rpm 5.2.1
|
00001 /* -------------------------------------------------------------------- */ 00002 /* 00003 * lookup3.c, by Bob Jenkins, May 2006, Public Domain. 00004 * 00005 * These are functions for producing 32-bit hashes for hash table lookup. 00006 * jlu32w(), jlu32l(), jlu32lpair(), jlu32b(), _JLU3_MIX(), and _JLU3_FINAL() 00007 * are externally useful functions. Routines to test the hash are included 00008 * if SELF_TEST is defined. You can use this free for any purpose. It's in 00009 * the public domain. It has no warranty. 00010 * 00011 * You probably want to use jlu32l(). jlu32l() and jlu32b() 00012 * hash byte arrays. jlu32l() is is faster than jlu32b() on 00013 * little-endian machines. Intel and AMD are little-endian machines. 00014 * On second thought, you probably want jlu32lpair(), which is identical to 00015 * jlu32l() except it returns two 32-bit hashes for the price of one. 00016 * You could implement jlu32bpair() if you wanted but I haven't bothered here. 00017 * 00018 * If you want to find a hash of, say, exactly 7 integers, do 00019 * a = i1; b = i2; c = i3; 00020 * _JLU3_MIX(a,b,c); 00021 * a += i4; b += i5; c += i6; 00022 * _JLU3_MIX(a,b,c); 00023 * a += i7; 00024 * _JLU3_FINAL(a,b,c); 00025 * then use c as the hash value. If you have a variable size array of 00026 * 4-byte integers to hash, use jlu32w(). If you have a byte array (like 00027 * a character string), use jlu32l(). If you have several byte arrays, or 00028 * a mix of things, see the comments above jlu32l(). 00029 * 00030 * Why is this so big? I read 12 bytes at a time into 3 4-byte integers, 00031 * then mix those integers. This is fast (you can do a lot more thorough 00032 * mixing with 12*3 instructions on 3 integers than you can with 3 instructions 00033 * on 1 byte), but shoehorning those bytes into integers efficiently is messy. 00034 */ 00035 /* -------------------------------------------------------------------- */ 00036 00037 #include "system.h" 00038 #include "rpmiotypes.h" 00039 #include "debug.h" 00040 00041 #if defined(_JLU3_SELFTEST) 00042 # define _JLU3_jlu32w 1 00043 # define _JLU3_jlu32l 1 00044 # define _JLU3_jlu32lpair 1 00045 # define _JLU3_jlu32b 1 00046 #endif 00047 00048 /*@-redef@*/ 00049 /*@unchecked@*/ 00050 static const union _dbswap { 00051 const rpmuint32_t ui; 00052 const unsigned char uc[4]; 00053 } endian = { .ui = 0x11223344 }; 00054 # define HASH_LITTLE_ENDIAN (endian.uc[0] == (unsigned char) 0x44) 00055 # define HASH_BIG_ENDIAN (endian.uc[0] == (unsigned char) 0x11) 00056 /*@=redef@*/ 00057 00058 #ifndef ROTL32 00059 # define ROTL32(x, s) (((x) << (s)) | ((x) >> (32 - (s)))) 00060 #endif 00061 00062 /* NOTE: The _size parameter should be in bytes. */ 00063 #define _JLU3_INIT(_h, _size) (0xdeadbeef + ((rpmuint32_t)(_size)) + (_h)) 00064 00065 /* -------------------------------------------------------------------- */ 00066 /* 00067 * _JLU3_MIX -- mix 3 32-bit values reversibly. 00068 * 00069 * This is reversible, so any information in (a,b,c) before _JLU3_MIX() is 00070 * still in (a,b,c) after _JLU3_MIX(). 00071 * 00072 * If four pairs of (a,b,c) inputs are run through _JLU3_MIX(), or through 00073 * _JLU3_MIX() in reverse, there are at least 32 bits of the output that 00074 * are sometimes the same for one pair and different for another pair. 00075 * This was tested for: 00076 * * pairs that differed by one bit, by two bits, in any combination 00077 * of top bits of (a,b,c), or in any combination of bottom bits of 00078 * (a,b,c). 00079 * * "differ" is defined as +, -, ^, or ~^. For + and -, I transformed 00080 * the output delta to a Gray code (a^(a>>1)) so a string of 1's (as 00081 * is commonly produced by subtraction) look like a single 1-bit 00082 * difference. 00083 * * the base values were pseudorandom, all zero but one bit set, or 00084 * all zero plus a counter that starts at zero. 00085 * 00086 * Some k values for my "a-=c; a^=ROTL32(c,k); c+=b;" arrangement that 00087 * satisfy this are 00088 * 4 6 8 16 19 4 00089 * 9 15 3 18 27 15 00090 * 14 9 3 7 17 3 00091 * Well, "9 15 3 18 27 15" didn't quite get 32 bits diffing 00092 * for "differ" defined as + with a one-bit base and a two-bit delta. I 00093 * used http://burtleburtle.net/bob/hash/avalanche.html to choose 00094 * the operations, constants, and arrangements of the variables. 00095 * 00096 * This does not achieve avalanche. There are input bits of (a,b,c) 00097 * that fail to affect some output bits of (a,b,c), especially of a. The 00098 * most thoroughly mixed value is c, but it doesn't really even achieve 00099 * avalanche in c. 00100 * 00101 * This allows some parallelism. Read-after-writes are good at doubling 00102 * the number of bits affected, so the goal of mixing pulls in the opposite 00103 * direction as the goal of parallelism. I did what I could. Rotates 00104 * seem to cost as much as shifts on every machine I could lay my hands 00105 * on, and rotates are much kinder to the top and bottom bits, so I used 00106 * rotates. 00107 */ 00108 /* -------------------------------------------------------------------- */ 00109 #define _JLU3_MIX(a,b,c) \ 00110 { \ 00111 a -= c; a ^= ROTL32(c, 4); c += b; \ 00112 b -= a; b ^= ROTL32(a, 6); a += c; \ 00113 c -= b; c ^= ROTL32(b, 8); b += a; \ 00114 a -= c; a ^= ROTL32(c,16); c += b; \ 00115 b -= a; b ^= ROTL32(a,19); a += c; \ 00116 c -= b; c ^= ROTL32(b, 4); b += a; \ 00117 } 00118 00119 /* -------------------------------------------------------------------- */ 00143 /* -------------------------------------------------------------------- */ 00144 #define _JLU3_FINAL(a,b,c) \ 00145 { \ 00146 c ^= b; c -= ROTL32(b,14); \ 00147 a ^= c; a -= ROTL32(c,11); \ 00148 b ^= a; b -= ROTL32(a,25); \ 00149 c ^= b; c -= ROTL32(b,16); \ 00150 a ^= c; a -= ROTL32(c,4); \ 00151 b ^= a; b -= ROTL32(a,14); \ 00152 c ^= b; c -= ROTL32(b,24); \ 00153 } 00154 00155 #if defined(_JLU3_jlu32w) 00156 rpmuint32_t jlu32w(rpmuint32_t h, /*@null@*/ const rpmuint32_t *k, size_t size) 00157 /*@*/; 00158 /* -------------------------------------------------------------------- */ 00175 /* -------------------------------------------------------------------- */ 00176 rpmuint32_t jlu32w(rpmuint32_t h, const rpmuint32_t *k, size_t size) 00177 { 00178 rpmuint32_t a = _JLU3_INIT(h, (size * sizeof(*k))); 00179 rpmuint32_t b = a; 00180 rpmuint32_t c = a; 00181 00182 if (k == NULL) 00183 goto exit; 00184 00185 /*----------------------------------------------- handle most of the key */ 00186 while (size > 3) { 00187 a += k[0]; 00188 b += k[1]; 00189 c += k[2]; 00190 _JLU3_MIX(a,b,c); 00191 size -= 3; 00192 k += 3; 00193 } 00194 00195 /*----------------------------------------- handle the last 3 rpmuint32_t's */ 00196 switch (size) { 00197 case 3 : c+=k[2]; 00198 case 2 : b+=k[1]; 00199 case 1 : a+=k[0]; 00200 _JLU3_FINAL(a,b,c); 00201 /*@fallthrough@*/ 00202 case 0: 00203 break; 00204 } 00205 /*---------------------------------------------------- report the result */ 00206 exit: 00207 return c; 00208 } 00209 #endif /* defined(_JLU3_jlu32w) */ 00210 00211 #if defined(_JLU3_jlu32l) 00212 rpmuint32_t jlu32l(rpmuint32_t h, const void *key, size_t size) 00213 /*@*/; 00214 /* -------------------------------------------------------------------- */ 00215 /* 00216 * jlu32l() -- hash a variable-length key into a 32-bit value 00217 * h : can be any 4-byte value 00218 * k : the key (the unaligned variable-length array of bytes) 00219 * size : the size of the key, counting by bytes 00220 * Returns a 32-bit value. Every bit of the key affects every bit of 00221 * the return value. Two keys differing by one or two bits will have 00222 * totally different hash values. 00223 * 00224 * The best hash table sizes are powers of 2. There is no need to do 00225 * mod a prime (mod is sooo slow!). If you need less than 32 bits, 00226 * use a bitmask. For example, if you need only 10 bits, do 00227 * h = (h & hashmask(10)); 00228 * In which case, the hash table should have hashsize(10) elements. 00229 * 00230 * If you are hashing n strings (rpmuint8_t **)k, do it like this: 00231 * for (i=0, h=0; i<n; ++i) h = jlu32l(h, k[i], len[i]); 00232 * 00233 * By Bob Jenkins, 2006. bob_jenkins@burtleburtle.net. You may use this 00234 * code any way you wish, private, educational, or commercial. It's free. 00235 * 00236 * Use for hash table lookup, or anything where one collision in 2^^32 is 00237 * acceptable. Do NOT use for cryptographic purposes. 00238 * 00239 * @param h the previous hash, or an arbitrary value 00240 * @param *k the key, an array of rpmuint8_t values 00241 * @param size the size of the key 00242 * @return the lookup3 hash 00243 */ 00244 /* -------------------------------------------------------------------- */ 00245 rpmuint32_t jlu32l(rpmuint32_t h, const void *key, size_t size) 00246 { 00247 union { const void *ptr; size_t i; } u; 00248 rpmuint32_t a = _JLU3_INIT(h, size); 00249 rpmuint32_t b = a; 00250 rpmuint32_t c = a; 00251 00252 if (key == NULL) 00253 goto exit; 00254 00255 u.ptr = key; 00256 if (HASH_LITTLE_ENDIAN && ((u.i & 0x3) == 0)) { 00257 const rpmuint32_t *k = (const rpmuint32_t *)key; /* read 32-bit chunks */ 00258 #ifdef VALGRIND 00259 const rpmuint8_t *k8; 00260 #endif 00261 00262 /*------ all but last block: aligned reads and affect 32 bits of (a,b,c) */ 00263 while (size > 12) { 00264 a += k[0]; 00265 b += k[1]; 00266 c += k[2]; 00267 _JLU3_MIX(a,b,c); 00268 size -= 12; 00269 k += 3; 00270 } 00271 00272 /*------------------------- handle the last (probably partial) block */ 00273 /* 00274 * "k[2]&0xffffff" actually reads beyond the end of the string, but 00275 * then masks off the part it's not allowed to read. Because the 00276 * string is aligned, the masked-off tail is in the same word as the 00277 * rest of the string. Every machine with memory protection I've seen 00278 * does it on word boundaries, so is OK with this. But VALGRIND will 00279 * still catch it and complain. The masking trick does make the hash 00280 * noticably faster for short strings (like English words). 00281 */ 00282 #ifndef VALGRIND 00283 00284 switch (size) { 00285 case 12: c += k[2]; b+=k[1]; a+=k[0]; break; 00286 case 11: c += k[2]&0xffffff; b+=k[1]; a+=k[0]; break; 00287 case 10: c += k[2]&0xffff; b+=k[1]; a+=k[0]; break; 00288 case 9: c += k[2]&0xff; b+=k[1]; a+=k[0]; break; 00289 case 8: b += k[1]; a+=k[0]; break; 00290 case 7: b += k[1]&0xffffff; a+=k[0]; break; 00291 case 6: b += k[1]&0xffff; a+=k[0]; break; 00292 case 5: b += k[1]&0xff; a+=k[0]; break; 00293 case 4: a += k[0]; break; 00294 case 3: a += k[0]&0xffffff; break; 00295 case 2: a += k[0]&0xffff; break; 00296 case 1: a += k[0]&0xff; break; 00297 case 0: goto exit; 00298 } 00299 00300 #else /* make valgrind happy */ 00301 00302 k8 = (const rpmuint8_t *)k; 00303 switch (size) { 00304 case 12: c += k[2]; b+=k[1]; a+=k[0] break; 00305 case 11: c += ((rpmuint32_t)k8[10])<<16; /*@fallthrough@*/ 00306 case 10: c += ((rpmuint32_t)k8[9])<<8; /*@fallthrough@*/ 00307 case 9: c += k8[8]; /*@fallthrough@*/ 00308 case 8: b += k[1]; a+=k[0]; break; 00309 case 7: b += ((rpmuint32_t)k8[6])<<16; /*@fallthrough@*/ 00310 case 6: b += ((rpmuint32_t)k8[5])<<8; /*@fallthrough@*/ 00311 case 5: b += k8[4]; /*@fallthrough@*/ 00312 case 4: a += k[0]; break; 00313 case 3: a += ((rpmuint32_t)k8[2])<<16; /*@fallthrough@*/ 00314 case 2: a += ((rpmuint32_t)k8[1])<<8; /*@fallthrough@*/ 00315 case 1: a += k8[0]; break; 00316 case 0: goto exit; 00317 } 00318 00319 #endif /* !valgrind */ 00320 00321 } else if (HASH_LITTLE_ENDIAN && ((u.i & 0x1) == 0)) { 00322 const rpmuint16_t *k = (const rpmuint16_t *)key; /* read 16-bit chunks */ 00323 const rpmuint8_t *k8; 00324 00325 /*----------- all but last block: aligned reads and different mixing */ 00326 while (size > 12) { 00327 a += k[0] + (((rpmuint32_t)k[1])<<16); 00328 b += k[2] + (((rpmuint32_t)k[3])<<16); 00329 c += k[4] + (((rpmuint32_t)k[5])<<16); 00330 _JLU3_MIX(a,b,c); 00331 size -= 12; 00332 k += 6; 00333 } 00334 00335 /*------------------------- handle the last (probably partial) block */ 00336 k8 = (const rpmuint8_t *)k; 00337 switch (size) { 00338 case 12: 00339 c += k[4]+(((rpmuint32_t)k[5])<<16); 00340 b += k[2]+(((rpmuint32_t)k[3])<<16); 00341 a += k[0]+(((rpmuint32_t)k[1])<<16); 00342 break; 00343 case 11: 00344 c += ((rpmuint32_t)k8[10])<<16; 00345 /*@fallthrough@*/ 00346 case 10: 00347 c += (rpmuint32_t)k[4]; 00348 b += k[2]+(((rpmuint32_t)k[3])<<16); 00349 a += k[0]+(((rpmuint32_t)k[1])<<16); 00350 break; 00351 case 9: 00352 c += (rpmuint32_t)k8[8]; 00353 /*@fallthrough@*/ 00354 case 8: 00355 b += k[2]+(((rpmuint32_t)k[3])<<16); 00356 a += k[0]+(((rpmuint32_t)k[1])<<16); 00357 break; 00358 case 7: 00359 b += ((rpmuint32_t)k8[6])<<16; 00360 /*@fallthrough@*/ 00361 case 6: 00362 b += (rpmuint32_t)k[2]; 00363 a += k[0]+(((rpmuint32_t)k[1])<<16); 00364 break; 00365 case 5: 00366 b += (rpmuint32_t)k8[4]; 00367 /*@fallthrough@*/ 00368 case 4: 00369 a += k[0]+(((rpmuint32_t)k[1])<<16); 00370 break; 00371 case 3: 00372 a += ((rpmuint32_t)k8[2])<<16; 00373 /*@fallthrough@*/ 00374 case 2: 00375 a += (rpmuint32_t)k[0]; 00376 break; 00377 case 1: 00378 a += (rpmuint32_t)k8[0]; 00379 break; 00380 case 0: 00381 goto exit; 00382 } 00383 00384 } else { /* need to read the key one byte at a time */ 00385 const rpmuint8_t *k = (const rpmuint8_t *)key; 00386 00387 /*----------- all but the last block: affect some 32 bits of (a,b,c) */ 00388 while (size > 12) { 00389 a += (rpmuint32_t)k[0]; 00390 a += ((rpmuint32_t)k[1])<<8; 00391 a += ((rpmuint32_t)k[2])<<16; 00392 a += ((rpmuint32_t)k[3])<<24; 00393 b += (rpmuint32_t)k[4]; 00394 b += ((rpmuint32_t)k[5])<<8; 00395 b += ((rpmuint32_t)k[6])<<16; 00396 b += ((rpmuint32_t)k[7])<<24; 00397 c += (rpmuint32_t)k[8]; 00398 c += ((rpmuint32_t)k[9])<<8; 00399 c += ((rpmuint32_t)k[10])<<16; 00400 c += ((rpmuint32_t)k[11])<<24; 00401 _JLU3_MIX(a,b,c); 00402 size -= 12; 00403 k += 12; 00404 } 00405 00406 /*---------------------------- last block: affect all 32 bits of (c) */ 00407 switch (size) { 00408 case 12: c += ((rpmuint32_t)k[11])<<24; /*@fallthrough@*/ 00409 case 11: c += ((rpmuint32_t)k[10])<<16; /*@fallthrough@*/ 00410 case 10: c += ((rpmuint32_t)k[9])<<8; /*@fallthrough@*/ 00411 case 9: c += (rpmuint32_t)k[8]; /*@fallthrough@*/ 00412 case 8: b += ((rpmuint32_t)k[7])<<24; /*@fallthrough@*/ 00413 case 7: b += ((rpmuint32_t)k[6])<<16; /*@fallthrough@*/ 00414 case 6: b += ((rpmuint32_t)k[5])<<8; /*@fallthrough@*/ 00415 case 5: b += (rpmuint32_t)k[4]; /*@fallthrough@*/ 00416 case 4: a += ((rpmuint32_t)k[3])<<24; /*@fallthrough@*/ 00417 case 3: a += ((rpmuint32_t)k[2])<<16; /*@fallthrough@*/ 00418 case 2: a += ((rpmuint32_t)k[1])<<8; /*@fallthrough@*/ 00419 case 1: a += (rpmuint32_t)k[0]; 00420 break; 00421 case 0: 00422 goto exit; 00423 } 00424 } 00425 00426 _JLU3_FINAL(a,b,c); 00427 00428 exit: 00429 return c; 00430 } 00431 #endif /* defined(_JLU3_jlu32l) */ 00432 00433 #if defined(_JLU3_jlu32lpair) 00434 void jlu32lpair(/*@null@*/ const void *key, size_t size, 00435 rpmuint32_t *pc, rpmuint32_t *pb) 00436 /*@modifies *pc, *pb@*/; 00453 void jlu32lpair(const void *key, size_t size, rpmuint32_t *pc, rpmuint32_t *pb) 00454 { 00455 union { const void *ptr; size_t i; } u; 00456 rpmuint32_t a = _JLU3_INIT(*pc, size); 00457 rpmuint32_t b = a; 00458 rpmuint32_t c = a; 00459 00460 if (key == NULL) 00461 goto exit; 00462 00463 c += *pb; /* Add the secondary hash. */ 00464 00465 u.ptr = key; 00466 if (HASH_LITTLE_ENDIAN && ((u.i & 0x3) == 0)) { 00467 const rpmuint32_t *k = (const rpmuint32_t *)key; /* read 32-bit chunks */ 00468 #ifdef VALGRIND 00469 const rpmuint8_t *k8; 00470 #endif 00471 00472 /*-- all but last block: aligned reads and affect 32 bits of (a,b,c) */ 00473 while (size > 12) { 00474 a += k[0]; 00475 b += k[1]; 00476 c += k[2]; 00477 _JLU3_MIX(a,b,c); 00478 size -= 12; 00479 k += 3; 00480 } 00481 /*------------------------- handle the last (probably partial) block */ 00482 /* 00483 * "k[2]&0xffffff" actually reads beyond the end of the string, but 00484 * then masks off the part it's not allowed to read. Because the 00485 * string is aligned, the masked-off tail is in the same word as the 00486 * rest of the string. Every machine with memory protection I've seen 00487 * does it on word boundaries, so is OK with this. But VALGRIND will 00488 * still catch it and complain. The masking trick does make the hash 00489 * noticably faster for short strings (like English words). 00490 */ 00491 #ifndef VALGRIND 00492 00493 switch (size) { 00494 case 12: c += k[2]; b+=k[1]; a+=k[0]; break; 00495 case 11: c += k[2]&0xffffff; b+=k[1]; a+=k[0]; break; 00496 case 10: c += k[2]&0xffff; b+=k[1]; a+=k[0]; break; 00497 case 9: c += k[2]&0xff; b+=k[1]; a+=k[0]; break; 00498 case 8: b += k[1]; a+=k[0]; break; 00499 case 7: b += k[1]&0xffffff; a+=k[0]; break; 00500 case 6: b += k[1]&0xffff; a+=k[0]; break; 00501 case 5: b += k[1]&0xff; a+=k[0]; break; 00502 case 4: a += k[0]; break; 00503 case 3: a += k[0]&0xffffff; break; 00504 case 2: a += k[0]&0xffff; break; 00505 case 1: a += k[0]&0xff; break; 00506 case 0: goto exit; 00507 } 00508 00509 #else /* make valgrind happy */ 00510 00511 k8 = (const rpmuint8_t *)k; 00512 switch (size) { 00513 case 12: c += k[2]; b+=k[1]; a+=k[0]; break; 00514 case 11: c += ((rpmuint32_t)k8[10])<<16; /*@fallthrough@*/ 00515 case 10: c += ((rpmuint32_t)k8[9])<<8; /*@fallthrough@*/ 00516 case 9: c += k8[8]; /*@fallthrough@*/ 00517 case 8: b += k[1]; a+=k[0]; break; 00518 case 7: b += ((rpmuint32_t)k8[6])<<16; /*@fallthrough@*/ 00519 case 6: b += ((rpmuint32_t)k8[5])<<8; /*@fallthrough@*/ 00520 case 5: b += k8[4]; /*@fallthrough@*/ 00521 case 4: a += k[0]; break; 00522 case 3: a += ((rpmuint32_t)k8[2])<<16; /*@fallthrough@*/ 00523 case 2: a += ((rpmuint32_t)k8[1])<<8; /*@fallthrough@*/ 00524 case 1: a += k8[0]; break; 00525 case 0: goto exit; 00526 } 00527 00528 #endif /* !valgrind */ 00529 00530 } else if (HASH_LITTLE_ENDIAN && ((u.i & 0x1) == 0)) { 00531 const rpmuint16_t *k = (const rpmuint16_t *)key; /* read 16-bit chunks */ 00532 const rpmuint8_t *k8; 00533 00534 /*----------- all but last block: aligned reads and different mixing */ 00535 while (size > 12) { 00536 a += k[0] + (((rpmuint32_t)k[1])<<16); 00537 b += k[2] + (((rpmuint32_t)k[3])<<16); 00538 c += k[4] + (((rpmuint32_t)k[5])<<16); 00539 _JLU3_MIX(a,b,c); 00540 size -= 12; 00541 k += 6; 00542 } 00543 00544 /*------------------------- handle the last (probably partial) block */ 00545 k8 = (const rpmuint8_t *)k; 00546 switch (size) { 00547 case 12: 00548 c += k[4]+(((rpmuint32_t)k[5])<<16); 00549 b += k[2]+(((rpmuint32_t)k[3])<<16); 00550 a += k[0]+(((rpmuint32_t)k[1])<<16); 00551 break; 00552 case 11: 00553 c += ((rpmuint32_t)k8[10])<<16; 00554 /*@fallthrough@*/ 00555 case 10: 00556 c += k[4]; 00557 b += k[2]+(((rpmuint32_t)k[3])<<16); 00558 a += k[0]+(((rpmuint32_t)k[1])<<16); 00559 break; 00560 case 9: 00561 c += k8[8]; 00562 /*@fallthrough@*/ 00563 case 8: 00564 b += k[2]+(((rpmuint32_t)k[3])<<16); 00565 a += k[0]+(((rpmuint32_t)k[1])<<16); 00566 break; 00567 case 7: 00568 b += ((rpmuint32_t)k8[6])<<16; 00569 /*@fallthrough@*/ 00570 case 6: 00571 b += k[2]; 00572 a += k[0]+(((rpmuint32_t)k[1])<<16); 00573 break; 00574 case 5: 00575 b += k8[4]; 00576 /*@fallthrough@*/ 00577 case 4: 00578 a += k[0]+(((rpmuint32_t)k[1])<<16); 00579 break; 00580 case 3: 00581 a += ((rpmuint32_t)k8[2])<<16; 00582 /*@fallthrough@*/ 00583 case 2: 00584 a += k[0]; 00585 break; 00586 case 1: 00587 a += k8[0]; 00588 break; 00589 case 0: 00590 goto exit; 00591 } 00592 00593 } else { /* need to read the key one byte at a time */ 00594 const rpmuint8_t *k = (const rpmuint8_t *)key; 00595 00596 /*----------- all but the last block: affect some 32 bits of (a,b,c) */ 00597 while (size > 12) { 00598 a += k[0]; 00599 a += ((rpmuint32_t)k[1])<<8; 00600 a += ((rpmuint32_t)k[2])<<16; 00601 a += ((rpmuint32_t)k[3])<<24; 00602 b += k[4]; 00603 b += ((rpmuint32_t)k[5])<<8; 00604 b += ((rpmuint32_t)k[6])<<16; 00605 b += ((rpmuint32_t)k[7])<<24; 00606 c += k[8]; 00607 c += ((rpmuint32_t)k[9])<<8; 00608 c += ((rpmuint32_t)k[10])<<16; 00609 c += ((rpmuint32_t)k[11])<<24; 00610 _JLU3_MIX(a,b,c); 00611 size -= 12; 00612 k += 12; 00613 } 00614 00615 /*---------------------------- last block: affect all 32 bits of (c) */ 00616 switch (size) { 00617 case 12: c += ((rpmuint32_t)k[11])<<24; /*@fallthrough@*/ 00618 case 11: c += ((rpmuint32_t)k[10])<<16; /*@fallthrough@*/ 00619 case 10: c += ((rpmuint32_t)k[9])<<8; /*@fallthrough@*/ 00620 case 9: c += k[8]; /*@fallthrough@*/ 00621 case 8: b += ((rpmuint32_t)k[7])<<24; /*@fallthrough@*/ 00622 case 7: b += ((rpmuint32_t)k[6])<<16; /*@fallthrough@*/ 00623 case 6: b += ((rpmuint32_t)k[5])<<8; /*@fallthrough@*/ 00624 case 5: b += k[4]; /*@fallthrough@*/ 00625 case 4: a += ((rpmuint32_t)k[3])<<24; /*@fallthrough@*/ 00626 case 3: a += ((rpmuint32_t)k[2])<<16; /*@fallthrough@*/ 00627 case 2: a += ((rpmuint32_t)k[1])<<8; /*@fallthrough@*/ 00628 case 1: a += k[0]; /*@fallthrough@*/ 00629 break; 00630 case 0: 00631 goto exit; 00632 } 00633 } 00634 00635 _JLU3_FINAL(a,b,c); 00636 00637 exit: 00638 *pc = c; 00639 *pb = b; 00640 return; 00641 } 00642 #endif /* defined(_JLU3_jlu32lpair) */ 00643 00644 #if defined(_JLU3_jlu32b) 00645 rpmuint32_t jlu32b(rpmuint32_t h, /*@null@*/ const void *key, size_t size) 00646 /*@*/; 00647 /* 00648 * jlu32b(): 00649 * This is the same as jlu32w() on big-endian machines. It is different 00650 * from jlu32l() on all machines. jlu32b() takes advantage of 00651 * big-endian byte ordering. 00652 * 00653 * @param h the previous hash, or an arbitrary value 00654 * @param *k the key, an array of rpmuint8_t values 00655 * @param size the size of the key 00656 * @return the lookup3 hash 00657 */ 00658 rpmuint32_t jlu32b(rpmuint32_t h, const void *key, size_t size) 00659 { 00660 union { const void *ptr; size_t i; } u; 00661 rpmuint32_t a = _JLU3_INIT(h, size); 00662 rpmuint32_t b = a; 00663 rpmuint32_t c = a; 00664 00665 if (key == NULL) 00666 return h; 00667 00668 u.ptr = key; 00669 if (HASH_BIG_ENDIAN && ((u.i & 0x3) == 0)) { 00670 const rpmuint32_t *k = (const rpmuint32_t *)key; /* read 32-bit chunks */ 00671 #ifdef VALGRIND 00672 const rpmuint8_t *k8; 00673 #endif 00674 00675 /*-- all but last block: aligned reads and affect 32 bits of (a,b,c) */ 00676 while (size > 12) { 00677 a += k[0]; 00678 b += k[1]; 00679 c += k[2]; 00680 _JLU3_MIX(a,b,c); 00681 size -= 12; 00682 k += 3; 00683 } 00684 00685 /*------------------------- handle the last (probably partial) block */ 00686 /* 00687 * "k[2]<<8" actually reads beyond the end of the string, but 00688 * then shifts out the part it's not allowed to read. Because the 00689 * string is aligned, the illegal read is in the same word as the 00690 * rest of the string. Every machine with memory protection I've seen 00691 * does it on word boundaries, so is OK with this. But VALGRIND will 00692 * still catch it and complain. The masking trick does make the hash 00693 * noticably faster for short strings (like English words). 00694 */ 00695 #ifndef VALGRIND 00696 00697 switch (size) { 00698 case 12: c += k[2]; b+=k[1]; a+=k[0]; break; 00699 case 11: c += k[2]&0xffffff00; b+=k[1]; a+=k[0]; break; 00700 case 10: c += k[2]&0xffff0000; b+=k[1]; a+=k[0]; break; 00701 case 9: c += k[2]&0xff000000; b+=k[1]; a+=k[0]; break; 00702 case 8: b += k[1]; a+=k[0]; break; 00703 case 7: b += k[1]&0xffffff00; a+=k[0]; break; 00704 case 6: b += k[1]&0xffff0000; a+=k[0]; break; 00705 case 5: b += k[1]&0xff000000; a+=k[0]; break; 00706 case 4: a += k[0]; break; 00707 case 3: a += k[0]&0xffffff00; break; 00708 case 2: a += k[0]&0xffff0000; break; 00709 case 1: a += k[0]&0xff000000; break; 00710 case 0: goto exit; 00711 } 00712 00713 #else /* make valgrind happy */ 00714 00715 k8 = (const rpmuint8_t *)k; 00716 switch (size) { /* all the case statements fall through */ 00717 case 12: c += k[2]; b+=k[1]; a+=k[0]; break; 00718 case 11: c += ((rpmuint32_t)k8[10])<<8; /*@fallthrough@*/ 00719 case 10: c += ((rpmuint32_t)k8[9])<<16; /*@fallthrough@*/ 00720 case 9: c += ((rpmuint32_t)k8[8])<<24; /*@fallthrough@*/ 00721 case 8: b += k[1]; a+=k[0]; break; 00722 case 7: b += ((rpmuint32_t)k8[6])<<8; /*@fallthrough@*/ 00723 case 6: b += ((rpmuint32_t)k8[5])<<16; /*@fallthrough@*/ 00724 case 5: b += ((rpmuint32_t)k8[4])<<24; /*@fallthrough@*/ 00725 case 4: a += k[0]; break; 00726 case 3: a += ((rpmuint32_t)k8[2])<<8; /*@fallthrough@*/ 00727 case 2: a += ((rpmuint32_t)k8[1])<<16; /*@fallthrough@*/ 00728 case 1: a += ((rpmuint32_t)k8[0])<<24; break; 00729 case 0: goto exit; 00730 } 00731 00732 #endif /* !VALGRIND */ 00733 00734 } else { /* need to read the key one byte at a time */ 00735 const rpmuint8_t *k = (const rpmuint8_t *)key; 00736 00737 /*----------- all but the last block: affect some 32 bits of (a,b,c) */ 00738 while (size > 12) { 00739 a += ((rpmuint32_t)k[0])<<24; 00740 a += ((rpmuint32_t)k[1])<<16; 00741 a += ((rpmuint32_t)k[2])<<8; 00742 a += ((rpmuint32_t)k[3]); 00743 b += ((rpmuint32_t)k[4])<<24; 00744 b += ((rpmuint32_t)k[5])<<16; 00745 b += ((rpmuint32_t)k[6])<<8; 00746 b += ((rpmuint32_t)k[7]); 00747 c += ((rpmuint32_t)k[8])<<24; 00748 c += ((rpmuint32_t)k[9])<<16; 00749 c += ((rpmuint32_t)k[10])<<8; 00750 c += ((rpmuint32_t)k[11]); 00751 _JLU3_MIX(a,b,c); 00752 size -= 12; 00753 k += 12; 00754 } 00755 00756 /*---------------------------- last block: affect all 32 bits of (c) */ 00757 switch (size) { /* all the case statements fall through */ 00758 case 12: c += k[11]; /*@fallthrough@*/ 00759 case 11: c += ((rpmuint32_t)k[10])<<8; /*@fallthrough@*/ 00760 case 10: c += ((rpmuint32_t)k[9])<<16; /*@fallthrough@*/ 00761 case 9: c += ((rpmuint32_t)k[8])<<24; /*@fallthrough@*/ 00762 case 8: b += k[7]; /*@fallthrough@*/ 00763 case 7: b += ((rpmuint32_t)k[6])<<8; /*@fallthrough@*/ 00764 case 6: b += ((rpmuint32_t)k[5])<<16; /*@fallthrough@*/ 00765 case 5: b += ((rpmuint32_t)k[4])<<24; /*@fallthrough@*/ 00766 case 4: a += k[3]; /*@fallthrough@*/ 00767 case 3: a += ((rpmuint32_t)k[2])<<8; /*@fallthrough@*/ 00768 case 2: a += ((rpmuint32_t)k[1])<<16; /*@fallthrough@*/ 00769 case 1: a += ((rpmuint32_t)k[0])<<24; /*@fallthrough@*/ 00770 break; 00771 case 0: 00772 goto exit; 00773 } 00774 } 00775 00776 _JLU3_FINAL(a,b,c); 00777 00778 exit: 00779 return c; 00780 } 00781 #endif /* defined(_JLU3_jlu32b) */ 00782 00783 #if defined(_JLU3_SELFTEST) 00784 00785 /* used for timings */ 00786 static void driver1(void) 00787 /*@*/ 00788 { 00789 rpmuint8_t buf[256]; 00790 rpmuint32_t i; 00791 rpmuint32_t h=0; 00792 time_t a,z; 00793 00794 time(&a); 00795 for (i=0; i<256; ++i) buf[i] = 'x'; 00796 for (i=0; i<1; ++i) { 00797 h = jlu32l(h, &buf[0], sizeof(buf[0])); 00798 } 00799 time(&z); 00800 if (z-a > 0) printf("time %d %.8x\n", (int)(z-a), h); 00801 } 00802 00803 /* check that every input bit changes every output bit half the time */ 00804 #define HASHSTATE 1 00805 #define HASHLEN 1 00806 #define MAXPAIR 60 00807 #define MAXLEN 70 00808 static void driver2(void) 00809 /*@*/ 00810 { 00811 rpmuint8_t qa[MAXLEN+1], qb[MAXLEN+2], *a = &qa[0], *b = &qb[1]; 00812 rpmuint32_t c[HASHSTATE], d[HASHSTATE], i=0, j=0, k, l, m=0, z; 00813 rpmuint32_t e[HASHSTATE],f[HASHSTATE],g[HASHSTATE],h[HASHSTATE]; 00814 rpmuint32_t x[HASHSTATE],y[HASHSTATE]; 00815 rpmuint32_t hlen; 00816 00817 printf("No more than %d trials should ever be needed \n",MAXPAIR/2); 00818 for (hlen=0; hlen < MAXLEN; ++hlen) { 00819 z=0; 00820 for (i=0; i<hlen; ++i) { /*-------------- for each input byte, */ 00821 for (j=0; j<8; ++j) { /*--------------- for each input bit, */ 00822 for (m=1; m<8; ++m) { /*--- for serveral possible initvals, */ 00823 for (l=0; l<HASHSTATE; ++l) 00824 e[l]=f[l]=g[l]=h[l]=x[l]=y[l]=~((rpmuint32_t)0); 00825 00826 /* check that every output bit is affected by that input bit */ 00827 for (k=0; k<MAXPAIR; k+=2) { 00828 rpmuint32_t finished=1; 00829 /* keys have one bit different */ 00830 for (l=0; l<hlen+1; ++l) {a[l] = b[l] = (rpmuint8_t)0;} 00831 /* have a and b be two keys differing in only one bit */ 00832 a[i] ^= (k<<j); 00833 a[i] ^= (k>>(8-j)); 00834 c[0] = jlu32l(m, a, hlen); 00835 b[i] ^= ((k+1)<<j); 00836 b[i] ^= ((k+1)>>(8-j)); 00837 d[0] = jlu32l(m, b, hlen); 00838 /* check every bit is 1, 0, set, and not set at least once */ 00839 for (l=0; l<HASHSTATE; ++l) { 00840 e[l] &= (c[l]^d[l]); 00841 f[l] &= ~(c[l]^d[l]); 00842 g[l] &= c[l]; 00843 h[l] &= ~c[l]; 00844 x[l] &= d[l]; 00845 y[l] &= ~d[l]; 00846 if (e[l]|f[l]|g[l]|h[l]|x[l]|y[l]) finished=0; 00847 } 00848 if (finished) break; 00849 } 00850 if (k>z) z=k; 00851 if (k == MAXPAIR) { 00852 printf("Some bit didn't change: "); 00853 printf("%.8x %.8x %.8x %.8x %.8x %.8x ", 00854 e[0],f[0],g[0],h[0],x[0],y[0]); 00855 printf("i %d j %d m %d len %d\n", i, j, m, hlen); 00856 } 00857 if (z == MAXPAIR) goto done; 00858 } 00859 } 00860 } 00861 done: 00862 if (z < MAXPAIR) { 00863 printf("Mix success %2d bytes %2d initvals ",i,m); 00864 printf("required %d trials\n", z/2); 00865 } 00866 } 00867 printf("\n"); 00868 } 00869 00870 /* Check for reading beyond the end of the buffer and alignment problems */ 00871 static void driver3(void) 00872 /*@*/ 00873 { 00874 rpmuint8_t buf[MAXLEN+20], *b; 00875 rpmuint32_t len; 00876 rpmuint8_t q[] = "This is the time for all good men to come to the aid of their country..."; 00877 rpmuint32_t h; 00878 rpmuint8_t qq[] = "xThis is the time for all good men to come to the aid of their country..."; 00879 rpmuint32_t i; 00880 rpmuint8_t qqq[] = "xxThis is the time for all good men to come to the aid of their country..."; 00881 rpmuint32_t j; 00882 rpmuint8_t qqqq[] = "xxxThis is the time for all good men to come to the aid of their country..."; 00883 rpmuint32_t ref,x,y; 00884 rpmuint8_t *p; 00885 rpmuint32_t m = 13; 00886 00887 printf("Endianness. These lines should all be the same (for values filled in):\n"); 00888 printf("%.8x %.8x %.8x\n", 00889 jlu32w(m, (const rpmuint32_t *)q, (sizeof(q)-1)/4), 00890 jlu32w(m, (const rpmuint32_t *)q, (sizeof(q)-5)/4), 00891 jlu32w(m, (const rpmuint32_t *)q, (sizeof(q)-9)/4)); 00892 p = q; 00893 printf("%.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x\n", 00894 jlu32l(m, p, sizeof(q)-1), jlu32l(m, p, sizeof(q)-2), 00895 jlu32l(m, p, sizeof(q)-3), jlu32l(m, p, sizeof(q)-4), 00896 jlu32l(m, p, sizeof(q)-5), jlu32l(m, p, sizeof(q)-6), 00897 jlu32l(m, p, sizeof(q)-7), jlu32l(m, p, sizeof(q)-8), 00898 jlu32l(m, p, sizeof(q)-9), jlu32l(m, p, sizeof(q)-10), 00899 jlu32l(m, p, sizeof(q)-11), jlu32l(m, p, sizeof(q)-12)); 00900 p = &qq[1]; 00901 printf("%.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x\n", 00902 jlu32l(m, p, sizeof(q)-1), jlu32l(m, p, sizeof(q)-2), 00903 jlu32l(m, p, sizeof(q)-3), jlu32l(m, p, sizeof(q)-4), 00904 jlu32l(m, p, sizeof(q)-5), jlu32l(m, p, sizeof(q)-6), 00905 jlu32l(m, p, sizeof(q)-7), jlu32l(m, p, sizeof(q)-8), 00906 jlu32l(m, p, sizeof(q)-9), jlu32l(m, p, sizeof(q)-10), 00907 jlu32l(m, p, sizeof(q)-11), jlu32l(m, p, sizeof(q)-12)); 00908 p = &qqq[2]; 00909 printf("%.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x\n", 00910 jlu32l(m, p, sizeof(q)-1), jlu32l(m, p, sizeof(q)-2), 00911 jlu32l(m, p, sizeof(q)-3), jlu32l(m, p, sizeof(q)-4), 00912 jlu32l(m, p, sizeof(q)-5), jlu32l(m, p, sizeof(q)-6), 00913 jlu32l(m, p, sizeof(q)-7), jlu32l(m, p, sizeof(q)-8), 00914 jlu32l(m, p, sizeof(q)-9), jlu32l(m, p, sizeof(q)-10), 00915 jlu32l(m, p, sizeof(q)-11), jlu32l(m, p, sizeof(q)-12)); 00916 p = &qqqq[3]; 00917 printf("%.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x\n", 00918 jlu32l(m, p, sizeof(q)-1), jlu32l(m, p, sizeof(q)-2), 00919 jlu32l(m, p, sizeof(q)-3), jlu32l(m, p, sizeof(q)-4), 00920 jlu32l(m, p, sizeof(q)-5), jlu32l(m, p, sizeof(q)-6), 00921 jlu32l(m, p, sizeof(q)-7), jlu32l(m, p, sizeof(q)-8), 00922 jlu32l(m, p, sizeof(q)-9), jlu32l(m, p, sizeof(q)-10), 00923 jlu32l(m, p, sizeof(q)-11), jlu32l(m, p, sizeof(q)-12)); 00924 printf("\n"); 00925 for (h=0, b=buf+1; h<8; ++h, ++b) { 00926 for (i=0; i<MAXLEN; ++i) { 00927 len = i; 00928 for (j=0; j<i; ++j) 00929 *(b+j)=0; 00930 00931 /* these should all be equal */ 00932 m = 1; 00933 ref = jlu32l(m, b, len); 00934 *(b+i)=(rpmuint8_t)~0; 00935 *(b-1)=(rpmuint8_t)~0; 00936 x = jlu32l(m, b, len); 00937 y = jlu32l(m, b, len); 00938 if ((ref != x) || (ref != y)) 00939 printf("alignment error: %.8x %.8x %.8x %d %d\n",ref,x,y, h, i); 00940 } 00941 } 00942 } 00943 00944 /* check for problems with nulls */ 00945 static void driver4(void) 00946 /*@*/ 00947 { 00948 rpmuint8_t buf[1]; 00949 rpmuint32_t h; 00950 rpmuint32_t i; 00951 rpmuint32_t state[HASHSTATE]; 00952 00953 buf[0] = ~0; 00954 for (i=0; i<HASHSTATE; ++i) 00955 state[i] = 1; 00956 printf("These should all be different\n"); 00957 h = 0; 00958 for (i=0; i<8; ++i) { 00959 h = jlu32l(h, buf, 0); 00960 printf("%2ld 0-byte strings, hash is %.8x\n", (long)i, h); 00961 } 00962 } 00963 00964 00965 int main(int argc, char ** argv) 00966 { 00967 driver1(); /* test that the key is hashed: used for timings */ 00968 driver2(); /* test that whole key is hashed thoroughly */ 00969 driver3(); /* test that nothing but the key is hashed */ 00970 driver4(); /* test hashing multiple buffers (all buffers are null) */ 00971 return 1; 00972 } 00973 00974 #endif /* _JLU3_SELFTEST */