|
rpm
5.4.4
|
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 #undef UNLIKELY 00042 #ifdef WITH_VALGRIND 00043 #if defined(__GNUC__) && (__GNUC__ > 2) && defined(__OPTIMIZE__) 00044 # define UNLIKELY(value) __builtin_expect((value), 0) && (value > 0 || (value = RUNNING_ON_VALGRIND)) 00045 #else 00046 # define UNLIKELY(value) (value) && (value > 0 || (value = RUNNING_ON_VALGRIND)) 00047 #endif 00048 static int _running_on_valgrind = -1; 00049 #endif 00050 00051 #if defined(_JLU3_SELFTEST) 00052 # define _JLU3_jlu32w 1 00053 # define _JLU3_jlu32l 1 00054 # define _JLU3_jlu32lpair 1 00055 # define _JLU3_jlu32b 1 00056 #endif 00057 00058 /*@-redef@*/ 00059 /*@unchecked@*/ 00060 static const union _dbswap { 00061 const rpmuint32_t ui; 00062 const unsigned char uc[4]; 00063 } endian = { .ui = 0x11223344 }; 00064 # define HASH_LITTLE_ENDIAN (endian.uc[0] == (unsigned char) 0x44) 00065 # define HASH_BIG_ENDIAN (endian.uc[0] == (unsigned char) 0x11) 00066 /*@=redef@*/ 00067 00068 #ifndef ROTL32 00069 # define ROTL32(x, s) (((x) << (s)) | ((x) >> (32 - (s)))) 00070 #endif 00071 00072 /* NOTE: The _size parameter should be in bytes. */ 00073 #define _JLU3_INIT(_h, _size) (0xdeadbeef + ((rpmuint32_t)(_size)) + (_h)) 00074 00075 /* -------------------------------------------------------------------- */ 00076 /* 00077 * _JLU3_MIX -- mix 3 32-bit values reversibly. 00078 * 00079 * This is reversible, so any information in (a,b,c) before _JLU3_MIX() is 00080 * still in (a,b,c) after _JLU3_MIX(). 00081 * 00082 * If four pairs of (a,b,c) inputs are run through _JLU3_MIX(), or through 00083 * _JLU3_MIX() in reverse, there are at least 32 bits of the output that 00084 * are sometimes the same for one pair and different for another pair. 00085 * This was tested for: 00086 * * pairs that differed by one bit, by two bits, in any combination 00087 * of top bits of (a,b,c), or in any combination of bottom bits of 00088 * (a,b,c). 00089 * * "differ" is defined as +, -, ^, or ~^. For + and -, I transformed 00090 * the output delta to a Gray code (a^(a>>1)) so a string of 1's (as 00091 * is commonly produced by subtraction) look like a single 1-bit 00092 * difference. 00093 * * the base values were pseudorandom, all zero but one bit set, or 00094 * all zero plus a counter that starts at zero. 00095 * 00096 * Some k values for my "a-=c; a^=ROTL32(c,k); c+=b;" arrangement that 00097 * satisfy this are 00098 * 4 6 8 16 19 4 00099 * 9 15 3 18 27 15 00100 * 14 9 3 7 17 3 00101 * Well, "9 15 3 18 27 15" didn't quite get 32 bits diffing 00102 * for "differ" defined as + with a one-bit base and a two-bit delta. I 00103 * used http://burtleburtle.net/bob/hash/avalanche.html to choose 00104 * the operations, constants, and arrangements of the variables. 00105 * 00106 * This does not achieve avalanche. There are input bits of (a,b,c) 00107 * that fail to affect some output bits of (a,b,c), especially of a. The 00108 * most thoroughly mixed value is c, but it doesn't really even achieve 00109 * avalanche in c. 00110 * 00111 * This allows some parallelism. Read-after-writes are good at doubling 00112 * the number of bits affected, so the goal of mixing pulls in the opposite 00113 * direction as the goal of parallelism. I did what I could. Rotates 00114 * seem to cost as much as shifts on every machine I could lay my hands 00115 * on, and rotates are much kinder to the top and bottom bits, so I used 00116 * rotates. 00117 */ 00118 /* -------------------------------------------------------------------- */ 00119 #define _JLU3_MIX(a,b,c) \ 00120 { \ 00121 a -= c; a ^= ROTL32(c, 4); c += b; \ 00122 b -= a; b ^= ROTL32(a, 6); a += c; \ 00123 c -= b; c ^= ROTL32(b, 8); b += a; \ 00124 a -= c; a ^= ROTL32(c,16); c += b; \ 00125 b -= a; b ^= ROTL32(a,19); a += c; \ 00126 c -= b; c ^= ROTL32(b, 4); b += a; \ 00127 } 00128 00129 /* -------------------------------------------------------------------- */ 00153 /* -------------------------------------------------------------------- */ 00154 #define _JLU3_FINAL(a,b,c) \ 00155 { \ 00156 c ^= b; c -= ROTL32(b,14); \ 00157 a ^= c; a -= ROTL32(c,11); \ 00158 b ^= a; b -= ROTL32(a,25); \ 00159 c ^= b; c -= ROTL32(b,16); \ 00160 a ^= c; a -= ROTL32(c,4); \ 00161 b ^= a; b -= ROTL32(a,14); \ 00162 c ^= b; c -= ROTL32(b,24); \ 00163 } 00164 00165 #if defined(_JLU3_jlu32w) 00166 rpmuint32_t jlu32w(rpmuint32_t h, /*@null@*/ const rpmuint32_t *k, size_t size) 00167 /*@*/; 00168 /* -------------------------------------------------------------------- */ 00185 /* -------------------------------------------------------------------- */ 00186 rpmuint32_t jlu32w(rpmuint32_t h, const rpmuint32_t *k, size_t size) 00187 { 00188 rpmuint32_t a = _JLU3_INIT(h, (size * sizeof(*k))); 00189 rpmuint32_t b = a; 00190 rpmuint32_t c = a; 00191 00192 if (k == NULL) 00193 goto exit; 00194 00195 /*----------------------------------------------- handle most of the key */ 00196 while (size > 3) { 00197 a += k[0]; 00198 b += k[1]; 00199 c += k[2]; 00200 _JLU3_MIX(a,b,c); 00201 size -= 3; 00202 k += 3; 00203 } 00204 00205 /*----------------------------------------- handle the last 3 rpmuint32_t's */ 00206 switch (size) { 00207 case 3 : c+=k[2]; 00208 case 2 : b+=k[1]; 00209 case 1 : a+=k[0]; 00210 _JLU3_FINAL(a,b,c); 00211 /*@fallthrough@*/ 00212 case 0: 00213 break; 00214 } 00215 /*---------------------------------------------------- report the result */ 00216 exit: 00217 return c; 00218 } 00219 #endif /* defined(_JLU3_jlu32w) */ 00220 00221 #if defined(_JLU3_jlu32l) 00222 rpmuint32_t jlu32l(rpmuint32_t h, const void *key, size_t size) 00223 /*@*/; 00224 /* -------------------------------------------------------------------- */ 00225 /* 00226 * jlu32l() -- hash a variable-length key into a 32-bit value 00227 * h : can be any 4-byte value 00228 * k : the key (the unaligned variable-length array of bytes) 00229 * size : the size of the key, counting by bytes 00230 * Returns a 32-bit value. Every bit of the key affects every bit of 00231 * the return value. Two keys differing by one or two bits will have 00232 * totally different hash values. 00233 * 00234 * The best hash table sizes are powers of 2. There is no need to do 00235 * mod a prime (mod is sooo slow!). If you need less than 32 bits, 00236 * use a bitmask. For example, if you need only 10 bits, do 00237 * h = (h & hashmask(10)); 00238 * In which case, the hash table should have hashsize(10) elements. 00239 * 00240 * If you are hashing n strings (rpmuint8_t **)k, do it like this: 00241 * for (i=0, h=0; i<n; ++i) h = jlu32l(h, k[i], len[i]); 00242 * 00243 * By Bob Jenkins, 2006. bob_jenkins@burtleburtle.net. You may use this 00244 * code any way you wish, private, educational, or commercial. It's free. 00245 * 00246 * Use for hash table lookup, or anything where one collision in 2^^32 is 00247 * acceptable. Do NOT use for cryptographic purposes. 00248 * 00249 * @param h the previous hash, or an arbitrary value 00250 * @param *k the key, an array of rpmuint8_t values 00251 * @param size the size of the key 00252 * @return the lookup3 hash 00253 */ 00254 /* -------------------------------------------------------------------- */ 00255 rpmuint32_t jlu32l(rpmuint32_t h, const void *key, size_t size) 00256 { 00257 union { const void *ptr; size_t i; } u; 00258 rpmuint32_t a = _JLU3_INIT(h, size); 00259 rpmuint32_t b = a; 00260 rpmuint32_t c = a; 00261 00262 if (key == NULL) 00263 goto exit; 00264 00265 u.ptr = key; 00266 if (HASH_LITTLE_ENDIAN && ((u.i & 0x3) == 0)) { 00267 const rpmuint32_t *k = (const rpmuint32_t *)key; /* read 32-bit chunks */ 00268 00269 /*------ all but last block: aligned reads and affect 32 bits of (a,b,c) */ 00270 while (size > 12) { 00271 a += k[0]; 00272 b += k[1]; 00273 c += k[2]; 00274 _JLU3_MIX(a,b,c); 00275 size -= 12; 00276 k += 3; 00277 } 00278 00279 /*------------------------- handle the last (probably partial) block */ 00280 /* 00281 * "k[2]&0xffffff" actually reads beyond the end of the string, but 00282 * then masks off the part it's not allowed to read. Because the 00283 * string is aligned, the masked-off tail is in the same word as the 00284 * rest of the string. Every machine with memory protection I've seen 00285 * does it on word boundaries, so is OK with this. But VALGRIND will 00286 * still catch it and complain. The masking trick does make the hash 00287 * noticably faster for short strings (like English words). 00288 */ 00289 #ifdef WITH_VALGRIND 00290 if (UNLIKELY(_running_on_valgrind)) { 00291 const rpmuint8_t * k8 = (const rpmuint8_t *)k; 00292 00293 switch (size) { 00294 case 12: c += k[2]; b+=k[1]; a+=k[0]; break; 00295 case 11: c += ((rpmuint32_t)k8[10])<<16; /*@fallthrough@*/ 00296 case 10: c += ((rpmuint32_t)k8[9])<<8; /*@fallthrough@*/ 00297 case 9: c += k8[8]; /*@fallthrough@*/ 00298 case 8: b += k[1]; a+=k[0]; break; 00299 case 7: b += ((rpmuint32_t)k8[6])<<16; /*@fallthrough@*/ 00300 case 6: b += ((rpmuint32_t)k8[5])<<8; /*@fallthrough@*/ 00301 case 5: b += k8[4]; /*@fallthrough@*/ 00302 case 4: a += k[0]; break; 00303 case 3: a += ((rpmuint32_t)k8[2])<<16; /*@fallthrough@*/ 00304 case 2: a += ((rpmuint32_t)k8[1])<<8; /*@fallthrough@*/ 00305 case 1: a += k8[0]; break; 00306 case 0: goto exit; 00307 } 00308 00309 } else 00310 #endif 00311 { 00312 switch (size) { 00313 case 12: c += k[2]; b+=k[1]; a+=k[0]; break; 00314 case 11: c += k[2]&0xffffff; b+=k[1]; a+=k[0]; break; 00315 case 10: c += k[2]&0xffff; b+=k[1]; a+=k[0]; break; 00316 case 9: c += k[2]&0xff; b+=k[1]; a+=k[0]; break; 00317 case 8: b += k[1]; a+=k[0]; break; 00318 case 7: b += k[1]&0xffffff; a+=k[0]; break; 00319 case 6: b += k[1]&0xffff; a+=k[0]; break; 00320 case 5: b += k[1]&0xff; a+=k[0]; break; 00321 case 4: a += k[0]; break; 00322 case 3: a += k[0]&0xffffff; break; 00323 case 2: a += k[0]&0xffff; break; 00324 case 1: a += k[0]&0xff; break; 00325 case 0: goto exit; 00326 } 00327 } 00328 } else if (HASH_LITTLE_ENDIAN && ((u.i & 0x1) == 0)) { 00329 const rpmuint16_t *k = (const rpmuint16_t *)key; /* read 16-bit chunks */ 00330 const rpmuint8_t *k8; 00331 00332 /*----------- all but last block: aligned reads and different mixing */ 00333 while (size > 12) { 00334 a += k[0] + (((rpmuint32_t)k[1])<<16); 00335 b += k[2] + (((rpmuint32_t)k[3])<<16); 00336 c += k[4] + (((rpmuint32_t)k[5])<<16); 00337 _JLU3_MIX(a,b,c); 00338 size -= 12; 00339 k += 6; 00340 } 00341 00342 /*------------------------- handle the last (probably partial) block */ 00343 k8 = (const rpmuint8_t *)k; 00344 switch (size) { 00345 case 12: 00346 c += k[4]+(((rpmuint32_t)k[5])<<16); 00347 b += k[2]+(((rpmuint32_t)k[3])<<16); 00348 a += k[0]+(((rpmuint32_t)k[1])<<16); 00349 break; 00350 case 11: 00351 c += ((rpmuint32_t)k8[10])<<16; 00352 /*@fallthrough@*/ 00353 case 10: 00354 c += (rpmuint32_t)k[4]; 00355 b += k[2]+(((rpmuint32_t)k[3])<<16); 00356 a += k[0]+(((rpmuint32_t)k[1])<<16); 00357 break; 00358 case 9: 00359 c += (rpmuint32_t)k8[8]; 00360 /*@fallthrough@*/ 00361 case 8: 00362 b += k[2]+(((rpmuint32_t)k[3])<<16); 00363 a += k[0]+(((rpmuint32_t)k[1])<<16); 00364 break; 00365 case 7: 00366 b += ((rpmuint32_t)k8[6])<<16; 00367 /*@fallthrough@*/ 00368 case 6: 00369 b += (rpmuint32_t)k[2]; 00370 a += k[0]+(((rpmuint32_t)k[1])<<16); 00371 break; 00372 case 5: 00373 b += (rpmuint32_t)k8[4]; 00374 /*@fallthrough@*/ 00375 case 4: 00376 a += k[0]+(((rpmuint32_t)k[1])<<16); 00377 break; 00378 case 3: 00379 a += ((rpmuint32_t)k8[2])<<16; 00380 /*@fallthrough@*/ 00381 case 2: 00382 a += (rpmuint32_t)k[0]; 00383 break; 00384 case 1: 00385 a += (rpmuint32_t)k8[0]; 00386 break; 00387 case 0: 00388 goto exit; 00389 } 00390 00391 } else { /* need to read the key one byte at a time */ 00392 const rpmuint8_t *k = (const rpmuint8_t *)key; 00393 00394 /*----------- all but the last block: affect some 32 bits of (a,b,c) */ 00395 while (size > 12) { 00396 a += (rpmuint32_t)k[0]; 00397 a += ((rpmuint32_t)k[1])<<8; 00398 a += ((rpmuint32_t)k[2])<<16; 00399 a += ((rpmuint32_t)k[3])<<24; 00400 b += (rpmuint32_t)k[4]; 00401 b += ((rpmuint32_t)k[5])<<8; 00402 b += ((rpmuint32_t)k[6])<<16; 00403 b += ((rpmuint32_t)k[7])<<24; 00404 c += (rpmuint32_t)k[8]; 00405 c += ((rpmuint32_t)k[9])<<8; 00406 c += ((rpmuint32_t)k[10])<<16; 00407 c += ((rpmuint32_t)k[11])<<24; 00408 _JLU3_MIX(a,b,c); 00409 size -= 12; 00410 k += 12; 00411 } 00412 00413 /*---------------------------- last block: affect all 32 bits of (c) */ 00414 switch (size) { 00415 case 12: c += ((rpmuint32_t)k[11])<<24; /*@fallthrough@*/ 00416 case 11: c += ((rpmuint32_t)k[10])<<16; /*@fallthrough@*/ 00417 case 10: c += ((rpmuint32_t)k[9])<<8; /*@fallthrough@*/ 00418 case 9: c += (rpmuint32_t)k[8]; /*@fallthrough@*/ 00419 case 8: b += ((rpmuint32_t)k[7])<<24; /*@fallthrough@*/ 00420 case 7: b += ((rpmuint32_t)k[6])<<16; /*@fallthrough@*/ 00421 case 6: b += ((rpmuint32_t)k[5])<<8; /*@fallthrough@*/ 00422 case 5: b += (rpmuint32_t)k[4]; /*@fallthrough@*/ 00423 case 4: a += ((rpmuint32_t)k[3])<<24; /*@fallthrough@*/ 00424 case 3: a += ((rpmuint32_t)k[2])<<16; /*@fallthrough@*/ 00425 case 2: a += ((rpmuint32_t)k[1])<<8; /*@fallthrough@*/ 00426 case 1: a += (rpmuint32_t)k[0]; 00427 break; 00428 case 0: 00429 goto exit; 00430 } 00431 } 00432 00433 _JLU3_FINAL(a,b,c); 00434 00435 exit: 00436 return c; 00437 } 00438 #endif /* defined(_JLU3_jlu32l) */ 00439 00440 #if defined(_JLU3_jlu32lpair) 00441 void jlu32lpair(/*@null@*/ const void *key, size_t size, 00442 rpmuint32_t *pc, rpmuint32_t *pb) 00443 /*@modifies *pc, *pb@*/; 00460 void jlu32lpair(const void *key, size_t size, rpmuint32_t *pc, rpmuint32_t *pb) 00461 { 00462 union { const void *ptr; size_t i; } u; 00463 rpmuint32_t a = _JLU3_INIT(*pc, size); 00464 rpmuint32_t b = a; 00465 rpmuint32_t c = a; 00466 00467 if (key == NULL) 00468 goto exit; 00469 00470 c += *pb; /* Add the secondary hash. */ 00471 00472 u.ptr = key; 00473 if (HASH_LITTLE_ENDIAN && ((u.i & 0x3) == 0)) { 00474 const rpmuint32_t *k = (const rpmuint32_t *)key; /* read 32-bit chunks */ 00475 00476 /*-- all but last block: aligned reads and affect 32 bits of (a,b,c) */ 00477 while (size > 12) { 00478 a += k[0]; 00479 b += k[1]; 00480 c += k[2]; 00481 _JLU3_MIX(a,b,c); 00482 size -= 12; 00483 k += 3; 00484 } 00485 /*------------------------- handle the last (probably partial) block */ 00486 /* 00487 * "k[2]&0xffffff" actually reads beyond the end of the string, but 00488 * then masks off the part it's not allowed to read. Because the 00489 * string is aligned, the masked-off tail is in the same word as the 00490 * rest of the string. Every machine with memory protection I've seen 00491 * does it on word boundaries, so is OK with this. But VALGRIND will 00492 * still catch it and complain. The masking trick does make the hash 00493 * noticably faster for short strings (like English words). 00494 */ 00495 #ifdef WITH_VALGRIND 00496 if (UNLIKELY(_running_on_valgrind)) { 00497 const rpmuint8_t * k8 = (const rpmuint8_t *)k; 00498 00499 switch (size) { 00500 case 12: c += k[2]; b+=k[1]; a+=k[0]; break; 00501 case 11: c += ((rpmuint32_t)k8[10])<<16; /*@fallthrough@*/ 00502 case 10: c += ((rpmuint32_t)k8[9])<<8; /*@fallthrough@*/ 00503 case 9: c += k8[8]; /*@fallthrough@*/ 00504 case 8: b += k[1]; a+=k[0]; break; 00505 case 7: b += ((rpmuint32_t)k8[6])<<16; /*@fallthrough@*/ 00506 case 6: b += ((rpmuint32_t)k8[5])<<8; /*@fallthrough@*/ 00507 case 5: b += k8[4]; /*@fallthrough@*/ 00508 case 4: a += k[0]; break; 00509 case 3: a += ((rpmuint32_t)k8[2])<<16; /*@fallthrough@*/ 00510 case 2: a += ((rpmuint32_t)k8[1])<<8; /*@fallthrough@*/ 00511 case 1: a += k8[0]; break; 00512 case 0: goto exit; 00513 } 00514 00515 } else 00516 #endif 00517 { 00518 switch (size) { 00519 case 12: c += k[2]; b+=k[1]; a+=k[0]; break; 00520 case 11: c += k[2]&0xffffff; b+=k[1]; a+=k[0]; break; 00521 case 10: c += k[2]&0xffff; b+=k[1]; a+=k[0]; break; 00522 case 9: c += k[2]&0xff; b+=k[1]; a+=k[0]; break; 00523 case 8: b += k[1]; a+=k[0]; break; 00524 case 7: b += k[1]&0xffffff; a+=k[0]; break; 00525 case 6: b += k[1]&0xffff; a+=k[0]; break; 00526 case 5: b += k[1]&0xff; a+=k[0]; break; 00527 case 4: a += k[0]; break; 00528 case 3: a += k[0]&0xffffff; break; 00529 case 2: a += k[0]&0xffff; break; 00530 case 1: a += k[0]&0xff; break; 00531 case 0: goto exit; 00532 } 00533 } 00534 } else if (HASH_LITTLE_ENDIAN && ((u.i & 0x1) == 0)) { 00535 const rpmuint16_t *k = (const rpmuint16_t *)key; /* read 16-bit chunks */ 00536 const rpmuint8_t *k8; 00537 00538 /*----------- all but last block: aligned reads and different mixing */ 00539 while (size > 12) { 00540 a += k[0] + (((rpmuint32_t)k[1])<<16); 00541 b += k[2] + (((rpmuint32_t)k[3])<<16); 00542 c += k[4] + (((rpmuint32_t)k[5])<<16); 00543 _JLU3_MIX(a,b,c); 00544 size -= 12; 00545 k += 6; 00546 } 00547 00548 /*------------------------- handle the last (probably partial) block */ 00549 k8 = (const rpmuint8_t *)k; 00550 switch (size) { 00551 case 12: 00552 c += k[4]+(((rpmuint32_t)k[5])<<16); 00553 b += k[2]+(((rpmuint32_t)k[3])<<16); 00554 a += k[0]+(((rpmuint32_t)k[1])<<16); 00555 break; 00556 case 11: 00557 c += ((rpmuint32_t)k8[10])<<16; 00558 /*@fallthrough@*/ 00559 case 10: 00560 c += k[4]; 00561 b += k[2]+(((rpmuint32_t)k[3])<<16); 00562 a += k[0]+(((rpmuint32_t)k[1])<<16); 00563 break; 00564 case 9: 00565 c += k8[8]; 00566 /*@fallthrough@*/ 00567 case 8: 00568 b += k[2]+(((rpmuint32_t)k[3])<<16); 00569 a += k[0]+(((rpmuint32_t)k[1])<<16); 00570 break; 00571 case 7: 00572 b += ((rpmuint32_t)k8[6])<<16; 00573 /*@fallthrough@*/ 00574 case 6: 00575 b += k[2]; 00576 a += k[0]+(((rpmuint32_t)k[1])<<16); 00577 break; 00578 case 5: 00579 b += k8[4]; 00580 /*@fallthrough@*/ 00581 case 4: 00582 a += k[0]+(((rpmuint32_t)k[1])<<16); 00583 break; 00584 case 3: 00585 a += ((rpmuint32_t)k8[2])<<16; 00586 /*@fallthrough@*/ 00587 case 2: 00588 a += k[0]; 00589 break; 00590 case 1: 00591 a += k8[0]; 00592 break; 00593 case 0: 00594 goto exit; 00595 } 00596 00597 } else { /* need to read the key one byte at a time */ 00598 const rpmuint8_t *k = (const rpmuint8_t *)key; 00599 00600 /*----------- all but the last block: affect some 32 bits of (a,b,c) */ 00601 while (size > 12) { 00602 a += k[0]; 00603 a += ((rpmuint32_t)k[1])<<8; 00604 a += ((rpmuint32_t)k[2])<<16; 00605 a += ((rpmuint32_t)k[3])<<24; 00606 b += k[4]; 00607 b += ((rpmuint32_t)k[5])<<8; 00608 b += ((rpmuint32_t)k[6])<<16; 00609 b += ((rpmuint32_t)k[7])<<24; 00610 c += k[8]; 00611 c += ((rpmuint32_t)k[9])<<8; 00612 c += ((rpmuint32_t)k[10])<<16; 00613 c += ((rpmuint32_t)k[11])<<24; 00614 _JLU3_MIX(a,b,c); 00615 size -= 12; 00616 k += 12; 00617 } 00618 00619 /*---------------------------- last block: affect all 32 bits of (c) */ 00620 switch (size) { 00621 case 12: c += ((rpmuint32_t)k[11])<<24; /*@fallthrough@*/ 00622 case 11: c += ((rpmuint32_t)k[10])<<16; /*@fallthrough@*/ 00623 case 10: c += ((rpmuint32_t)k[9])<<8; /*@fallthrough@*/ 00624 case 9: c += k[8]; /*@fallthrough@*/ 00625 case 8: b += ((rpmuint32_t)k[7])<<24; /*@fallthrough@*/ 00626 case 7: b += ((rpmuint32_t)k[6])<<16; /*@fallthrough@*/ 00627 case 6: b += ((rpmuint32_t)k[5])<<8; /*@fallthrough@*/ 00628 case 5: b += k[4]; /*@fallthrough@*/ 00629 case 4: a += ((rpmuint32_t)k[3])<<24; /*@fallthrough@*/ 00630 case 3: a += ((rpmuint32_t)k[2])<<16; /*@fallthrough@*/ 00631 case 2: a += ((rpmuint32_t)k[1])<<8; /*@fallthrough@*/ 00632 case 1: a += k[0]; /*@fallthrough@*/ 00633 break; 00634 case 0: 00635 goto exit; 00636 } 00637 } 00638 00639 _JLU3_FINAL(a,b,c); 00640 00641 exit: 00642 *pc = c; 00643 *pb = b; 00644 return; 00645 } 00646 #endif /* defined(_JLU3_jlu32lpair) */ 00647 00648 #if defined(_JLU3_jlu32b) 00649 rpmuint32_t jlu32b(rpmuint32_t h, /*@null@*/ const void *key, size_t size) 00650 /*@*/; 00651 /* 00652 * jlu32b(): 00653 * This is the same as jlu32w() on big-endian machines. It is different 00654 * from jlu32l() on all machines. jlu32b() takes advantage of 00655 * big-endian byte ordering. 00656 * 00657 * @param h the previous hash, or an arbitrary value 00658 * @param *k the key, an array of rpmuint8_t values 00659 * @param size the size of the key 00660 * @return the lookup3 hash 00661 */ 00662 rpmuint32_t jlu32b(rpmuint32_t h, const void *key, size_t size) 00663 { 00664 union { const void *ptr; size_t i; } u; 00665 rpmuint32_t a = _JLU3_INIT(h, size); 00666 rpmuint32_t b = a; 00667 rpmuint32_t c = a; 00668 00669 if (key == NULL) 00670 return h; 00671 00672 u.ptr = key; 00673 if (HASH_BIG_ENDIAN && ((u.i & 0x3) == 0)) { 00674 const rpmuint32_t *k = (const rpmuint32_t *)key; /* read 32-bit chunks */ 00675 00676 /*-- all but last block: aligned reads and affect 32 bits of (a,b,c) */ 00677 while (size > 12) { 00678 a += k[0]; 00679 b += k[1]; 00680 c += k[2]; 00681 _JLU3_MIX(a,b,c); 00682 size -= 12; 00683 k += 3; 00684 } 00685 00686 /*------------------------- handle the last (probably partial) block */ 00687 /* 00688 * "k[2]<<8" actually reads beyond the end of the string, but 00689 * then shifts out the part it's not allowed to read. Because the 00690 * string is aligned, the illegal read is in the same word as the 00691 * rest of the string. Every machine with memory protection I've seen 00692 * does it on word boundaries, so is OK with this. But VALGRIND will 00693 * still catch it and complain. The masking trick does make the hash 00694 * noticably faster for short strings (like English words). 00695 */ 00696 #ifdef WITH_VALGRIND 00697 if (UNLIKELY(_running_on_valgrind)) { 00698 const rpmuint8_t * k8 = (const rpmuint8_t *)k; 00699 00700 switch (size) { /* all the case statements fall through */ 00701 case 12: c += k[2]; b+=k[1]; a+=k[0]; break; 00702 case 11: c += ((rpmuint32_t)k8[10])<<8; /*@fallthrough@*/ 00703 case 10: c += ((rpmuint32_t)k8[9])<<16; /*@fallthrough@*/ 00704 case 9: c += ((rpmuint32_t)k8[8])<<24; /*@fallthrough@*/ 00705 case 8: b += k[1]; a+=k[0]; break; 00706 case 7: b += ((rpmuint32_t)k8[6])<<8; /*@fallthrough@*/ 00707 case 6: b += ((rpmuint32_t)k8[5])<<16; /*@fallthrough@*/ 00708 case 5: b += ((rpmuint32_t)k8[4])<<24; /*@fallthrough@*/ 00709 case 4: a += k[0]; break; 00710 case 3: a += ((rpmuint32_t)k8[2])<<8; /*@fallthrough@*/ 00711 case 2: a += ((rpmuint32_t)k8[1])<<16; /*@fallthrough@*/ 00712 case 1: a += ((rpmuint32_t)k8[0])<<24; break; 00713 case 0: goto exit; 00714 } 00715 00716 } else 00717 #endif 00718 { 00719 switch (size) { 00720 case 12: c += k[2]; b+=k[1]; a+=k[0]; break; 00721 case 11: c += k[2]&0xffffff00; b+=k[1]; a+=k[0]; break; 00722 case 10: c += k[2]&0xffff0000; b+=k[1]; a+=k[0]; break; 00723 case 9: c += k[2]&0xff000000; b+=k[1]; a+=k[0]; break; 00724 case 8: b += k[1]; a+=k[0]; break; 00725 case 7: b += k[1]&0xffffff00; a+=k[0]; break; 00726 case 6: b += k[1]&0xffff0000; a+=k[0]; break; 00727 case 5: b += k[1]&0xff000000; a+=k[0]; break; 00728 case 4: a += k[0]; break; 00729 case 3: a += k[0]&0xffffff00; break; 00730 case 2: a += k[0]&0xffff0000; break; 00731 case 1: a += k[0]&0xff000000; break; 00732 case 0: goto exit; 00733 } 00734 } 00735 } else { /* need to read the key one byte at a time */ 00736 const rpmuint8_t *k = (const rpmuint8_t *)key; 00737 00738 /*----------- all but the last block: affect some 32 bits of (a,b,c) */ 00739 while (size > 12) { 00740 a += ((rpmuint32_t)k[0])<<24; 00741 a += ((rpmuint32_t)k[1])<<16; 00742 a += ((rpmuint32_t)k[2])<<8; 00743 a += ((rpmuint32_t)k[3]); 00744 b += ((rpmuint32_t)k[4])<<24; 00745 b += ((rpmuint32_t)k[5])<<16; 00746 b += ((rpmuint32_t)k[6])<<8; 00747 b += ((rpmuint32_t)k[7]); 00748 c += ((rpmuint32_t)k[8])<<24; 00749 c += ((rpmuint32_t)k[9])<<16; 00750 c += ((rpmuint32_t)k[10])<<8; 00751 c += ((rpmuint32_t)k[11]); 00752 _JLU3_MIX(a,b,c); 00753 size -= 12; 00754 k += 12; 00755 } 00756 00757 /*---------------------------- last block: affect all 32 bits of (c) */ 00758 switch (size) { /* all the case statements fall through */ 00759 case 12: c += k[11]; /*@fallthrough@*/ 00760 case 11: c += ((rpmuint32_t)k[10])<<8; /*@fallthrough@*/ 00761 case 10: c += ((rpmuint32_t)k[9])<<16; /*@fallthrough@*/ 00762 case 9: c += ((rpmuint32_t)k[8])<<24; /*@fallthrough@*/ 00763 case 8: b += k[7]; /*@fallthrough@*/ 00764 case 7: b += ((rpmuint32_t)k[6])<<8; /*@fallthrough@*/ 00765 case 6: b += ((rpmuint32_t)k[5])<<16; /*@fallthrough@*/ 00766 case 5: b += ((rpmuint32_t)k[4])<<24; /*@fallthrough@*/ 00767 case 4: a += k[3]; /*@fallthrough@*/ 00768 case 3: a += ((rpmuint32_t)k[2])<<8; /*@fallthrough@*/ 00769 case 2: a += ((rpmuint32_t)k[1])<<16; /*@fallthrough@*/ 00770 case 1: a += ((rpmuint32_t)k[0])<<24; /*@fallthrough@*/ 00771 break; 00772 case 0: 00773 goto exit; 00774 } 00775 } 00776 00777 _JLU3_FINAL(a,b,c); 00778 00779 exit: 00780 return c; 00781 } 00782 #endif /* defined(_JLU3_jlu32b) */ 00783 00784 #if defined(_JLU3_SELFTEST) 00785 00786 /* used for timings */ 00787 static void driver1(void) 00788 /*@*/ 00789 { 00790 rpmuint8_t buf[256]; 00791 rpmuint32_t i; 00792 rpmuint32_t h=0; 00793 time_t a,z; 00794 00795 time(&a); 00796 for (i=0; i<256; ++i) buf[i] = 'x'; 00797 for (i=0; i<1; ++i) { 00798 h = jlu32l(h, &buf[0], sizeof(buf[0])); 00799 } 00800 time(&z); 00801 if (z-a > 0) printf("time %d %.8x\n", (int)(z-a), h); 00802 } 00803 00804 /* check that every input bit changes every output bit half the time */ 00805 #define HASHSTATE 1 00806 #define HASHLEN 1 00807 #define MAXPAIR 60 00808 #define MAXLEN 70 00809 static void driver2(void) 00810 /*@*/ 00811 { 00812 rpmuint8_t qa[MAXLEN+1], qb[MAXLEN+2], *a = &qa[0], *b = &qb[1]; 00813 rpmuint32_t c[HASHSTATE], d[HASHSTATE], i=0, j=0, k, l, m=0, z; 00814 rpmuint32_t e[HASHSTATE],f[HASHSTATE],g[HASHSTATE],h[HASHSTATE]; 00815 rpmuint32_t x[HASHSTATE],y[HASHSTATE]; 00816 rpmuint32_t hlen; 00817 00818 printf("No more than %d trials should ever be needed \n",MAXPAIR/2); 00819 for (hlen=0; hlen < MAXLEN; ++hlen) { 00820 z=0; 00821 for (i=0; i<hlen; ++i) { /*-------------- for each input byte, */ 00822 for (j=0; j<8; ++j) { /*--------------- for each input bit, */ 00823 for (m=1; m<8; ++m) { /*--- for serveral possible initvals, */ 00824 for (l=0; l<HASHSTATE; ++l) 00825 e[l]=f[l]=g[l]=h[l]=x[l]=y[l]=~((rpmuint32_t)0); 00826 00827 /* check that every output bit is affected by that input bit */ 00828 for (k=0; k<MAXPAIR; k+=2) { 00829 rpmuint32_t finished=1; 00830 /* keys have one bit different */ 00831 for (l=0; l<hlen+1; ++l) {a[l] = b[l] = (rpmuint8_t)0;} 00832 /* have a and b be two keys differing in only one bit */ 00833 a[i] ^= (k<<j); 00834 a[i] ^= (k>>(8-j)); 00835 c[0] = jlu32l(m, a, hlen); 00836 b[i] ^= ((k+1)<<j); 00837 b[i] ^= ((k+1)>>(8-j)); 00838 d[0] = jlu32l(m, b, hlen); 00839 /* check every bit is 1, 0, set, and not set at least once */ 00840 for (l=0; l<HASHSTATE; ++l) { 00841 e[l] &= (c[l]^d[l]); 00842 f[l] &= ~(c[l]^d[l]); 00843 g[l] &= c[l]; 00844 h[l] &= ~c[l]; 00845 x[l] &= d[l]; 00846 y[l] &= ~d[l]; 00847 if (e[l]|f[l]|g[l]|h[l]|x[l]|y[l]) finished=0; 00848 } 00849 if (finished) break; 00850 } 00851 if (k>z) z=k; 00852 if (k == MAXPAIR) { 00853 printf("Some bit didn't change: "); 00854 printf("%.8x %.8x %.8x %.8x %.8x %.8x ", 00855 e[0],f[0],g[0],h[0],x[0],y[0]); 00856 printf("i %d j %d m %d len %d\n", i, j, m, hlen); 00857 } 00858 if (z == MAXPAIR) goto done; 00859 } 00860 } 00861 } 00862 done: 00863 if (z < MAXPAIR) { 00864 printf("Mix success %2d bytes %2d initvals ",i,m); 00865 printf("required %d trials\n", z/2); 00866 } 00867 } 00868 printf("\n"); 00869 } 00870 00871 /* Check for reading beyond the end of the buffer and alignment problems */ 00872 static void driver3(void) 00873 /*@*/ 00874 { 00875 rpmuint8_t buf[MAXLEN+20], *b; 00876 rpmuint32_t len; 00877 rpmuint8_t q[] = "This is the time for all good men to come to the aid of their country..."; 00878 rpmuint32_t h; 00879 rpmuint8_t qq[] = "xThis is the time for all good men to come to the aid of their country..."; 00880 rpmuint32_t i; 00881 rpmuint8_t qqq[] = "xxThis is the time for all good men to come to the aid of their country..."; 00882 rpmuint32_t j; 00883 rpmuint8_t qqqq[] = "xxxThis is the time for all good men to come to the aid of their country..."; 00884 rpmuint32_t ref,x,y; 00885 rpmuint8_t *p; 00886 rpmuint32_t m = 13; 00887 00888 printf("Endianness. These lines should all be the same (for values filled in):\n"); 00889 printf("%.8x %.8x %.8x\n", 00890 jlu32w(m, (const rpmuint32_t *)q, (sizeof(q)-1)/4), 00891 jlu32w(m, (const rpmuint32_t *)q, (sizeof(q)-5)/4), 00892 jlu32w(m, (const rpmuint32_t *)q, (sizeof(q)-9)/4)); 00893 p = q; 00894 printf("%.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x\n", 00895 jlu32l(m, p, sizeof(q)-1), jlu32l(m, p, sizeof(q)-2), 00896 jlu32l(m, p, sizeof(q)-3), jlu32l(m, p, sizeof(q)-4), 00897 jlu32l(m, p, sizeof(q)-5), jlu32l(m, p, sizeof(q)-6), 00898 jlu32l(m, p, sizeof(q)-7), jlu32l(m, p, sizeof(q)-8), 00899 jlu32l(m, p, sizeof(q)-9), jlu32l(m, p, sizeof(q)-10), 00900 jlu32l(m, p, sizeof(q)-11), jlu32l(m, p, sizeof(q)-12)); 00901 p = &qq[1]; 00902 printf("%.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x\n", 00903 jlu32l(m, p, sizeof(q)-1), jlu32l(m, p, sizeof(q)-2), 00904 jlu32l(m, p, sizeof(q)-3), jlu32l(m, p, sizeof(q)-4), 00905 jlu32l(m, p, sizeof(q)-5), jlu32l(m, p, sizeof(q)-6), 00906 jlu32l(m, p, sizeof(q)-7), jlu32l(m, p, sizeof(q)-8), 00907 jlu32l(m, p, sizeof(q)-9), jlu32l(m, p, sizeof(q)-10), 00908 jlu32l(m, p, sizeof(q)-11), jlu32l(m, p, sizeof(q)-12)); 00909 p = &qqq[2]; 00910 printf("%.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x\n", 00911 jlu32l(m, p, sizeof(q)-1), jlu32l(m, p, sizeof(q)-2), 00912 jlu32l(m, p, sizeof(q)-3), jlu32l(m, p, sizeof(q)-4), 00913 jlu32l(m, p, sizeof(q)-5), jlu32l(m, p, sizeof(q)-6), 00914 jlu32l(m, p, sizeof(q)-7), jlu32l(m, p, sizeof(q)-8), 00915 jlu32l(m, p, sizeof(q)-9), jlu32l(m, p, sizeof(q)-10), 00916 jlu32l(m, p, sizeof(q)-11), jlu32l(m, p, sizeof(q)-12)); 00917 p = &qqqq[3]; 00918 printf("%.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x\n", 00919 jlu32l(m, p, sizeof(q)-1), jlu32l(m, p, sizeof(q)-2), 00920 jlu32l(m, p, sizeof(q)-3), jlu32l(m, p, sizeof(q)-4), 00921 jlu32l(m, p, sizeof(q)-5), jlu32l(m, p, sizeof(q)-6), 00922 jlu32l(m, p, sizeof(q)-7), jlu32l(m, p, sizeof(q)-8), 00923 jlu32l(m, p, sizeof(q)-9), jlu32l(m, p, sizeof(q)-10), 00924 jlu32l(m, p, sizeof(q)-11), jlu32l(m, p, sizeof(q)-12)); 00925 printf("\n"); 00926 for (h=0, b=buf+1; h<8; ++h, ++b) { 00927 for (i=0; i<MAXLEN; ++i) { 00928 len = i; 00929 for (j=0; j<i; ++j) 00930 *(b+j)=0; 00931 00932 /* these should all be equal */ 00933 m = 1; 00934 ref = jlu32l(m, b, len); 00935 *(b+i)=(rpmuint8_t)~0; 00936 *(b-1)=(rpmuint8_t)~0; 00937 x = jlu32l(m, b, len); 00938 y = jlu32l(m, b, len); 00939 if ((ref != x) || (ref != y)) 00940 printf("alignment error: %.8x %.8x %.8x %d %d\n",ref,x,y, h, i); 00941 } 00942 } 00943 } 00944 00945 /* check for problems with nulls */ 00946 static void driver4(void) 00947 /*@*/ 00948 { 00949 rpmuint8_t buf[1]; 00950 rpmuint32_t h; 00951 rpmuint32_t i; 00952 rpmuint32_t state[HASHSTATE]; 00953 00954 buf[0] = ~0; 00955 for (i=0; i<HASHSTATE; ++i) 00956 state[i] = 1; 00957 printf("These should all be different\n"); 00958 h = 0; 00959 for (i=0; i<8; ++i) { 00960 h = jlu32l(h, buf, 0); 00961 printf("%2ld 0-byte strings, hash is %.8x\n", (long)i, h); 00962 } 00963 } 00964 00965 00966 int main(int argc, char ** argv) 00967 { 00968 driver1(); /* test that the key is hashed: used for timings */ 00969 driver2(); /* test that whole key is hashed thoroughly */ 00970 driver3(); /* test that nothing but the key is hashed */ 00971 driver4(); /* test hashing multiple buffers (all buffers are null) */ 00972 return 1; 00973 } 00974 00975 #endif /* _JLU3_SELFTEST */
1.7.5.1