rpm 5.3.12
|
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 #ifdef WITH_VALGRIND 00042 #if !defined(RUNNING_ON_VALGRIND) 00043 #define RUNNING_ON_VALGRIND 0 00044 #endif 00045 #if defined(__GNUC__) && (__GNUC__ > 2) && defined(__OPTIMIZE__) 00046 # define UNLIKELY(value) __builtin_expect((value), 0) && (value > 0 || (value = RUNNING_ON_VALGRIND)) 00047 #else 00048 # define UNLIKELY(value) (value) && (value > 0 || (value = RUNNING_ON_VALGRIND)) 00049 #endif 00050 static int _running_on_valgrind = -1; 00051 #endif 00052 00053 #if defined(_JLU3_SELFTEST) 00054 # define _JLU3_jlu32w 1 00055 # define _JLU3_jlu32l 1 00056 # define _JLU3_jlu32lpair 1 00057 # define _JLU3_jlu32b 1 00058 #endif 00059 00060 /*@-redef@*/ 00061 /*@unchecked@*/ 00062 static const union _dbswap { 00063 const rpmuint32_t ui; 00064 const unsigned char uc[4]; 00065 } endian = { .ui = 0x11223344 }; 00066 # define HASH_LITTLE_ENDIAN (endian.uc[0] == (unsigned char) 0x44) 00067 # define HASH_BIG_ENDIAN (endian.uc[0] == (unsigned char) 0x11) 00068 /*@=redef@*/ 00069 00070 #ifndef ROTL32 00071 # define ROTL32(x, s) (((x) << (s)) | ((x) >> (32 - (s)))) 00072 #endif 00073 00074 /* NOTE: The _size parameter should be in bytes. */ 00075 #define _JLU3_INIT(_h, _size) (0xdeadbeef + ((rpmuint32_t)(_size)) + (_h)) 00076 00077 /* -------------------------------------------------------------------- */ 00078 /* 00079 * _JLU3_MIX -- mix 3 32-bit values reversibly. 00080 * 00081 * This is reversible, so any information in (a,b,c) before _JLU3_MIX() is 00082 * still in (a,b,c) after _JLU3_MIX(). 00083 * 00084 * If four pairs of (a,b,c) inputs are run through _JLU3_MIX(), or through 00085 * _JLU3_MIX() in reverse, there are at least 32 bits of the output that 00086 * are sometimes the same for one pair and different for another pair. 00087 * This was tested for: 00088 * * pairs that differed by one bit, by two bits, in any combination 00089 * of top bits of (a,b,c), or in any combination of bottom bits of 00090 * (a,b,c). 00091 * * "differ" is defined as +, -, ^, or ~^. For + and -, I transformed 00092 * the output delta to a Gray code (a^(a>>1)) so a string of 1's (as 00093 * is commonly produced by subtraction) look like a single 1-bit 00094 * difference. 00095 * * the base values were pseudorandom, all zero but one bit set, or 00096 * all zero plus a counter that starts at zero. 00097 * 00098 * Some k values for my "a-=c; a^=ROTL32(c,k); c+=b;" arrangement that 00099 * satisfy this are 00100 * 4 6 8 16 19 4 00101 * 9 15 3 18 27 15 00102 * 14 9 3 7 17 3 00103 * Well, "9 15 3 18 27 15" didn't quite get 32 bits diffing 00104 * for "differ" defined as + with a one-bit base and a two-bit delta. I 00105 * used http://burtleburtle.net/bob/hash/avalanche.html to choose 00106 * the operations, constants, and arrangements of the variables. 00107 * 00108 * This does not achieve avalanche. There are input bits of (a,b,c) 00109 * that fail to affect some output bits of (a,b,c), especially of a. The 00110 * most thoroughly mixed value is c, but it doesn't really even achieve 00111 * avalanche in c. 00112 * 00113 * This allows some parallelism. Read-after-writes are good at doubling 00114 * the number of bits affected, so the goal of mixing pulls in the opposite 00115 * direction as the goal of parallelism. I did what I could. Rotates 00116 * seem to cost as much as shifts on every machine I could lay my hands 00117 * on, and rotates are much kinder to the top and bottom bits, so I used 00118 * rotates. 00119 */ 00120 /* -------------------------------------------------------------------- */ 00121 #define _JLU3_MIX(a,b,c) \ 00122 { \ 00123 a -= c; a ^= ROTL32(c, 4); c += b; \ 00124 b -= a; b ^= ROTL32(a, 6); a += c; \ 00125 c -= b; c ^= ROTL32(b, 8); b += a; \ 00126 a -= c; a ^= ROTL32(c,16); c += b; \ 00127 b -= a; b ^= ROTL32(a,19); a += c; \ 00128 c -= b; c ^= ROTL32(b, 4); b += a; \ 00129 } 00130 00131 /* -------------------------------------------------------------------- */ 00155 /* -------------------------------------------------------------------- */ 00156 #define _JLU3_FINAL(a,b,c) \ 00157 { \ 00158 c ^= b; c -= ROTL32(b,14); \ 00159 a ^= c; a -= ROTL32(c,11); \ 00160 b ^= a; b -= ROTL32(a,25); \ 00161 c ^= b; c -= ROTL32(b,16); \ 00162 a ^= c; a -= ROTL32(c,4); \ 00163 b ^= a; b -= ROTL32(a,14); \ 00164 c ^= b; c -= ROTL32(b,24); \ 00165 } 00166 00167 #if defined(_JLU3_jlu32w) 00168 rpmuint32_t jlu32w(rpmuint32_t h, /*@null@*/ const rpmuint32_t *k, size_t size) 00169 /*@*/; 00170 /* -------------------------------------------------------------------- */ 00187 /* -------------------------------------------------------------------- */ 00188 rpmuint32_t jlu32w(rpmuint32_t h, const rpmuint32_t *k, size_t size) 00189 { 00190 rpmuint32_t a = _JLU3_INIT(h, (size * sizeof(*k))); 00191 rpmuint32_t b = a; 00192 rpmuint32_t c = a; 00193 00194 if (k == NULL) 00195 goto exit; 00196 00197 /*----------------------------------------------- handle most of the key */ 00198 while (size > 3) { 00199 a += k[0]; 00200 b += k[1]; 00201 c += k[2]; 00202 _JLU3_MIX(a,b,c); 00203 size -= 3; 00204 k += 3; 00205 } 00206 00207 /*----------------------------------------- handle the last 3 rpmuint32_t's */ 00208 switch (size) { 00209 case 3 : c+=k[2]; 00210 case 2 : b+=k[1]; 00211 case 1 : a+=k[0]; 00212 _JLU3_FINAL(a,b,c); 00213 /*@fallthrough@*/ 00214 case 0: 00215 break; 00216 } 00217 /*---------------------------------------------------- report the result */ 00218 exit: 00219 return c; 00220 } 00221 #endif /* defined(_JLU3_jlu32w) */ 00222 00223 #if defined(_JLU3_jlu32l) 00224 rpmuint32_t jlu32l(rpmuint32_t h, const void *key, size_t size) 00225 /*@*/; 00226 /* -------------------------------------------------------------------- */ 00227 /* 00228 * jlu32l() -- hash a variable-length key into a 32-bit value 00229 * h : can be any 4-byte value 00230 * k : the key (the unaligned variable-length array of bytes) 00231 * size : the size of the key, counting by bytes 00232 * Returns a 32-bit value. Every bit of the key affects every bit of 00233 * the return value. Two keys differing by one or two bits will have 00234 * totally different hash values. 00235 * 00236 * The best hash table sizes are powers of 2. There is no need to do 00237 * mod a prime (mod is sooo slow!). If you need less than 32 bits, 00238 * use a bitmask. For example, if you need only 10 bits, do 00239 * h = (h & hashmask(10)); 00240 * In which case, the hash table should have hashsize(10) elements. 00241 * 00242 * If you are hashing n strings (rpmuint8_t **)k, do it like this: 00243 * for (i=0, h=0; i<n; ++i) h = jlu32l(h, k[i], len[i]); 00244 * 00245 * By Bob Jenkins, 2006. bob_jenkins@burtleburtle.net. You may use this 00246 * code any way you wish, private, educational, or commercial. It's free. 00247 * 00248 * Use for hash table lookup, or anything where one collision in 2^^32 is 00249 * acceptable. Do NOT use for cryptographic purposes. 00250 * 00251 * @param h the previous hash, or an arbitrary value 00252 * @param *k the key, an array of rpmuint8_t values 00253 * @param size the size of the key 00254 * @return the lookup3 hash 00255 */ 00256 /* -------------------------------------------------------------------- */ 00257 rpmuint32_t jlu32l(rpmuint32_t h, const void *key, size_t size) 00258 { 00259 union { const void *ptr; size_t i; } u; 00260 rpmuint32_t a = _JLU3_INIT(h, size); 00261 rpmuint32_t b = a; 00262 rpmuint32_t c = a; 00263 00264 if (key == NULL) 00265 goto exit; 00266 00267 u.ptr = key; 00268 if (HASH_LITTLE_ENDIAN && ((u.i & 0x3) == 0)) { 00269 const rpmuint32_t *k = (const rpmuint32_t *)key; /* read 32-bit chunks */ 00270 00271 /*------ all but last block: aligned reads and affect 32 bits of (a,b,c) */ 00272 while (size > 12) { 00273 a += k[0]; 00274 b += k[1]; 00275 c += k[2]; 00276 _JLU3_MIX(a,b,c); 00277 size -= 12; 00278 k += 3; 00279 } 00280 00281 /*------------------------- handle the last (probably partial) block */ 00282 /* 00283 * "k[2]&0xffffff" actually reads beyond the end of the string, but 00284 * then masks off the part it's not allowed to read. Because the 00285 * string is aligned, the masked-off tail is in the same word as the 00286 * rest of the string. Every machine with memory protection I've seen 00287 * does it on word boundaries, so is OK with this. But VALGRIND will 00288 * still catch it and complain. The masking trick does make the hash 00289 * noticably faster for short strings (like English words). 00290 */ 00291 #ifdef WITH_VALGRIND 00292 if (UNLIKELY(_running_on_valgrind)) { 00293 const rpmuint8_t * k8 = (const rpmuint8_t *)k; 00294 00295 switch (size) { 00296 case 12: c += k[2]; b+=k[1]; a+=k[0]; break; 00297 case 11: c += ((rpmuint32_t)k8[10])<<16; /*@fallthrough@*/ 00298 case 10: c += ((rpmuint32_t)k8[9])<<8; /*@fallthrough@*/ 00299 case 9: c += k8[8]; /*@fallthrough@*/ 00300 case 8: b += k[1]; a+=k[0]; break; 00301 case 7: b += ((rpmuint32_t)k8[6])<<16; /*@fallthrough@*/ 00302 case 6: b += ((rpmuint32_t)k8[5])<<8; /*@fallthrough@*/ 00303 case 5: b += k8[4]; /*@fallthrough@*/ 00304 case 4: a += k[0]; break; 00305 case 3: a += ((rpmuint32_t)k8[2])<<16; /*@fallthrough@*/ 00306 case 2: a += ((rpmuint32_t)k8[1])<<8; /*@fallthrough@*/ 00307 case 1: a += k8[0]; break; 00308 case 0: goto exit; 00309 } 00310 00311 } else 00312 #endif 00313 { 00314 switch (size) { 00315 case 12: c += k[2]; b+=k[1]; a+=k[0]; break; 00316 case 11: c += k[2]&0xffffff; b+=k[1]; a+=k[0]; break; 00317 case 10: c += k[2]&0xffff; b+=k[1]; a+=k[0]; break; 00318 case 9: c += k[2]&0xff; b+=k[1]; a+=k[0]; break; 00319 case 8: b += k[1]; a+=k[0]; break; 00320 case 7: b += k[1]&0xffffff; a+=k[0]; break; 00321 case 6: b += k[1]&0xffff; a+=k[0]; break; 00322 case 5: b += k[1]&0xff; a+=k[0]; break; 00323 case 4: a += k[0]; break; 00324 case 3: a += k[0]&0xffffff; break; 00325 case 2: a += k[0]&0xffff; break; 00326 case 1: a += k[0]&0xff; break; 00327 case 0: goto exit; 00328 } 00329 } 00330 } else if (HASH_LITTLE_ENDIAN && ((u.i & 0x1) == 0)) { 00331 const rpmuint16_t *k = (const rpmuint16_t *)key; /* read 16-bit chunks */ 00332 const rpmuint8_t *k8; 00333 00334 /*----------- all but last block: aligned reads and different mixing */ 00335 while (size > 12) { 00336 a += k[0] + (((rpmuint32_t)k[1])<<16); 00337 b += k[2] + (((rpmuint32_t)k[3])<<16); 00338 c += k[4] + (((rpmuint32_t)k[5])<<16); 00339 _JLU3_MIX(a,b,c); 00340 size -= 12; 00341 k += 6; 00342 } 00343 00344 /*------------------------- handle the last (probably partial) block */ 00345 k8 = (const rpmuint8_t *)k; 00346 switch (size) { 00347 case 12: 00348 c += k[4]+(((rpmuint32_t)k[5])<<16); 00349 b += k[2]+(((rpmuint32_t)k[3])<<16); 00350 a += k[0]+(((rpmuint32_t)k[1])<<16); 00351 break; 00352 case 11: 00353 c += ((rpmuint32_t)k8[10])<<16; 00354 /*@fallthrough@*/ 00355 case 10: 00356 c += (rpmuint32_t)k[4]; 00357 b += k[2]+(((rpmuint32_t)k[3])<<16); 00358 a += k[0]+(((rpmuint32_t)k[1])<<16); 00359 break; 00360 case 9: 00361 c += (rpmuint32_t)k8[8]; 00362 /*@fallthrough@*/ 00363 case 8: 00364 b += k[2]+(((rpmuint32_t)k[3])<<16); 00365 a += k[0]+(((rpmuint32_t)k[1])<<16); 00366 break; 00367 case 7: 00368 b += ((rpmuint32_t)k8[6])<<16; 00369 /*@fallthrough@*/ 00370 case 6: 00371 b += (rpmuint32_t)k[2]; 00372 a += k[0]+(((rpmuint32_t)k[1])<<16); 00373 break; 00374 case 5: 00375 b += (rpmuint32_t)k8[4]; 00376 /*@fallthrough@*/ 00377 case 4: 00378 a += k[0]+(((rpmuint32_t)k[1])<<16); 00379 break; 00380 case 3: 00381 a += ((rpmuint32_t)k8[2])<<16; 00382 /*@fallthrough@*/ 00383 case 2: 00384 a += (rpmuint32_t)k[0]; 00385 break; 00386 case 1: 00387 a += (rpmuint32_t)k8[0]; 00388 break; 00389 case 0: 00390 goto exit; 00391 } 00392 00393 } else { /* need to read the key one byte at a time */ 00394 const rpmuint8_t *k = (const rpmuint8_t *)key; 00395 00396 /*----------- all but the last block: affect some 32 bits of (a,b,c) */ 00397 while (size > 12) { 00398 a += (rpmuint32_t)k[0]; 00399 a += ((rpmuint32_t)k[1])<<8; 00400 a += ((rpmuint32_t)k[2])<<16; 00401 a += ((rpmuint32_t)k[3])<<24; 00402 b += (rpmuint32_t)k[4]; 00403 b += ((rpmuint32_t)k[5])<<8; 00404 b += ((rpmuint32_t)k[6])<<16; 00405 b += ((rpmuint32_t)k[7])<<24; 00406 c += (rpmuint32_t)k[8]; 00407 c += ((rpmuint32_t)k[9])<<8; 00408 c += ((rpmuint32_t)k[10])<<16; 00409 c += ((rpmuint32_t)k[11])<<24; 00410 _JLU3_MIX(a,b,c); 00411 size -= 12; 00412 k += 12; 00413 } 00414 00415 /*---------------------------- last block: affect all 32 bits of (c) */ 00416 switch (size) { 00417 case 12: c += ((rpmuint32_t)k[11])<<24; /*@fallthrough@*/ 00418 case 11: c += ((rpmuint32_t)k[10])<<16; /*@fallthrough@*/ 00419 case 10: c += ((rpmuint32_t)k[9])<<8; /*@fallthrough@*/ 00420 case 9: c += (rpmuint32_t)k[8]; /*@fallthrough@*/ 00421 case 8: b += ((rpmuint32_t)k[7])<<24; /*@fallthrough@*/ 00422 case 7: b += ((rpmuint32_t)k[6])<<16; /*@fallthrough@*/ 00423 case 6: b += ((rpmuint32_t)k[5])<<8; /*@fallthrough@*/ 00424 case 5: b += (rpmuint32_t)k[4]; /*@fallthrough@*/ 00425 case 4: a += ((rpmuint32_t)k[3])<<24; /*@fallthrough@*/ 00426 case 3: a += ((rpmuint32_t)k[2])<<16; /*@fallthrough@*/ 00427 case 2: a += ((rpmuint32_t)k[1])<<8; /*@fallthrough@*/ 00428 case 1: a += (rpmuint32_t)k[0]; 00429 break; 00430 case 0: 00431 goto exit; 00432 } 00433 } 00434 00435 _JLU3_FINAL(a,b,c); 00436 00437 exit: 00438 return c; 00439 } 00440 #endif /* defined(_JLU3_jlu32l) */ 00441 00442 #if defined(_JLU3_jlu32lpair) 00443 void jlu32lpair(/*@null@*/ const void *key, size_t size, 00444 rpmuint32_t *pc, rpmuint32_t *pb) 00445 /*@modifies *pc, *pb@*/; 00462 void jlu32lpair(const void *key, size_t size, rpmuint32_t *pc, rpmuint32_t *pb) 00463 { 00464 union { const void *ptr; size_t i; } u; 00465 rpmuint32_t a = _JLU3_INIT(*pc, size); 00466 rpmuint32_t b = a; 00467 rpmuint32_t c = a; 00468 00469 if (key == NULL) 00470 goto exit; 00471 00472 c += *pb; /* Add the secondary hash. */ 00473 00474 u.ptr = key; 00475 if (HASH_LITTLE_ENDIAN && ((u.i & 0x3) == 0)) { 00476 const rpmuint32_t *k = (const rpmuint32_t *)key; /* read 32-bit chunks */ 00477 00478 /*-- all but last block: aligned reads and affect 32 bits of (a,b,c) */ 00479 while (size > 12) { 00480 a += k[0]; 00481 b += k[1]; 00482 c += k[2]; 00483 _JLU3_MIX(a,b,c); 00484 size -= 12; 00485 k += 3; 00486 } 00487 /*------------------------- handle the last (probably partial) block */ 00488 /* 00489 * "k[2]&0xffffff" actually reads beyond the end of the string, but 00490 * then masks off the part it's not allowed to read. Because the 00491 * string is aligned, the masked-off tail is in the same word as the 00492 * rest of the string. Every machine with memory protection I've seen 00493 * does it on word boundaries, so is OK with this. But VALGRIND will 00494 * still catch it and complain. The masking trick does make the hash 00495 * noticably faster for short strings (like English words). 00496 */ 00497 #ifdef WITH_VALGRIND 00498 if (UNLIKELY(_running_on_valgrind)) { 00499 const rpmuint8_t * k8 = (const rpmuint8_t *)k; 00500 00501 switch (size) { 00502 case 12: c += k[2]; b+=k[1]; a+=k[0]; break; 00503 case 11: c += ((rpmuint32_t)k8[10])<<16; /*@fallthrough@*/ 00504 case 10: c += ((rpmuint32_t)k8[9])<<8; /*@fallthrough@*/ 00505 case 9: c += k8[8]; /*@fallthrough@*/ 00506 case 8: b += k[1]; a+=k[0]; break; 00507 case 7: b += ((rpmuint32_t)k8[6])<<16; /*@fallthrough@*/ 00508 case 6: b += ((rpmuint32_t)k8[5])<<8; /*@fallthrough@*/ 00509 case 5: b += k8[4]; /*@fallthrough@*/ 00510 case 4: a += k[0]; break; 00511 case 3: a += ((rpmuint32_t)k8[2])<<16; /*@fallthrough@*/ 00512 case 2: a += ((rpmuint32_t)k8[1])<<8; /*@fallthrough@*/ 00513 case 1: a += k8[0]; break; 00514 case 0: goto exit; 00515 } 00516 00517 } else 00518 #endif 00519 { 00520 switch (size) { 00521 case 12: c += k[2]; b+=k[1]; a+=k[0]; break; 00522 case 11: c += k[2]&0xffffff; b+=k[1]; a+=k[0]; break; 00523 case 10: c += k[2]&0xffff; b+=k[1]; a+=k[0]; break; 00524 case 9: c += k[2]&0xff; b+=k[1]; a+=k[0]; break; 00525 case 8: b += k[1]; a+=k[0]; break; 00526 case 7: b += k[1]&0xffffff; a+=k[0]; break; 00527 case 6: b += k[1]&0xffff; a+=k[0]; break; 00528 case 5: b += k[1]&0xff; a+=k[0]; break; 00529 case 4: a += k[0]; break; 00530 case 3: a += k[0]&0xffffff; break; 00531 case 2: a += k[0]&0xffff; break; 00532 case 1: a += k[0]&0xff; break; 00533 case 0: goto exit; 00534 } 00535 } 00536 } else if (HASH_LITTLE_ENDIAN && ((u.i & 0x1) == 0)) { 00537 const rpmuint16_t *k = (const rpmuint16_t *)key; /* read 16-bit chunks */ 00538 const rpmuint8_t *k8; 00539 00540 /*----------- all but last block: aligned reads and different mixing */ 00541 while (size > 12) { 00542 a += k[0] + (((rpmuint32_t)k[1])<<16); 00543 b += k[2] + (((rpmuint32_t)k[3])<<16); 00544 c += k[4] + (((rpmuint32_t)k[5])<<16); 00545 _JLU3_MIX(a,b,c); 00546 size -= 12; 00547 k += 6; 00548 } 00549 00550 /*------------------------- handle the last (probably partial) block */ 00551 k8 = (const rpmuint8_t *)k; 00552 switch (size) { 00553 case 12: 00554 c += k[4]+(((rpmuint32_t)k[5])<<16); 00555 b += k[2]+(((rpmuint32_t)k[3])<<16); 00556 a += k[0]+(((rpmuint32_t)k[1])<<16); 00557 break; 00558 case 11: 00559 c += ((rpmuint32_t)k8[10])<<16; 00560 /*@fallthrough@*/ 00561 case 10: 00562 c += k[4]; 00563 b += k[2]+(((rpmuint32_t)k[3])<<16); 00564 a += k[0]+(((rpmuint32_t)k[1])<<16); 00565 break; 00566 case 9: 00567 c += k8[8]; 00568 /*@fallthrough@*/ 00569 case 8: 00570 b += k[2]+(((rpmuint32_t)k[3])<<16); 00571 a += k[0]+(((rpmuint32_t)k[1])<<16); 00572 break; 00573 case 7: 00574 b += ((rpmuint32_t)k8[6])<<16; 00575 /*@fallthrough@*/ 00576 case 6: 00577 b += k[2]; 00578 a += k[0]+(((rpmuint32_t)k[1])<<16); 00579 break; 00580 case 5: 00581 b += k8[4]; 00582 /*@fallthrough@*/ 00583 case 4: 00584 a += k[0]+(((rpmuint32_t)k[1])<<16); 00585 break; 00586 case 3: 00587 a += ((rpmuint32_t)k8[2])<<16; 00588 /*@fallthrough@*/ 00589 case 2: 00590 a += k[0]; 00591 break; 00592 case 1: 00593 a += k8[0]; 00594 break; 00595 case 0: 00596 goto exit; 00597 } 00598 00599 } else { /* need to read the key one byte at a time */ 00600 const rpmuint8_t *k = (const rpmuint8_t *)key; 00601 00602 /*----------- all but the last block: affect some 32 bits of (a,b,c) */ 00603 while (size > 12) { 00604 a += k[0]; 00605 a += ((rpmuint32_t)k[1])<<8; 00606 a += ((rpmuint32_t)k[2])<<16; 00607 a += ((rpmuint32_t)k[3])<<24; 00608 b += k[4]; 00609 b += ((rpmuint32_t)k[5])<<8; 00610 b += ((rpmuint32_t)k[6])<<16; 00611 b += ((rpmuint32_t)k[7])<<24; 00612 c += k[8]; 00613 c += ((rpmuint32_t)k[9])<<8; 00614 c += ((rpmuint32_t)k[10])<<16; 00615 c += ((rpmuint32_t)k[11])<<24; 00616 _JLU3_MIX(a,b,c); 00617 size -= 12; 00618 k += 12; 00619 } 00620 00621 /*---------------------------- last block: affect all 32 bits of (c) */ 00622 switch (size) { 00623 case 12: c += ((rpmuint32_t)k[11])<<24; /*@fallthrough@*/ 00624 case 11: c += ((rpmuint32_t)k[10])<<16; /*@fallthrough@*/ 00625 case 10: c += ((rpmuint32_t)k[9])<<8; /*@fallthrough@*/ 00626 case 9: c += k[8]; /*@fallthrough@*/ 00627 case 8: b += ((rpmuint32_t)k[7])<<24; /*@fallthrough@*/ 00628 case 7: b += ((rpmuint32_t)k[6])<<16; /*@fallthrough@*/ 00629 case 6: b += ((rpmuint32_t)k[5])<<8; /*@fallthrough@*/ 00630 case 5: b += k[4]; /*@fallthrough@*/ 00631 case 4: a += ((rpmuint32_t)k[3])<<24; /*@fallthrough@*/ 00632 case 3: a += ((rpmuint32_t)k[2])<<16; /*@fallthrough@*/ 00633 case 2: a += ((rpmuint32_t)k[1])<<8; /*@fallthrough@*/ 00634 case 1: a += k[0]; /*@fallthrough@*/ 00635 break; 00636 case 0: 00637 goto exit; 00638 } 00639 } 00640 00641 _JLU3_FINAL(a,b,c); 00642 00643 exit: 00644 *pc = c; 00645 *pb = b; 00646 return; 00647 } 00648 #endif /* defined(_JLU3_jlu32lpair) */ 00649 00650 #if defined(_JLU3_jlu32b) 00651 rpmuint32_t jlu32b(rpmuint32_t h, /*@null@*/ const void *key, size_t size) 00652 /*@*/; 00653 /* 00654 * jlu32b(): 00655 * This is the same as jlu32w() on big-endian machines. It is different 00656 * from jlu32l() on all machines. jlu32b() takes advantage of 00657 * big-endian byte ordering. 00658 * 00659 * @param h the previous hash, or an arbitrary value 00660 * @param *k the key, an array of rpmuint8_t values 00661 * @param size the size of the key 00662 * @return the lookup3 hash 00663 */ 00664 rpmuint32_t jlu32b(rpmuint32_t h, const void *key, size_t size) 00665 { 00666 union { const void *ptr; size_t i; } u; 00667 rpmuint32_t a = _JLU3_INIT(h, size); 00668 rpmuint32_t b = a; 00669 rpmuint32_t c = a; 00670 00671 if (key == NULL) 00672 return h; 00673 00674 u.ptr = key; 00675 if (HASH_BIG_ENDIAN && ((u.i & 0x3) == 0)) { 00676 const rpmuint32_t *k = (const rpmuint32_t *)key; /* read 32-bit chunks */ 00677 00678 /*-- all but last block: aligned reads and affect 32 bits of (a,b,c) */ 00679 while (size > 12) { 00680 a += k[0]; 00681 b += k[1]; 00682 c += k[2]; 00683 _JLU3_MIX(a,b,c); 00684 size -= 12; 00685 k += 3; 00686 } 00687 00688 /*------------------------- handle the last (probably partial) block */ 00689 /* 00690 * "k[2]<<8" actually reads beyond the end of the string, but 00691 * then shifts out the part it's not allowed to read. Because the 00692 * string is aligned, the illegal read is in the same word as the 00693 * rest of the string. Every machine with memory protection I've seen 00694 * does it on word boundaries, so is OK with this. But VALGRIND will 00695 * still catch it and complain. The masking trick does make the hash 00696 * noticably faster for short strings (like English words). 00697 */ 00698 #ifdef WITH_VALGRIND 00699 if (UNLIKELY(_running_on_valgrind)) { 00700 const rpmuint8_t * k8 = (const rpmuint8_t *)k; 00701 00702 switch (size) { /* all the case statements fall through */ 00703 case 12: c += k[2]; b+=k[1]; a+=k[0]; break; 00704 case 11: c += ((rpmuint32_t)k8[10])<<8; /*@fallthrough@*/ 00705 case 10: c += ((rpmuint32_t)k8[9])<<16; /*@fallthrough@*/ 00706 case 9: c += ((rpmuint32_t)k8[8])<<24; /*@fallthrough@*/ 00707 case 8: b += k[1]; a+=k[0]; break; 00708 case 7: b += ((rpmuint32_t)k8[6])<<8; /*@fallthrough@*/ 00709 case 6: b += ((rpmuint32_t)k8[5])<<16; /*@fallthrough@*/ 00710 case 5: b += ((rpmuint32_t)k8[4])<<24; /*@fallthrough@*/ 00711 case 4: a += k[0]; break; 00712 case 3: a += ((rpmuint32_t)k8[2])<<8; /*@fallthrough@*/ 00713 case 2: a += ((rpmuint32_t)k8[1])<<16; /*@fallthrough@*/ 00714 case 1: a += ((rpmuint32_t)k8[0])<<24; break; 00715 case 0: goto exit; 00716 } 00717 00718 } else 00719 #endif 00720 { 00721 switch (size) { 00722 case 12: c += k[2]; b+=k[1]; a+=k[0]; break; 00723 case 11: c += k[2]&0xffffff00; b+=k[1]; a+=k[0]; break; 00724 case 10: c += k[2]&0xffff0000; b+=k[1]; a+=k[0]; break; 00725 case 9: c += k[2]&0xff000000; b+=k[1]; a+=k[0]; break; 00726 case 8: b += k[1]; a+=k[0]; break; 00727 case 7: b += k[1]&0xffffff00; a+=k[0]; break; 00728 case 6: b += k[1]&0xffff0000; a+=k[0]; break; 00729 case 5: b += k[1]&0xff000000; a+=k[0]; break; 00730 case 4: a += k[0]; break; 00731 case 3: a += k[0]&0xffffff00; break; 00732 case 2: a += k[0]&0xffff0000; break; 00733 case 1: a += k[0]&0xff000000; break; 00734 case 0: goto exit; 00735 } 00736 } 00737 } else { /* need to read the key one byte at a time */ 00738 const rpmuint8_t *k = (const rpmuint8_t *)key; 00739 00740 /*----------- all but the last block: affect some 32 bits of (a,b,c) */ 00741 while (size > 12) { 00742 a += ((rpmuint32_t)k[0])<<24; 00743 a += ((rpmuint32_t)k[1])<<16; 00744 a += ((rpmuint32_t)k[2])<<8; 00745 a += ((rpmuint32_t)k[3]); 00746 b += ((rpmuint32_t)k[4])<<24; 00747 b += ((rpmuint32_t)k[5])<<16; 00748 b += ((rpmuint32_t)k[6])<<8; 00749 b += ((rpmuint32_t)k[7]); 00750 c += ((rpmuint32_t)k[8])<<24; 00751 c += ((rpmuint32_t)k[9])<<16; 00752 c += ((rpmuint32_t)k[10])<<8; 00753 c += ((rpmuint32_t)k[11]); 00754 _JLU3_MIX(a,b,c); 00755 size -= 12; 00756 k += 12; 00757 } 00758 00759 /*---------------------------- last block: affect all 32 bits of (c) */ 00760 switch (size) { /* all the case statements fall through */ 00761 case 12: c += k[11]; /*@fallthrough@*/ 00762 case 11: c += ((rpmuint32_t)k[10])<<8; /*@fallthrough@*/ 00763 case 10: c += ((rpmuint32_t)k[9])<<16; /*@fallthrough@*/ 00764 case 9: c += ((rpmuint32_t)k[8])<<24; /*@fallthrough@*/ 00765 case 8: b += k[7]; /*@fallthrough@*/ 00766 case 7: b += ((rpmuint32_t)k[6])<<8; /*@fallthrough@*/ 00767 case 6: b += ((rpmuint32_t)k[5])<<16; /*@fallthrough@*/ 00768 case 5: b += ((rpmuint32_t)k[4])<<24; /*@fallthrough@*/ 00769 case 4: a += k[3]; /*@fallthrough@*/ 00770 case 3: a += ((rpmuint32_t)k[2])<<8; /*@fallthrough@*/ 00771 case 2: a += ((rpmuint32_t)k[1])<<16; /*@fallthrough@*/ 00772 case 1: a += ((rpmuint32_t)k[0])<<24; /*@fallthrough@*/ 00773 break; 00774 case 0: 00775 goto exit; 00776 } 00777 } 00778 00779 _JLU3_FINAL(a,b,c); 00780 00781 exit: 00782 return c; 00783 } 00784 #endif /* defined(_JLU3_jlu32b) */ 00785 00786 #if defined(_JLU3_SELFTEST) 00787 00788 /* used for timings */ 00789 static void driver1(void) 00790 /*@*/ 00791 { 00792 rpmuint8_t buf[256]; 00793 rpmuint32_t i; 00794 rpmuint32_t h=0; 00795 time_t a,z; 00796 00797 time(&a); 00798 for (i=0; i<256; ++i) buf[i] = 'x'; 00799 for (i=0; i<1; ++i) { 00800 h = jlu32l(h, &buf[0], sizeof(buf[0])); 00801 } 00802 time(&z); 00803 if (z-a > 0) printf("time %d %.8x\n", (int)(z-a), h); 00804 } 00805 00806 /* check that every input bit changes every output bit half the time */ 00807 #define HASHSTATE 1 00808 #define HASHLEN 1 00809 #define MAXPAIR 60 00810 #define MAXLEN 70 00811 static void driver2(void) 00812 /*@*/ 00813 { 00814 rpmuint8_t qa[MAXLEN+1], qb[MAXLEN+2], *a = &qa[0], *b = &qb[1]; 00815 rpmuint32_t c[HASHSTATE], d[HASHSTATE], i=0, j=0, k, l, m=0, z; 00816 rpmuint32_t e[HASHSTATE],f[HASHSTATE],g[HASHSTATE],h[HASHSTATE]; 00817 rpmuint32_t x[HASHSTATE],y[HASHSTATE]; 00818 rpmuint32_t hlen; 00819 00820 printf("No more than %d trials should ever be needed \n",MAXPAIR/2); 00821 for (hlen=0; hlen < MAXLEN; ++hlen) { 00822 z=0; 00823 for (i=0; i<hlen; ++i) { /*-------------- for each input byte, */ 00824 for (j=0; j<8; ++j) { /*--------------- for each input bit, */ 00825 for (m=1; m<8; ++m) { /*--- for serveral possible initvals, */ 00826 for (l=0; l<HASHSTATE; ++l) 00827 e[l]=f[l]=g[l]=h[l]=x[l]=y[l]=~((rpmuint32_t)0); 00828 00829 /* check that every output bit is affected by that input bit */ 00830 for (k=0; k<MAXPAIR; k+=2) { 00831 rpmuint32_t finished=1; 00832 /* keys have one bit different */ 00833 for (l=0; l<hlen+1; ++l) {a[l] = b[l] = (rpmuint8_t)0;} 00834 /* have a and b be two keys differing in only one bit */ 00835 a[i] ^= (k<<j); 00836 a[i] ^= (k>>(8-j)); 00837 c[0] = jlu32l(m, a, hlen); 00838 b[i] ^= ((k+1)<<j); 00839 b[i] ^= ((k+1)>>(8-j)); 00840 d[0] = jlu32l(m, b, hlen); 00841 /* check every bit is 1, 0, set, and not set at least once */ 00842 for (l=0; l<HASHSTATE; ++l) { 00843 e[l] &= (c[l]^d[l]); 00844 f[l] &= ~(c[l]^d[l]); 00845 g[l] &= c[l]; 00846 h[l] &= ~c[l]; 00847 x[l] &= d[l]; 00848 y[l] &= ~d[l]; 00849 if (e[l]|f[l]|g[l]|h[l]|x[l]|y[l]) finished=0; 00850 } 00851 if (finished) break; 00852 } 00853 if (k>z) z=k; 00854 if (k == MAXPAIR) { 00855 printf("Some bit didn't change: "); 00856 printf("%.8x %.8x %.8x %.8x %.8x %.8x ", 00857 e[0],f[0],g[0],h[0],x[0],y[0]); 00858 printf("i %d j %d m %d len %d\n", i, j, m, hlen); 00859 } 00860 if (z == MAXPAIR) goto done; 00861 } 00862 } 00863 } 00864 done: 00865 if (z < MAXPAIR) { 00866 printf("Mix success %2d bytes %2d initvals ",i,m); 00867 printf("required %d trials\n", z/2); 00868 } 00869 } 00870 printf("\n"); 00871 } 00872 00873 /* Check for reading beyond the end of the buffer and alignment problems */ 00874 static void driver3(void) 00875 /*@*/ 00876 { 00877 rpmuint8_t buf[MAXLEN+20], *b; 00878 rpmuint32_t len; 00879 rpmuint8_t q[] = "This is the time for all good men to come to the aid of their country..."; 00880 rpmuint32_t h; 00881 rpmuint8_t qq[] = "xThis is the time for all good men to come to the aid of their country..."; 00882 rpmuint32_t i; 00883 rpmuint8_t qqq[] = "xxThis is the time for all good men to come to the aid of their country..."; 00884 rpmuint32_t j; 00885 rpmuint8_t qqqq[] = "xxxThis is the time for all good men to come to the aid of their country..."; 00886 rpmuint32_t ref,x,y; 00887 rpmuint8_t *p; 00888 rpmuint32_t m = 13; 00889 00890 printf("Endianness. These lines should all be the same (for values filled in):\n"); 00891 printf("%.8x %.8x %.8x\n", 00892 jlu32w(m, (const rpmuint32_t *)q, (sizeof(q)-1)/4), 00893 jlu32w(m, (const rpmuint32_t *)q, (sizeof(q)-5)/4), 00894 jlu32w(m, (const rpmuint32_t *)q, (sizeof(q)-9)/4)); 00895 p = q; 00896 printf("%.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x\n", 00897 jlu32l(m, p, sizeof(q)-1), jlu32l(m, p, sizeof(q)-2), 00898 jlu32l(m, p, sizeof(q)-3), jlu32l(m, p, sizeof(q)-4), 00899 jlu32l(m, p, sizeof(q)-5), jlu32l(m, p, sizeof(q)-6), 00900 jlu32l(m, p, sizeof(q)-7), jlu32l(m, p, sizeof(q)-8), 00901 jlu32l(m, p, sizeof(q)-9), jlu32l(m, p, sizeof(q)-10), 00902 jlu32l(m, p, sizeof(q)-11), jlu32l(m, p, sizeof(q)-12)); 00903 p = &qq[1]; 00904 printf("%.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x\n", 00905 jlu32l(m, p, sizeof(q)-1), jlu32l(m, p, sizeof(q)-2), 00906 jlu32l(m, p, sizeof(q)-3), jlu32l(m, p, sizeof(q)-4), 00907 jlu32l(m, p, sizeof(q)-5), jlu32l(m, p, sizeof(q)-6), 00908 jlu32l(m, p, sizeof(q)-7), jlu32l(m, p, sizeof(q)-8), 00909 jlu32l(m, p, sizeof(q)-9), jlu32l(m, p, sizeof(q)-10), 00910 jlu32l(m, p, sizeof(q)-11), jlu32l(m, p, sizeof(q)-12)); 00911 p = &qqq[2]; 00912 printf("%.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x\n", 00913 jlu32l(m, p, sizeof(q)-1), jlu32l(m, p, sizeof(q)-2), 00914 jlu32l(m, p, sizeof(q)-3), jlu32l(m, p, sizeof(q)-4), 00915 jlu32l(m, p, sizeof(q)-5), jlu32l(m, p, sizeof(q)-6), 00916 jlu32l(m, p, sizeof(q)-7), jlu32l(m, p, sizeof(q)-8), 00917 jlu32l(m, p, sizeof(q)-9), jlu32l(m, p, sizeof(q)-10), 00918 jlu32l(m, p, sizeof(q)-11), jlu32l(m, p, sizeof(q)-12)); 00919 p = &qqqq[3]; 00920 printf("%.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x\n", 00921 jlu32l(m, p, sizeof(q)-1), jlu32l(m, p, sizeof(q)-2), 00922 jlu32l(m, p, sizeof(q)-3), jlu32l(m, p, sizeof(q)-4), 00923 jlu32l(m, p, sizeof(q)-5), jlu32l(m, p, sizeof(q)-6), 00924 jlu32l(m, p, sizeof(q)-7), jlu32l(m, p, sizeof(q)-8), 00925 jlu32l(m, p, sizeof(q)-9), jlu32l(m, p, sizeof(q)-10), 00926 jlu32l(m, p, sizeof(q)-11), jlu32l(m, p, sizeof(q)-12)); 00927 printf("\n"); 00928 for (h=0, b=buf+1; h<8; ++h, ++b) { 00929 for (i=0; i<MAXLEN; ++i) { 00930 len = i; 00931 for (j=0; j<i; ++j) 00932 *(b+j)=0; 00933 00934 /* these should all be equal */ 00935 m = 1; 00936 ref = jlu32l(m, b, len); 00937 *(b+i)=(rpmuint8_t)~0; 00938 *(b-1)=(rpmuint8_t)~0; 00939 x = jlu32l(m, b, len); 00940 y = jlu32l(m, b, len); 00941 if ((ref != x) || (ref != y)) 00942 printf("alignment error: %.8x %.8x %.8x %d %d\n",ref,x,y, h, i); 00943 } 00944 } 00945 } 00946 00947 /* check for problems with nulls */ 00948 static void driver4(void) 00949 /*@*/ 00950 { 00951 rpmuint8_t buf[1]; 00952 rpmuint32_t h; 00953 rpmuint32_t i; 00954 rpmuint32_t state[HASHSTATE]; 00955 00956 buf[0] = ~0; 00957 for (i=0; i<HASHSTATE; ++i) 00958 state[i] = 1; 00959 printf("These should all be different\n"); 00960 h = 0; 00961 for (i=0; i<8; ++i) { 00962 h = jlu32l(h, buf, 0); 00963 printf("%2ld 0-byte strings, hash is %.8x\n", (long)i, h); 00964 } 00965 } 00966 00967 00968 int main(int argc, char ** argv) 00969 { 00970 driver1(); /* test that the key is hashed: used for timings */ 00971 driver2(); /* test that whole key is hashed thoroughly */ 00972 driver3(); /* test that nothing but the key is hashed */ 00973 driver4(); /* test hashing multiple buffers (all buffers are null) */ 00974 return 1; 00975 } 00976 00977 #endif /* _JLU3_SELFTEST */