| 1 |
#include "jenkins_hash.h" |
| 2 |
#include <stdlib.h> |
| 3 |
#ifdef WIN32 |
| 4 |
#define _USE_MATH_DEFINES //For M_LOG2E |
| 5 |
#endif |
| 6 |
#include <math.h> |
| 7 |
#include <limits.h> |
| 8 |
#include <string.h> |
| 9 |
|
| 10 |
//#define DEBUG |
| 11 |
#include "debug.h" |
| 12 |
|
| 13 |
#define hashsize(n) ((cmph_uint32)1<<(n)) |
| 14 |
#define hashmask(n) (hashsize(n)-1) |
| 15 |
|
| 16 |
|
| 17 |
|
| 18 |
//#define NM2 /* Define this if you do not want power of 2 table sizes*/ |
| 19 |
|
| 20 |
|
| 21 |
/* |
| 22 |
-------------------------------------------------------------------- |
| 23 |
mix -- mix 3 32-bit values reversibly. |
| 24 |
For every delta with one or two bits set, and the deltas of all three |
| 25 |
high bits or all three low bits, whether the original value of a,b,c |
| 26 |
is almost all zero or is uniformly distributed, |
| 27 |
* If mix() is run forward or backward, at least 32 bits in a,b,c |
| 28 |
have at least 1/4 probability of changing. |
| 29 |
* If mix() is run forward, every bit of c will change between 1/3 and |
| 30 |
2/3 of the time. (Well, 22/100 and 78/100 for some 2-bit deltas.) |
| 31 |
mix() was built out of 36 single-cycle latency instructions in a |
| 32 |
structure that could supported 2x parallelism, like so: |
| 33 |
a -= b; |
| 34 |
a -= c; x = (c>>13); |
| 35 |
b -= c; a ^= x; |
| 36 |
b -= a; x = (a<<8); |
| 37 |
c -= a; b ^= x; |
| 38 |
c -= b; x = (b>>13); |
| 39 |
... |
| 40 |
Unfortunately, superscalar Pentiums and Sparcs can't take advantage |
| 41 |
of that parallelism. They've also turned some of those single-cycle |
| 42 |
latency instructions into multi-cycle latency instructions. Still, |
| 43 |
this is the fastest good hash I could find. There were about 2^^68 |
| 44 |
to choose from. I only looked at a billion or so. |
| 45 |
-------------------------------------------------------------------- |
| 46 |
*/ |
| 47 |
#define mix(a,b,c) \ |
| 48 |
{ \ |
| 49 |
a -= b; a -= c; a ^= (c>>13); \ |
| 50 |
b -= c; b -= a; b ^= (a<<8); \ |
| 51 |
c -= a; c -= b; c ^= (b>>13); \ |
| 52 |
a -= b; a -= c; a ^= (c>>12); \ |
| 53 |
b -= c; b -= a; b ^= (a<<16); \ |
| 54 |
c -= a; c -= b; c ^= (b>>5); \ |
| 55 |
a -= b; a -= c; a ^= (c>>3); \ |
| 56 |
b -= c; b -= a; b ^= (a<<10); \ |
| 57 |
c -= a; c -= b; c ^= (b>>15); \ |
| 58 |
} |
| 59 |
|
| 60 |
/* |
| 61 |
-------------------------------------------------------------------- |
| 62 |
hash() -- hash a variable-length key into a 32-bit value |
| 63 |
k : the key (the unaligned variable-length array of bytes) |
| 64 |
len : the length of the key, counting by bytes |
| 65 |
initval : can be any 4-byte value |
| 66 |
Returns a 32-bit value. Every bit of the key affects every bit of |
| 67 |
the return value. Every 1-bit and 2-bit delta achieves avalanche. |
| 68 |
About 6*len+35 instructions. |
| 69 |
|
| 70 |
The best hash table sizes are powers of 2. There is no need to do |
| 71 |
mod a prime (mod is sooo slow!). If you need less than 32 bits, |
| 72 |
use a bitmask. For example, if you need only 10 bits, do |
| 73 |
h = (h & hashmask(10)); |
| 74 |
In which case, the hash table should have hashsize(10) elements. |
| 75 |
|
| 76 |
If you are hashing n strings (cmph_uint8 **)k, do it like this: |
| 77 |
for (i=0, h=0; i<n; ++i) h = hash( k[i], len[i], h); |
| 78 |
|
| 79 |
By Bob Jenkins, 1996. [email protected]. You may use this |
| 80 |
code any way you wish, private, educational, or commercial. It's free. |
| 81 |
|
| 82 |
See http://burtleburtle.net/bob/hash/evahash.html |
| 83 |
Use for hash table lookup, or anything where one collision in 2^^32 is |
| 84 |
acceptable. Do NOT use for cryptographic purposes. |
| 85 |
-------------------------------------------------------------------- |
| 86 |
*/ |
| 87 |
jenkins_state_t *jenkins_state_new(cmph_uint32 size) //size of hash table |
| 88 |
{ |
| 89 |
jenkins_state_t *state = (jenkins_state_t *)malloc(sizeof(jenkins_state_t)); |
| 90 |
DEBUGP("Initializing jenkins hash\n"); |
| 91 |
state->seed = ((cmph_uint32)rand() % size); |
| 92 |
return state; |
| 93 |
} |
| 94 |
void jenkins_state_destroy(jenkins_state_t *state) |
| 95 |
{ |
| 96 |
free(state); |
| 97 |
} |
| 98 |
|
| 99 |
|
| 100 |
static inline void __jenkins_hash_vector(cmph_uint32 seed, const char *k, cmph_uint32 keylen, cmph_uint32 * hashes) |
| 101 |
{ |
| 102 |
register cmph_uint32 len, length; |
| 103 |
|
| 104 |
/* Set up the internal state */ |
| 105 |
length = keylen; |
| 106 |
len = length; |
| 107 |
hashes[0] = hashes[1] = 0x9e3779b9; /* the golden ratio; an arbitrary value */ |
| 108 |
hashes[2] = seed; /* the previous hash value - seed in our case */ |
| 109 |
|
| 110 |
/*---------------------------------------- handle most of the key */ |
| 111 |
while (len >= 12) |
| 112 |
{ |
| 113 |
hashes[0] += ((cmph_uint32)k[0] +((cmph_uint32)k[1]<<8) +((cmph_uint32)k[2]<<16) +((cmph_uint32)k[3]<<24)); |
| 114 |
hashes[1] += ((cmph_uint32)k[4] +((cmph_uint32)k[5]<<8) +((cmph_uint32)k[6]<<16) +((cmph_uint32)k[7]<<24)); |
| 115 |
hashes[2] += ((cmph_uint32)k[8] +((cmph_uint32)k[9]<<8) +((cmph_uint32)k[10]<<16)+((cmph_uint32)k[11]<<24)); |
| 116 |
mix(hashes[0],hashes[1],hashes[2]); |
| 117 |
k += 12; len -= 12; |
| 118 |
} |
| 119 |
|
| 120 |
/*------------------------------------- handle the last 11 bytes */ |
| 121 |
hashes[2] += length; |
| 122 |
switch(len) /* all the case statements fall through */ |
| 123 |
{ |
| 124 |
case 11: |
| 125 |
hashes[2] +=((cmph_uint32)k[10]<<24); |
| 126 |
case 10: |
| 127 |
hashes[2] +=((cmph_uint32)k[9]<<16); |
| 128 |
case 9 : |
| 129 |
hashes[2] +=((cmph_uint32)k[8]<<8); |
| 130 |
/* the first byte of hashes[2] is reserved for the length */ |
| 131 |
case 8 : |
| 132 |
hashes[1] +=((cmph_uint32)k[7]<<24); |
| 133 |
case 7 : |
| 134 |
hashes[1] +=((cmph_uint32)k[6]<<16); |
| 135 |
case 6 : |
| 136 |
hashes[1] +=((cmph_uint32)k[5]<<8); |
| 137 |
case 5 : |
| 138 |
hashes[1] +=(cmph_uint8) k[4]; |
| 139 |
case 4 : |
| 140 |
hashes[0] +=((cmph_uint32)k[3]<<24); |
| 141 |
case 3 : |
| 142 |
hashes[0] +=((cmph_uint32)k[2]<<16); |
| 143 |
case 2 : |
| 144 |
hashes[0] +=((cmph_uint32)k[1]<<8); |
| 145 |
case 1 : |
| 146 |
hashes[0] +=(cmph_uint8)k[0]; |
| 147 |
/* case 0: nothing left to add */ |
| 148 |
} |
| 149 |
|
| 150 |
mix(hashes[0],hashes[1],hashes[2]); |
| 151 |
} |
| 152 |
|
| 153 |
cmph_uint32 jenkins_hash(jenkins_state_t *state, const char *k, cmph_uint32 keylen) |
| 154 |
{ |
| 155 |
cmph_uint32 hashes[3]; |
| 156 |
__jenkins_hash_vector(state->seed, k, keylen, hashes); |
| 157 |
return hashes[2]; |
| 158 |
/* cmph_uint32 a, b, c; |
| 159 |
cmph_uint32 len, length; |
| 160 |
|
| 161 |
// Set up the internal state |
| 162 |
length = keylen; |
| 163 |
len = length; |
| 164 |
a = b = 0x9e3779b9; // the golden ratio; an arbitrary value |
| 165 |
c = state->seed; // the previous hash value - seed in our case |
| 166 |
|
| 167 |
// handle most of the key |
| 168 |
while (len >= 12) |
| 169 |
{ |
| 170 |
a += (k[0] +((cmph_uint32)k[1]<<8) +((cmph_uint32)k[2]<<16) +((cmph_uint32)k[3]<<24)); |
| 171 |
b += (k[4] +((cmph_uint32)k[5]<<8) +((cmph_uint32)k[6]<<16) +((cmph_uint32)k[7]<<24)); |
| 172 |
c += (k[8] +((cmph_uint32)k[9]<<8) +((cmph_uint32)k[10]<<16)+((cmph_uint32)k[11]<<24)); |
| 173 |
mix(a,b,c); |
| 174 |
k += 12; len -= 12; |
| 175 |
} |
| 176 |
|
| 177 |
// handle the last 11 bytes |
| 178 |
c += length; |
| 179 |
switch(len) /// all the case statements fall through |
| 180 |
{ |
| 181 |
case 11: |
| 182 |
c +=((cmph_uint32)k[10]<<24); |
| 183 |
case 10: |
| 184 |
c +=((cmph_uint32)k[9]<<16); |
| 185 |
case 9 : |
| 186 |
c +=((cmph_uint32)k[8]<<8); |
| 187 |
// the first byte of c is reserved for the length |
| 188 |
case 8 : |
| 189 |
b +=((cmph_uint32)k[7]<<24); |
| 190 |
case 7 : |
| 191 |
b +=((cmph_uint32)k[6]<<16); |
| 192 |
case 6 : |
| 193 |
b +=((cmph_uint32)k[5]<<8); |
| 194 |
case 5 : |
| 195 |
b +=k[4]; |
| 196 |
case 4 : |
| 197 |
a +=((cmph_uint32)k[3]<<24); |
| 198 |
case 3 : |
| 199 |
a +=((cmph_uint32)k[2]<<16); |
| 200 |
case 2 : |
| 201 |
a +=((cmph_uint32)k[1]<<8); |
| 202 |
case 1 : |
| 203 |
a +=k[0]; |
| 204 |
// case 0: nothing left to add |
| 205 |
} |
| 206 |
|
| 207 |
mix(a,b,c); |
| 208 |
|
| 209 |
/// report the result |
| 210 |
|
| 211 |
return c; |
| 212 |
*/ |
| 213 |
} |
| 214 |
|
| 215 |
void jenkins_hash_vector_(jenkins_state_t *state, const char *k, cmph_uint32 keylen, cmph_uint32 * hashes) |
| 216 |
{ |
| 217 |
__jenkins_hash_vector(state->seed, k, keylen, hashes); |
| 218 |
} |
| 219 |
|
| 220 |
void jenkins_state_dump(jenkins_state_t *state, char **buf, cmph_uint32 *buflen) |
| 221 |
{ |
| 222 |
*buflen = sizeof(cmph_uint32); |
| 223 |
*buf = (char *)malloc(sizeof(cmph_uint32)); |
| 224 |
if (!*buf) |
| 225 |
{ |
| 226 |
*buflen = UINT_MAX; |
| 227 |
return; |
| 228 |
} |
| 229 |
memcpy(*buf, &(state->seed), sizeof(cmph_uint32)); |
| 230 |
DEBUGP("Dumped jenkins state with seed %u\n", state->seed); |
| 231 |
return; |
| 232 |
} |
| 233 |
|
| 234 |
jenkins_state_t *jenkins_state_copy(jenkins_state_t *src_state) |
| 235 |
{ |
| 236 |
jenkins_state_t *dest_state = (jenkins_state_t *)malloc(sizeof(jenkins_state_t)); |
| 237 |
dest_state->hashfunc = src_state->hashfunc; |
| 238 |
dest_state->seed = src_state->seed; |
| 239 |
return dest_state; |
| 240 |
} |
| 241 |
|
| 242 |
jenkins_state_t *jenkins_state_load(const char *buf, cmph_uint32 buflen) |
| 243 |
{ |
| 244 |
jenkins_state_t *state = (jenkins_state_t *)malloc(sizeof(jenkins_state_t)); |
| 245 |
state->seed = *(cmph_uint32 *)buf; |
| 246 |
state->hashfunc = CMPH_HASH_JENKINS; |
| 247 |
DEBUGP("Loaded jenkins state with seed %u\n", state->seed); |
| 248 |
return state; |
| 249 |
} |
| 250 |
|
| 251 |
|
| 252 |
/** \fn void jenkins_state_pack(jenkins_state_t *state, void *jenkins_packed); |
| 253 |
* \brief Support the ability to pack a jenkins function into a preallocated contiguous memory space pointed by jenkins_packed. |
| 254 |
* \param state points to the jenkins function |
| 255 |
* \param jenkins_packed pointer to the contiguous memory area used to store the jenkins function. The size of jenkins_packed must be at least jenkins_state_packed_size() |
| 256 |
*/ |
| 257 |
void jenkins_state_pack(jenkins_state_t *state, void *jenkins_packed) |
| 258 |
{ |
| 259 |
if (state && jenkins_packed) |
| 260 |
{ |
| 261 |
memcpy(jenkins_packed, &(state->seed), sizeof(cmph_uint32)); |
| 262 |
} |
| 263 |
} |
| 264 |
|
| 265 |
/** \fn cmph_uint32 jenkins_state_packed_size(jenkins_state_t *state); |
| 266 |
* \brief Return the amount of space needed to pack a jenkins function. |
| 267 |
* \return the size of the packed function or zero for failures |
| 268 |
*/ |
| 269 |
cmph_uint32 jenkins_state_packed_size(void) |
| 270 |
{ |
| 271 |
return sizeof(cmph_uint32); |
| 272 |
} |
| 273 |
|
| 274 |
|
| 275 |
/** \fn cmph_uint32 jenkins_hash_packed(void *jenkins_packed, const char *k, cmph_uint32 keylen); |
| 276 |
* \param jenkins_packed is a pointer to a contiguous memory area |
| 277 |
* \param key is a pointer to a key |
| 278 |
* \param keylen is the key length |
| 279 |
* \return an integer that represents a hash value of 32 bits. |
| 280 |
*/ |
| 281 |
cmph_uint32 jenkins_hash_packed(void *jenkins_packed, const char *k, cmph_uint32 keylen) |
| 282 |
{ |
| 283 |
cmph_uint32 hashes[3]; |
| 284 |
__jenkins_hash_vector(*((cmph_uint32 *)jenkins_packed), k, keylen, hashes); |
| 285 |
return hashes[2]; |
| 286 |
} |
| 287 |
|
| 288 |
/** \fn jenkins_hash_vector_packed(void *jenkins_packed, const char *k, cmph_uint32 keylen, cmph_uint32 * hashes); |
| 289 |
* \param jenkins_packed is a pointer to a contiguous memory area |
| 290 |
* \param key is a pointer to a key |
| 291 |
* \param keylen is the key length |
| 292 |
* \param hashes is a pointer to a memory large enough to fit three 32-bit integers. |
| 293 |
*/ |
| 294 |
void jenkins_hash_vector_packed(void *jenkins_packed, const char *k, cmph_uint32 keylen, cmph_uint32 * hashes) |
| 295 |
{ |
| 296 |
__jenkins_hash_vector(*((cmph_uint32 *)jenkins_packed), k, keylen, hashes); |
| 297 |
} |