case 9 : c+=k8[8]; /* fall through */ 532 case 8 : b+=k[1]; a+=k[0]; break; 533 case 7 : b+=((uint32_t)k8[6])<<16; /* fall through */ 534 case 6 : b+=((uint32_t)k8[5])<<8; /* fall through */ 535 case 5 : b+=k8[4]; /* fall through */ 536 case 4 : a+=k[0]; break; 537 case 3 : a+=((uint32_t)k8[2])<<16; /* fall through */ 538 case 2 : a+=((uint32_t)k8[1])<<8; /* fall through */ 539 case 1 : a+=k8[0]; break; 540 case 0 : *pc=c; *pb=b; return; /* zero length strings require no mixing */ 541 } 542 543 #endif /* !valgrind */ 544 545 } else if (HASH_LITTLE_ENDIAN && ((u.i & 0x1) == 0)) { 546 const uint16_t *k = (const uint16_t *)key; /* read 16-bit chunks */ 547 const uint8_t *k8; 548 549 /*--------------- all but last block: aligned reads and different mixing */ 550 while (length > 12) 551 { 552 a += k[0] + (((uint32_t)k[1])<<16); 553 b += k[2] + (((uint32_t)k[3])<<16); 554 c += k[4] + (((uint32_t)k[5])<<16); 555 mix(a,b,c); 556 length -= 12; 557 k += 6; 558 } 559 560 /*----------------------------- handle the last (probably partial) block */ 561 k8 = (const uint8_t *)k; 562 switch(length) 563 { 564 case 12: c+=k[4]+(((uint32_t)k[5])<<16); 565 b+=k[2]+(((uint32_t)k[3])<<16); 566 a+=k[0]+(((uint32_t)k[1])<<16); 567 break; 568 case 11: c+=((uint32_t)k8[10])<<16; /* fall through */ 569 case 10: c+=k[4]; 570 b+=k[2]+(((uint32_t)k[3])<<16); 571 a+=k[0]+(((uint32_t)k[1])<<16); 572 break; 573 case 9 : c+=k8[8]; /* fall through */ 574 case 8 : b+=k[2]+(((uint32_t)k[3])<<16); 575 a+=k[0]+(((uint32_t)k[1])<<16); 576 break; 577 case 7 : b+=((uint32_t)k8[6])<<16; /* fall through */ 578 case 6 : b+=k[2]; 579 a+=k[0]+(((uint32_t)k[1])<<16); 580 break; 581 case 5 : b+=k8[4]; /* fall through */ 582 case 4 : a+=k[0]+(((uint32_t)k[1])<<16); 583 break; 584 case 3 : a+=((uint32_t)k8[2])<<16; /* fall through */ 585 case 2 : a+=k[0]; 586 break; 587 case 1 : a+=k8[0]; 588 break; 589 case 0 : *pc=c; *pb=b; return; /* zero length strings require no mixing */ 590 } 591 592 } else { /* need to read the key one byte at a time */ 593 const uint8_t *k = (const uint8_t *)key; 594 595 /*--------------- all but the last block: affect some 32 bits of (a,b,c) */ 596 while (length > 12) 597 { 598 a += k[0]; 599 a += ((uint32_t)k[1])<<8; 600 a += ((uint32_t)k[2])<<16; 601 a += ((uint32_t)k[3])<<24; 602 b += k[4]; 603 b += ((uint32_t)k[5])<<8; 604 b += ((uint32_t)k[6])<<16; 605 b += ((uint32_t)k[7])<<24; 606 c += k[8]; 607 c += ((uint32_t)k[9])<<8; 608 c += ((uint32_t)k[10])<<16; 609 c += ((uint32_t)k[11])<<24; 610 mix(a,b,c); 611 length -= 12; 612 k += 12; 613 } 614 615 /*-------------------------------- last block: affect all 32 bits of (c) */ 616 switch(length) /* all the case statements fall through */ 617 { 618 case 12: c+=((uint32_t)k[11])<<24; 619 case 11: c+=((uint32_t)k[10])<<16; 620 case 10: c+=((uint32_t)k[9])<<8; 621 case 9 : c+=k[8]; 622 case 8 : b+=((uint32_t)k[7])<<24; 623 case 7 : b+=((uint32_t)k[6])<<16; 624 case 6 : b+=((uint32_t)k[5])<<8; 625 case 5 : b+=k[4]; 626 case 4 : a+=((uint32_t)k[3])<<24; 627 case 3 : a+=((uint32_t)k[2])<<16; 628 case 2 : a+=((uint32_t)k[1])<<8; 629 case 1 : a+=k[0]; 630 break; 631 case 0 : *pc=c; *pb=b; return; /* zero length strings require no mixing */ 632 } 633 } 634 635 final(a,b,c); 636 *pc=c; *pb=b; 637 } 638 639 640 641 /* 642 * hashbig(): 643 * This is the same as hashword() on big-endian machines. It is different 644 * from hashlittle() on all machines. hashbig() takes advantage of 645 * big-endian byte ordering. 646 */ 647 uint32_t hashbig( const void *key, size_t length, uint32_t initval) 648 { 649 uint32_t a,b,c; 650 union { const void *ptr; size_t i; } u; /* to cast key to (size_t) happily */ 651 652 /* Set up the internal state */ 653 a = b = c = 0xdeadbeef + ((uint32_t)length) + initval; 654 655 u.ptr = key; 656 if (HASH_BIG_ENDIAN && ((u.i & 0x3) == 0)) { 657 const uint32_t *k = (const uint32_t *)key; /* read 32-bit chunks */ 658 const uint8_t *k8; 659 660 /*------ all but last block: aligned reads and affect 32 bits of (a,b,c) */ 661 while (length > 12) 662 { 663 a += k[0]; 664 b += k[1]; 665 c += k[2]; 666 mix(a,b,c); 667 length -= 12; 668 k += 3; 669 } 670 671 /*----------------------------- handle the last (probably partial) block */ 672 /* 673 * "k[2]<<8" actually reads beyond the end of the string, but 674 * then shifts out the part it's not allowed to read. Because the 675 * string is aligned, the illegal read is in the same word as the 676 * rest of the string. Every machine with memory protection I've seen 677 * does it on word boundaries, so is OK with this. But VALGRIND will 678 * still catch it and complain. The masking trick does make the hash 679 * noticably faster for short strings (like English words). 680 */ 681 #ifndef VALGRIND 682 683 (void) k8; // unused 684 switch(length) 685 { 686 case 12: c+=k[2]; b+=k[1]; a+=k[0]; break; 687 case 11: c+=k[2]&0xffffff00; b+=k[1]; a+=k[0]; break; 688 case 10: c+=k[2]&0xffff0000; b+=k[1]; a+=k[0]; break; 689 case 9 : c+=k[2]&0xff000000; b+=k[1]; a+=k[0]; break; 690 case 8 : b+=k[1]; a+=k[0]; break; 691 case 7 : b+=k[1]&0xffffff00; a+=k[0]; break; 692 case 6 : b+=k[1]&0xffff0000; a+=k[0]; break; 693 case 5 : b+=k[1]&0xff000000; a+=k[0]; break; 694 case 4 : a+=k[0]; break; 695 case 3 : a+=k[0]&0xffffff00; break; 696 case 2 : a+=k[0]&0xffff0000; break; 697 case 1 : a+=k[0]&0xff000000; break; 698 case 0 : return c; /* zero length strings require no mixing */ 699 } 700 701 #else /* make valgrind happy */ 702 703 k8 = (const uint8_t *)k; 704 switch(length) /* all the case statements fall through */ 705 { 706 case 12: c+=k[2]; b+=k[1]; a+=k[0]; break; 707 case 11: c+=((uint32_t)k8[10])<<8; /* fall through */ 708 case 10: c+=((uint32_t)k8[9])<<16; /* fall through */ 709 case 9 : c+=((uint32_t)k8[8])<<24; /* fall through */ 710 case 8 : b+=k[1]; a+=k[0]; break; 711 case 7 : b+=((uint32_t)k8[6])<<8; /* fall through */ 712 case 6 : b+=((uint32_t)k8[5])<<16; /* fall through */ 713 case 5 : b+=((uint32_t)k8[4])<<24; /* fall through */ 714 case 4 : a+=k[0]; break; 715 case 3 : a+=((uint32_t)k8[2])<<8; /* fall through */ 716 case 2 : a+=((uint32_t)k8[1])<<16; /* fall through */ 717 case 1 : a+=((uint32_t)k8[0])<<24; break; 718 case 0 : return c; 719 } 720 721 #endif /* !VALGRIND */ 722 723 } else { /* need to read the key one byte at a time */ 724 const uint8_t *k = (const uint8_t *)key; 725 726 /*--------------- all but the last block: affect some 32 bits of (a,b,c) */ 727 while (length > 12) 728 { 729 a += ((uint32_t)k[0])<<24; 730 a += ((uint32_t)k[1])<<16; 731 a += ((uint32_t)k[2])<<8; 732 a += ((uint32_t)k[3]); 733 b += ((uint32_t)k[4])<<24; 734 b += ((uint32_t)k[5])<<16; 735 b += ((uint32_t)k[6])<<8; 736 b += ((uint32_t)k[7]); 737 c += ((uint32_t)k[8])<<24; 738 c += ((uint32_t)k[9])<<16; 739 c += ((uint32_t)k[10])<<8; 740 c += ((uint32_t)k[11]); 741 mix(a,b,c); 742 length -= 12; 743 k += 12; 744 } 745 746 /*-------------------------------- last block: affect all 32 bits of (c) */ 747 switch(length) /* all the case statements fall through */ 748 { 749 case 12: c+=k[11]; 750 case 11: c+=((uint32_t)k[10])<<8; 751 case 10: c+=((uint32_t)k[9])<<16; 752 case 9 : c+=((uint32_t)k[8])<<24; 753 case 8 : b+=k[7]; 754 case 7 : b+=((uint32_t)k[6])<<8; 755 case 6 : b+=((uint32_t)k[5])<<16; 756 case 5 : b+=((uint32_t)k[4])<<24; 757 case 4 : a+=k[3]; 758 case 3 : a+=((uint32_t)k[2])<<8; 759 case 2 : a+=((uint32_t)k[1])<<16; 760 case 1 : a+=((uint32_t)k[0])<<24; 761 break; 762 case 0 : return c; 763 } 764 } 765 766 final(a,b,c); 767 return c; 768 } 769 770 771 #ifdef SELF_TEST 772 773 /* used for timings */ 774 void driver1() 775 { 776 uint8_t buf[256]; 777 uint32_t i; 778 uint32_t h=0; 779 time_t a,z; 780 781 time(&a); 782 for (i=0; i<256; ++i) buf[i] = 'x'; 783 for (i=0; i<1; ++i) 784 { 785 h = hashlittle(&buf[0],1,h); 786 } 787 time(&z); 788 if (z-a > 0) printf("time %d %.8x\n", z-a, h); 789 } 790 791 /* check that every input bit changes every output bit half the time */ 792 #define HASHSTATE 1 793 #define HASHLEN 1 794 #define MAXPAIR 60 795 #define MAXLEN 70 796 void driver2() 797 { 798 uint8_t qa[MAXLEN+1], qb[MAXLEN+2], *a = &qa[0], *b = &qb[1]; 799 uint32_t c[HASHSTATE], d[HASHSTATE], i=0, j=0, k, l, m=0, z; 800 uint32_t e[HASHSTATE],f[HASHSTATE],g[HASHSTATE],h[HASHSTATE]; 801 uint32_t x[HASHSTATE],y[HASHSTATE]; 802 uint32_t hlen; 803 804 printf("No more than %d trials should ever be needed \n",MAXPAIR/2); 805 for (hlen=0; hlen < MAXLEN; ++hlen) 806 { 807 z=0; 808 for (i=0; i<hlen; ++i) /*----------------------- for each input byte, */ 809 { 810 for (j=0; j<8; ++j) /*------------------------ for each input bit, */ 811 { 812 for (m=1; m<8; ++m) /*------------ for serveral possible initvals, */ 813 { 814 for (l=0; l<HASHSTATE; ++l) 815 e[l]=f[l]=g[l]=h[l]=x[l]=y[l]=~((uint32_t)0); 816 817 /*---- check that every output bit is affected by that input bit */ 818 for (k=0; k<MAXPAIR; k+=2) 819 { 820 uint32_t finished=1; 821 /* keys have one bit different */ 822 for (l=0; l<hlen+1; ++l) {a[l] = b[l] = (uint8_t)0;} 823 /* have a and b be two keys differing in only one bit */ 824 a[i] ^= (k<<j); 825 a[i] ^= (k>>(8-j)); 826 c[0] = hashlittle(a, hlen, m); 827 b[i] ^= ((k+1)<<j); 828 b[i] ^= ((k+1)>>(8-j)); 829 d[0] = hashlittle(b, hlen, m); 830 /* check every bit is 1, 0, set, and not set at least once */ 831 for (l=0; l<HASHSTATE; ++l) 832 { 833 e[l] &= (c[l]^d[l]); 834 f[l] &= ~(c[l]^d[l]); 835 g[l] &= c[l]; 836 h[l] &= ~c[l]; 837 x[l] &= d[l]; 838 y[l] &= ~d[l]; 839 if (e[l]|f[l]|g[l]|h[l]|x[l]|y[l]) finished=0; 840 } 841 if (finished) break; 842 } 843 if (k>z) z=k; 844 if (k==MAXPAIR) 845 { 846 printf("Some bit didn't change: "); 847 printf("%.8x %.8x %.8x %.8x %.8x %.8x ", 848 e[0],f[0],g[0],h[0],x[0],y[0]); 849 printf("i %d j %d m %d len %d\n", i, j, m, hlen); 850 } 851 if (z==MAXPAIR) goto done; 852 } 853 } 854 } 855 done: 856 if (z < MAXPAIR) 857 { 858 printf("Mix success %2d bytes %2d initvals ",i,m); 859 printf("required %d trials\n", z/2); 860 } 861 } 862 printf("\n"); 863 } 864 865 /* Check for reading beyond the end of the buffer and alignment problems */ 866 void driver3() 867 { 868 uint8_t buf[MAXLEN+20], *b; 869 uint32_t len; 870 uint8_t q[] = "This is the time for all good men to come to the aid of their country..."; 871 uint32_t h; 872 uint8_t qq[] = "xThis is the time for all good men to come to the aid of their country..."; 873 uint32_t i; 874 uint8_t qqq[] = "xxThis is the time for all good men to come to the aid of their country..."; 875 uint32_t j; 876 uint8_t qqqq[] = "xxxThis is the time for all good men to come to the aid of their country..."; 877 uint32_t ref,x,y; 878 uint8_t *p; 879 880 printf("Endianness. These lines should all be the same (for values filled in):\n"); 881 printf("%.8x %.8x %.8x\n", 882 hashword((const uint32_t *)q, (sizeof(q)-1)/4, 13), 883 hashword((const uint32_t *)q, (sizeof(q)-5)/4, 13), 884 hashword((const uint32_t *)q, (sizeof(q)-9)/4, 13)); 885 p = q; 886 printf("%.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x\n", 887 hashlittle(p, sizeof(q)-1, 13), hashlittle(p, sizeof(q)-2, 13), 888 hashlittle(p, sizeof(q)-3, 13), hashlittle(p, sizeof(q)-4, 13), 889 hashlittle(p, sizeof(q)-5, 13), hashlittle(p, sizeof(q)-6, 13), 890 hashlittle(p, sizeof(q)-7, 13), hashlittle(p, sizeof(q)-8, 13), 891 hashlittle(p, sizeof(q)-9, 13), hashlittle(p, sizeof(q)-10, 13), 892 hashlittle(p, sizeof(q)-11, 13), hashlittle(p, sizeof(q)-12, 13)); 893 p = &qq[1]; 894 printf("%.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x\n", 895 hashlittle(p, sizeof(q)-1, 13), hashlittle(p, sizeof(q)-2, 13), 896 hashlittle(p, sizeof(q)-3, 13), hashlittle(p, sizeof(q)-4, 13), 897 hashlittle(p, sizeof(q)-5, 13), hashlittle(p, sizeof(q)-6, 13), 898 hashlittle(p, sizeof(q)-7, 13), hashlittle(p, sizeof(q)-8, 13), 899 hashlittle(p, sizeof(q)-9, 13), hashlittle(p, sizeof(q)-10, 13), 900 hashlittle(p, sizeof(q)-11, 13), hashlittle(p, sizeof(q)-12, 13)); 901 p = &qqq[2]; 902 printf("%.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x\n", 903 hashlittle(p, sizeof(q)-1, 13), hashlittle(p, sizeof(q)-2, 13), 904 hashlittle(p, sizeof(q)-3, 13), hashlittle(p, sizeof(q)-4, 13), 905 hashlittle(p, sizeof(q)-5, 13), hashlittle(p, sizeof(q)-6, 13), 906 hashlittle(p, sizeof(q)-7, 13), hashlittle(p, sizeof(q)-8, 13), 907 hashlittle(p, sizeof(q)-9, 13), hashlittle(p, sizeof(q)-10, 13), 908 hashlittle(p, sizeof(q)-11, 13), hashlittle(p, sizeof(q)-12, 13)); 909 p = &qqqq[3]; 910 printf("%.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x\n", 911 hashlittle(p, sizeof(q)-1, 13), hashlittle(p, sizeof(q)-2, 13), 912 hashlittle(p, sizeof(q)-3, 13), hashlittle(p, sizeof(q)-4, 13), 913 hashlittle(p, sizeof(q)-5, 13), hashlittle(p, sizeof(q)-6, 13), 914 hashlittle(p, sizeof(q)-7, 13), hashlittle(p, sizeof(q)-8, 13), 915 hashlittle(p, sizeof(q)-9, 13), hashlittle(p, sizeof(q)-10, 13), 916 hashlittle(p, sizeof(q)-11, 13), hashlittle(p, sizeof(q)-12, 13)); 917 printf("\n"); 918 919 /* check that hashlittle2 and hashlittle produce the same results */ 920 i=47; j=0; 921 hashlittle2(q, sizeof(q), &i, &j); 922 if (hashlittle(q, sizeof(q), 47) != i) 923 printf("hashlittle2 and hashlittle mismatch\n"); 924 925 /* check that hashword2 and hashword produce the same results */ 926 len = 0xdeadbeef; 927 i=47, j=0; 928 hashword2(&len, 1, &i, &j); 929 if (hashword(&len, 1, 47) != i) 930 printf("hashword2 and hashword mismatch %x %x\n", 931 i, hashword(&len, 1, 47)); 932 933 /* check hashlittle doesn't read before or after the ends of the string */ 934 for (h=0, b=buf+1; h<8; ++h, ++b) 935 { 936 for (i=0; i<MAXLEN; ++i) 937 { 938 len = i; 939 for (j=0; j<i; ++j) *(b+j)=0; 940 941 /* these should all be equal */ 942 ref = hashlittle(b, len, (uint32_t)1); 943 *(b+i)=(uint8_t)~0; 944 *(b-1)=(uint8_t)~0; 945 x = hashlittle(b, len, (uint32_t)1); 946 y = hashlittle(b, len, (uint32_t)1); 947 if ((ref != x) || (ref != y)) 948 { 949 printf("alignment error: %.8x %.8x %.8x %d %d\n",ref,x,y, 950 h, i); 951 } 952 } 953 } 954 } 955 956 /* check for problems with nulls */ 957 void driver4() 958 { 959 uint8_t buf[1]; 960 uint32_t h,i,state[HASHSTATE]; 961 962 963 buf[0] = ~0; 964 for (i=0; i<HASHSTATE; ++i) state[i] = 1; 965 printf("These should all be different\n"); 966 for (i=0, h=0; i<8; ++i) 967 { 968 h = hashlittle(buf, 0, h); 969 printf("%2ld 0-byte strings, hash is %.8x\n", i, h); 970 } 971 } 972 973 void driver5() 974 { 975 uint32_t b,c; 976 b=0, c=0, hashlittle2("", 0, &c, &b); 977 printf("hash is %.8lx %.8lx\n", c, b); /* deadbeef deadbeef */ 978 b=0xdeadbeef, c=0, hashlittle2("", 0, &c, &b); 979 printf("hash is %.8lx %.8lx\n", c, b); /* bd5b7dde deadbeef */ 980 b=0xdeadbeef, c=0xdeadbeef, hashlittle2("", 0, &c, &b); 981 printf("hash is %.8lx %.8lx\n", c, b); /* 9c093ccd bd5b7dde */ 982 b=0, c=0, hashlittle2("Four score and seven years ago", 30, &c, &b); 983 printf("hash is %.8lx %.8lx\n", c, b); /* 17770551 ce7226e6 */ 984 b=1, c=0, hashlittle2("Four score and seven years ago", 30, &c, &b); 985 printf("hash is %.8lx %.8lx\n", c, b); /* e3607cae bd371de4 */ 986 b=0, c=1, hashlittle2("Four score and seven years ago", 30, &c, &b); 987 printf("hash is %.8lx %.8lx\n", c, b); /* cd628161 6cbea4b3 */ 988 c = hashlittle("Four score and seven years ago", 30, 0); 989 printf("hash is %.8lx\n", c); /* 17770551 */ 990 c = hashlittle("Four score and seven years ago", 30, 1); 991 printf("hash is %.8lx\n", c); /* cd628161 */ 992 } 993 994 995 int main() 996 { 997 driver1(); /* test that the key is hashed: used for timings */ 998 driver2(); /* test that whole key is hashed thoroughly */ 999 driver3(); /* test that nothing but the key is hashed */ 1000 driver4(); /* test hashing multiple buffers (all buffers are null) */ 1001 driver5(); /* test the hash against known vectors */ 1002 return 1; 1003 } 1004 1005 #endif /* SELF_TEST */ 1006
