| 1 | // MurmurHash3 was written by Austin Appleby, and is placed in the public |
| 2 | // domain. The author hereby disclaims copyright to this source code. |
| 3 | |
| 4 | // Note - The x86 and x64 versions do _not_ produce the same results, as the |
| 5 | // algorithms are optimized for their respective platforms. You can still |
| 6 | // compile and run any of them on any platform, but your performance with the |
| 7 | // non-native version will be less than optimal. |
| 8 | |
| 9 | #include "murmurhash3.h" |
| 10 | #include <cstring> |
| 11 | |
| 12 | //----------------------------------------------------------------------------- |
| 13 | // Platform-specific functions and macros |
| 14 | |
| 15 | // Microsoft Visual Studio |
| 16 | |
| 17 | #if defined(_MSC_VER) |
| 18 | |
| 19 | #define FORCE_INLINE __forceinline |
| 20 | |
| 21 | #include <stdlib.h> |
| 22 | |
| 23 | #define ROTL32(x,y) _rotl(x,y) |
| 24 | #define ROTL64(x,y) _rotl64(x,y) |
| 25 | |
| 26 | #define BIG_CONSTANT(x) (x) |
| 27 | |
| 28 | // Other compilers |
| 29 | |
| 30 | #else // defined(_MSC_VER) |
| 31 | |
| 32 | #define FORCE_INLINE inline __attribute__((always_inline)) |
| 33 | |
| 34 | inline uint32_t rotl32 ( uint32_t x, int8_t r ) |
| 35 | { |
| 36 | return (x << r) | (x >> (32 - r)); |
| 37 | } |
| 38 | |
| 39 | inline uint64_t rotl64 ( uint64_t x, int8_t r ) |
| 40 | { |
| 41 | return (x << r) | (x >> (64 - r)); |
| 42 | } |
| 43 | |
| 44 | #define ROTL32(x,y) rotl32(x,y) |
| 45 | #define ROTL64(x,y) rotl64(x,y) |
| 46 | |
| 47 | #define BIG_CONSTANT(x) (x##LLU) |
| 48 | |
| 49 | #endif // !defined(_MSC_VER) |
| 50 | |
| 51 | //----------------------------------------------------------------------------- |
| 52 | // Block read - if your platform needs to do endian-swapping or can only |
| 53 | // handle aligned reads, do the conversion here |
| 54 | |
| 55 | FORCE_INLINE uint32_t getblock32 ( const uint32_t * p, int i ) |
| 56 | { |
| 57 | uint32_t res; |
| 58 | memcpy(&res, p + i, sizeof(res)); |
| 59 | return res; |
| 60 | } |
| 61 | |
| 62 | FORCE_INLINE uint64_t getblock64 ( const uint64_t * p, int i ) |
| 63 | { |
| 64 | return p[i]; |
| 65 | } |
| 66 | |
| 67 | //----------------------------------------------------------------------------- |
| 68 | // Finalization mix - force all bits of a hash block to avalanche |
| 69 | |
| 70 | FORCE_INLINE uint32_t fmix32 ( uint32_t h ) |
| 71 | { |
| 72 | h ^= h >> 16; |
| 73 | h *= 0x85ebca6b; |
| 74 | h ^= h >> 13; |
| 75 | h *= 0xc2b2ae35; |
| 76 | h ^= h >> 16; |
| 77 | |
| 78 | return h; |
| 79 | } |
| 80 | |
| 81 | //---------- |
| 82 | |
| 83 | FORCE_INLINE uint64_t fmix64 ( uint64_t k ) |
| 84 | { |
| 85 | k ^= k >> 33; |
| 86 | k *= BIG_CONSTANT(0xff51afd7ed558ccd); |
| 87 | k ^= k >> 33; |
| 88 | k *= BIG_CONSTANT(0xc4ceb9fe1a85ec53); |
| 89 | k ^= k >> 33; |
| 90 | |
| 91 | return k; |
| 92 | } |
| 93 | |
| 94 | //----------------------------------------------------------------------------- |
| 95 | |
| 96 | void MurmurHash3_x86_32 ( const void * key, int len, |
| 97 | uint32_t seed, void * out ) |
| 98 | { |
| 99 | const uint8_t * data = (const uint8_t*)key; |
| 100 | const int nblocks = len / 4; |
| 101 | |
| 102 | uint32_t h1 = seed; |
| 103 | |
| 104 | const uint32_t c1 = 0xcc9e2d51; |
| 105 | const uint32_t c2 = 0x1b873593; |
| 106 | |
| 107 | //---------- |
| 108 | // body |
| 109 | |
| 110 | const uint32_t * blocks = (const uint32_t *)(data + nblocks*4); |
| 111 | |
| 112 | for(int i = -nblocks; i; i++) |
| 113 | { |
| 114 | uint32_t k1 = getblock32(blocks,i); |
| 115 | |
| 116 | k1 *= c1; |
| 117 | k1 = ROTL32(k1,15); |
| 118 | k1 *= c2; |
| 119 | |
| 120 | h1 ^= k1; |
| 121 | h1 = ROTL32(h1,13); |
| 122 | h1 = h1*5+0xe6546b64; |
| 123 | } |
| 124 | |
| 125 | //---------- |
| 126 | // tail |
| 127 | |
| 128 | const uint8_t * tail = (const uint8_t*)(data + nblocks*4); |
| 129 | |
| 130 | uint32_t k1 = 0; |
| 131 | |
| 132 | switch(len & 3) |
| 133 | { |
| 134 | case 3: k1 ^= tail[2] << 16; |
| 135 | case 2: k1 ^= tail[1] << 8; |
| 136 | case 1: k1 ^= tail[0]; |
| 137 | k1 *= c1; k1 = ROTL32(k1,15); k1 *= c2; h1 ^= k1; |
| 138 | }; |
| 139 | |
| 140 | //---------- |
| 141 | // finalization |
| 142 | |
| 143 | h1 ^= len; |
| 144 | |
| 145 | h1 = fmix32(h1); |
| 146 | |
| 147 | *(uint32_t*)out = h1; |
| 148 | } |
| 149 | |
| 150 | //----------------------------------------------------------------------------- |
| 151 | |
| 152 | void MurmurHash3_x86_128 ( const void * key, const int len, |
| 153 | uint32_t seed, void * out ) |
| 154 | { |
| 155 | const uint8_t * data = (const uint8_t*)key; |
| 156 | const int nblocks = len / 16; |
| 157 | |
| 158 | uint32_t h1 = seed; |
| 159 | uint32_t h2 = seed; |
| 160 | uint32_t h3 = seed; |
| 161 | uint32_t h4 = seed; |
| 162 | |
| 163 | const uint32_t c1 = 0x239b961b; |
| 164 | const uint32_t c2 = 0xab0e9789; |
| 165 | const uint32_t c3 = 0x38b34ae5; |
| 166 | const uint32_t c4 = 0xa1e38b93; |
| 167 | |
| 168 | //---------- |
| 169 | // body |
| 170 | |
| 171 | const uint32_t * blocks = (const uint32_t *)(data + nblocks*16); |
| 172 | |
| 173 | for(int i = -nblocks; i; i++) |
| 174 | { |
| 175 | uint32_t k1 = getblock32(blocks,i*4+0); |
| 176 | uint32_t k2 = getblock32(blocks,i*4+1); |
| 177 | uint32_t k3 = getblock32(blocks,i*4+2); |
| 178 | uint32_t k4 = getblock32(blocks,i*4+3); |
| 179 | |
| 180 | k1 *= c1; k1 = ROTL32(k1,15); k1 *= c2; h1 ^= k1; |
| 181 | |
| 182 | h1 = ROTL32(h1,19); h1 += h2; h1 = h1*5+0x561ccd1b; |
| 183 | |
| 184 | k2 *= c2; k2 = ROTL32(k2,16); k2 *= c3; h2 ^= k2; |
| 185 | |
| 186 | h2 = ROTL32(h2,17); h2 += h3; h2 = h2*5+0x0bcaa747; |
| 187 | |
| 188 | k3 *= c3; k3 = ROTL32(k3,17); k3 *= c4; h3 ^= k3; |
| 189 | |
| 190 | h3 = ROTL32(h3,15); h3 += h4; h3 = h3*5+0x96cd1c35; |
| 191 | |
| 192 | k4 *= c4; k4 = ROTL32(k4,18); k4 *= c1; h4 ^= k4; |
| 193 | |
| 194 | h4 = ROTL32(h4,13); h4 += h1; h4 = h4*5+0x32ac3b17; |
| 195 | } |
| 196 | |
| 197 | //---------- |
| 198 | // tail |
| 199 | |
| 200 | const uint8_t * tail = (const uint8_t*)(data + nblocks*16); |
| 201 | |
| 202 | uint32_t k1 = 0; |
| 203 | uint32_t k2 = 0; |
| 204 | uint32_t k3 = 0; |
| 205 | uint32_t k4 = 0; |
| 206 | |
| 207 | switch(len & 15) |
| 208 | { |
| 209 | case 15: k4 ^= tail[14] << 16; |
| 210 | case 14: k4 ^= tail[13] << 8; |
| 211 | case 13: k4 ^= tail[12] << 0; |
| 212 | k4 *= c4; k4 = ROTL32(k4,18); k4 *= c1; h4 ^= k4; |
| 213 | |
| 214 | case 12: k3 ^= tail[11] << 24; |
| 215 | case 11: k3 ^= tail[10] << 16; |
| 216 | case 10: k3 ^= tail[ 9] << 8; |
| 217 | case 9: k3 ^= tail[ 8] << 0; |
| 218 | k3 *= c3; k3 = ROTL32(k3,17); k3 *= c4; h3 ^= k3; |
| 219 | |
| 220 | case 8: k2 ^= tail[ 7] << 24; |
| 221 | case 7: k2 ^= tail[ 6] << 16; |
| 222 | case 6: k2 ^= tail[ 5] << 8; |
| 223 | case 5: k2 ^= tail[ 4] << 0; |
| 224 | k2 *= c2; k2 = ROTL32(k2,16); k2 *= c3; h2 ^= k2; |
| 225 | |
| 226 | case 4: k1 ^= tail[ 3] << 24; |
| 227 | case 3: k1 ^= tail[ 2] << 16; |
| 228 | case 2: k1 ^= tail[ 1] << 8; |
| 229 | case 1: k1 ^= tail[ 0] << 0; |
| 230 | k1 *= c1; k1 = ROTL32(k1,15); k1 *= c2; h1 ^= k1; |
| 231 | }; |
| 232 | |
| 233 | //---------- |
| 234 | // finalization |
| 235 | |
| 236 | h1 ^= len; h2 ^= len; h3 ^= len; h4 ^= len; |
| 237 | |
| 238 | h1 += h2; h1 += h3; h1 += h4; |
| 239 | h2 += h1; h3 += h1; h4 += h1; |
| 240 | |
| 241 | h1 = fmix32(h1); |
| 242 | h2 = fmix32(h2); |
| 243 | h3 = fmix32(h3); |
| 244 | h4 = fmix32(h4); |
| 245 | |
| 246 | h1 += h2; h1 += h3; h1 += h4; |
| 247 | h2 += h1; h3 += h1; h4 += h1; |
| 248 | |
| 249 | ((uint32_t*)out)[0] = h1; |
| 250 | ((uint32_t*)out)[1] = h2; |
| 251 | ((uint32_t*)out)[2] = h3; |
| 252 | ((uint32_t*)out)[3] = h4; |
| 253 | } |
| 254 | |
| 255 | //----------------------------------------------------------------------------- |
| 256 | |
| 257 | void MurmurHash3_x64_128 ( const void * key, const int len, |
| 258 | const uint32_t seed, void * out ) |
| 259 | { |
| 260 | const uint8_t * data = (const uint8_t*)key; |
| 261 | const int nblocks = len / 16; |
| 262 | |
| 263 | uint64_t h1 = seed; |
| 264 | uint64_t h2 = seed; |
| 265 | |
| 266 | const uint64_t c1 = BIG_CONSTANT(0x87c37b91114253d5); |
| 267 | const uint64_t c2 = BIG_CONSTANT(0x4cf5ad432745937f); |
| 268 | |
| 269 | //---------- |
| 270 | // body |
| 271 | |
| 272 | const uint64_t * blocks = (const uint64_t *)(data); |
| 273 | |
| 274 | for(int i = 0; i < nblocks; i++) |
| 275 | { |
| 276 | uint64_t k1 = getblock64(blocks,i*2+0); |
| 277 | uint64_t k2 = getblock64(blocks,i*2+1); |
| 278 | |
| 279 | k1 *= c1; k1 = ROTL64(k1,31); k1 *= c2; h1 ^= k1; |
| 280 | |
| 281 | h1 = ROTL64(h1,27); h1 += h2; h1 = h1*5+0x52dce729; |
| 282 | |
| 283 | k2 *= c2; k2 = ROTL64(k2,33); k2 *= c1; h2 ^= k2; |
| 284 | |
| 285 | h2 = ROTL64(h2,31); h2 += h1; h2 = h2*5+0x38495ab5; |
| 286 | } |
| 287 | |
| 288 | //---------- |
| 289 | // tail |
| 290 | |
| 291 | const uint8_t * tail = (const uint8_t*)(data + nblocks*16); |
| 292 | |
| 293 | uint64_t k1 = 0; |
| 294 | uint64_t k2 = 0; |
| 295 | |
| 296 | switch(len & 15) |
| 297 | { |
| 298 | case 15: k2 ^= ((uint64_t)tail[14]) << 48; |
| 299 | case 14: k2 ^= ((uint64_t)tail[13]) << 40; |
| 300 | case 13: k2 ^= ((uint64_t)tail[12]) << 32; |
| 301 | case 12: k2 ^= ((uint64_t)tail[11]) << 24; |
| 302 | case 11: k2 ^= ((uint64_t)tail[10]) << 16; |
| 303 | case 10: k2 ^= ((uint64_t)tail[ 9]) << 8; |
| 304 | case 9: k2 ^= ((uint64_t)tail[ 8]) << 0; |
| 305 | k2 *= c2; k2 = ROTL64(k2,33); k2 *= c1; h2 ^= k2; |
| 306 | |
| 307 | case 8: k1 ^= ((uint64_t)tail[ 7]) << 56; |
| 308 | case 7: k1 ^= ((uint64_t)tail[ 6]) << 48; |
| 309 | case 6: k1 ^= ((uint64_t)tail[ 5]) << 40; |
| 310 | case 5: k1 ^= ((uint64_t)tail[ 4]) << 32; |
| 311 | case 4: k1 ^= ((uint64_t)tail[ 3]) << 24; |
| 312 | case 3: k1 ^= ((uint64_t)tail[ 2]) << 16; |
| 313 | case 2: k1 ^= ((uint64_t)tail[ 1]) << 8; |
| 314 | case 1: k1 ^= ((uint64_t)tail[ 0]) << 0; |
| 315 | k1 *= c1; k1 = ROTL64(k1,31); k1 *= c2; h1 ^= k1; |
| 316 | }; |
| 317 | |
| 318 | //---------- |
| 319 | // finalization |
| 320 | |
| 321 | h1 ^= len; h2 ^= len; |
| 322 | |
| 323 | h1 += h2; |
| 324 | h2 += h1; |
| 325 | |
| 326 | h1 = fmix64(h1); |
| 327 | h2 = fmix64(h2); |
| 328 | |
| 329 | h1 += h2; |
| 330 | h2 += h1; |
| 331 | |
| 332 | ((uint64_t*)out)[0] = h1; |
| 333 | ((uint64_t*)out)[1] = h2; |
| 334 | } |
| 335 | |