src/shared/MurmurHash3.c

   1 //-----------------------------------------------------------------------------
   2 // MurmurHash3 was written by Austin Appleby, and is placed in the public
   3 // domain. The author hereby disclaims copyright to this source code.
   4
   5 // Note - The x86 and x64 versions do _not_ produce the same results, as the
   6 // algorithms are optimized for their respective platforms. You can still
   7 // compile and run any of them on any platform, but your performance with the
   8 // non-native version will be less than optimal.
   9
  10 #include "MurmurHash3.h"
  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 __attribute__((always_inline))
  33
  34 static inline uint32_t rotl32 ( uint32_t x, int8_t r )
  35 {
  36   return (x << r) | (x >> (32 - r));
  37 }
  38
  39 static 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 static FORCE_INLINE uint32_t getblock32 ( const uint32_t * p, int i )
  56 {
  57   return p[i];
  58 }
  59
  60 static FORCE_INLINE uint64_t getblock64 ( const uint64_t * p, int i )
  61 {
  62   return p[i];
  63 }
  64
  65 //-----------------------------------------------------------------------------
  66 // Finalization mix - force all bits of a hash block to avalanche
  67
  68 static FORCE_INLINE uint32_t fmix32 ( uint32_t h )
  69 {
  70   h ^= h >> 16;
  71   h *= 0x85ebca6b;
  72   h ^= h >> 13;
  73   h *= 0xc2b2ae35;
  74   h ^= h >> 16;
  75
  76   return h;
  77 }
  78
  79 //----------
  80
  81 static FORCE_INLINE uint64_t fmix64 ( uint64_t k )
  82 {
  83   k ^= k >> 33;
  84   k *= BIG_CONSTANT(0xff51afd7ed558ccd);
  85   k ^= k >> 33;
  86   k *= BIG_CONSTANT(0xc4ceb9fe1a85ec53);
  87   k ^= k >> 33;
  88
  89   return k;
  90 }
  91
  92 //-----------------------------------------------------------------------------
  93
  94 void MurmurHash3_x86_32 ( const void * key, size_t len,
  95                           uint32_t seed, void * out )
  96 {
  97   const uint8_t * data = (const uint8_t*)key;
  98   const int nblocks = len / 4;
  99
 100   uint32_t h1 = seed;
 101
 102   const uint32_t c1 = 0xcc9e2d51;
 103   const uint32_t c2 = 0x1b873593;
 104
 105   const uint8_t * tail;
 106   uint32_t k1;
 107
 108   //----------
 109   // body
 110
 111   const uint32_t * blocks = (const uint32_t *)(data + nblocks*4);
 112
 113   for(int i = -nblocks; i; i++)
 114   {
 115     k1 = getblock32(blocks,i);
 116
 117     k1 *= c1;
 118     k1 = ROTL32(k1,15);
 119     k1 *= c2;
 120
 121     h1 ^= k1;
 122     h1 = ROTL32(h1,13);
 123     h1 = h1*5+0xe6546b64;
 124   }
 125
 126   //----------
 127   // tail
 128
 129   tail = (const uint8_t*)(data + nblocks*4);
 130
 131   k1 = 0;
 132
 133   switch(len & 3)
 134   {
 135   case 3: k1 ^= tail[2] << 16;
 136   case 2: k1 ^= tail[1] << 8;
 137   case 1: k1 ^= tail[0];
 138           k1 *= c1; k1 = ROTL32(k1,15); k1 *= c2; h1 ^= k1;
 139   };
 140
 141   //----------
 142   // finalization
 143
 144   h1 ^= len;
 145
 146   h1 = fmix32(h1);
 147
 148   *(uint32_t*)out = h1;
 149 }
 150
 151 //-----------------------------------------------------------------------------
 152
 153 void MurmurHash3_x86_128 ( const void * key, const size_t len,
 154                            uint32_t seed, void * out )
 155 {
 156   const uint8_t * data = (const uint8_t*)key;
 157   const int nblocks = len / 16;
 158
 159   uint32_t h1 = seed;
 160   uint32_t h2 = seed;
 161   uint32_t h3 = seed;
 162   uint32_t h4 = seed;
 163
 164   const uint32_t c1 = 0x239b961b;
 165   const uint32_t c2 = 0xab0e9789;
 166   const uint32_t c3 = 0x38b34ae5;
 167   const uint32_t c4 = 0xa1e38b93;
 168
 169   const uint8_t * tail;
 170
 171   uint32_t k1;
 172   uint32_t k2;
 173   uint32_t k3;
 174   uint32_t k4;
 175
 176   //----------
 177   // body
 178
 179   const uint32_t * blocks = (const uint32_t *)(data + nblocks*16);
 180
 181   for(int i = -nblocks; i; i++)
 182   {
 183     k1 = getblock32(blocks,i*4+0);
 184     k2 = getblock32(blocks,i*4+1);
 185     k3 = getblock32(blocks,i*4+2);
 186     k4 = getblock32(blocks,i*4+3);
 187
 188     k1 *= c1; k1  = ROTL32(k1,15); k1 *= c2; h1 ^= k1;
 189
 190     h1 = ROTL32(h1,19); h1 += h2; h1 = h1*5+0x561ccd1b;
 191
 192     k2 *= c2; k2  = ROTL32(k2,16); k2 *= c3; h2 ^= k2;
 193
 194     h2 = ROTL32(h2,17); h2 += h3; h2 = h2*5+0x0bcaa747;
 195
 196     k3 *= c3; k3  = ROTL32(k3,17); k3 *= c4; h3 ^= k3;
 197
 198     h3 = ROTL32(h3,15); h3 += h4; h3 = h3*5+0x96cd1c35;
 199
 200     k4 *= c4; k4  = ROTL32(k4,18); k4 *= c1; h4 ^= k4;
 201
 202     h4 = ROTL32(h4,13); h4 += h1; h4 = h4*5+0x32ac3b17;
 203   }
 204
 205   //----------
 206   // tail
 207
 208   tail = (const uint8_t*)(data + nblocks*16);
 209
 210   k1 = 0;
 211   k2 = 0;
 212   k3 = 0;
 213   k4 = 0;
 214
 215   switch(len & 15)
 216   {
 217   case 15: k4 ^= tail[14] << 16;
 218   case 14: k4 ^= tail[13] << 8;
 219   case 13: k4 ^= tail[12] << 0;
 220            k4 *= c4; k4  = ROTL32(k4,18); k4 *= c1; h4 ^= k4;
 221
 222   case 12: k3 ^= tail[11] << 24;
 223   case 11: k3 ^= tail[10] << 16;
 224   case 10: k3 ^= tail[ 9] << 8;
 225   case  9: k3 ^= tail[ 8] << 0;
 226            k3 *= c3; k3  = ROTL32(k3,17); k3 *= c4; h3 ^= k3;
 227
 228   case  8: k2 ^= tail[ 7] << 24;
 229   case  7: k2 ^= tail[ 6] << 16;
 230   case  6: k2 ^= tail[ 5] << 8;
 231   case  5: k2 ^= tail[ 4] << 0;
 232            k2 *= c2; k2  = ROTL32(k2,16); k2 *= c3; h2 ^= k2;
 233
 234   case  4: k1 ^= tail[ 3] << 24;
 235   case  3: k1 ^= tail[ 2] << 16;
 236   case  2: k1 ^= tail[ 1] << 8;
 237   case  1: k1 ^= tail[ 0] << 0;
 238            k1 *= c1; k1  = ROTL32(k1,15); k1 *= c2; h1 ^= k1;
 239   };
 240
 241   //----------
 242   // finalization
 243
 244   h1 ^= len; h2 ^= len; h3 ^= len; h4 ^= len;
 245
 246   h1 += h2; h1 += h3; h1 += h4;
 247   h2 += h1; h3 += h1; h4 += h1;
 248
 249   h1 = fmix32(h1);
 250   h2 = fmix32(h2);
 251   h3 = fmix32(h3);
 252   h4 = fmix32(h4);
 253
 254   h1 += h2; h1 += h3; h1 += h4;
 255   h2 += h1; h3 += h1; h4 += h1;
 256
 257   ((uint32_t*)out)[0] = h1;
 258   ((uint32_t*)out)[1] = h2;
 259   ((uint32_t*)out)[2] = h3;
 260   ((uint32_t*)out)[3] = h4;
 261 }
 262
 263 //-----------------------------------------------------------------------------
 264
 265 void MurmurHash3_x64_128 ( const void * key, const size_t len,
 266                            const uint32_t seed, void * out )
 267 {
 268   const uint8_t * data = (const uint8_t*)key;
 269   const int nblocks = len / 16;
 270
 271   uint64_t h1 = seed;
 272   uint64_t h2 = seed;
 273
 274   const uint64_t c1 = BIG_CONSTANT(0x87c37b91114253d5);
 275   const uint64_t c2 = BIG_CONSTANT(0x4cf5ad432745937f);
 276
 277   const uint8_t *tail;
 278
 279   uint64_t k1;
 280   uint64_t k2;
 281
 282   //----------
 283   // body
 284
 285   const uint64_t * blocks = (const uint64_t *)(data);
 286
 287   for(int i = 0; i < nblocks; i++)
 288   {
 289     k1 = getblock64(blocks,i*2+0);
 290     k2 = getblock64(blocks,i*2+1);
 291
 292     k1 *= c1; k1  = ROTL64(k1,31); k1 *= c2; h1 ^= k1;
 293
 294     h1 = ROTL64(h1,27); h1 += h2; h1 = h1*5+0x52dce729;
 295
 296     k2 *= c2; k2  = ROTL64(k2,33); k2 *= c1; h2 ^= k2;
 297
 298     h2 = ROTL64(h2,31); h2 += h1; h2 = h2*5+0x38495ab5;
 299   }
 300
 301   //----------
 302   // tail
 303
 304   tail = (const uint8_t*)(data + nblocks*16);
 305
 306   k1 = 0;
 307   k2 = 0;
 308
 309   switch(len & 15)
 310   {
 311   case 15: k2 ^= (uint64_t)(tail[14]) << 48;
 312   case 14: k2 ^= (uint64_t)(tail[13]) << 40;
 313   case 13: k2 ^= (uint64_t)(tail[12]) << 32;
 314   case 12: k2 ^= (uint64_t)(tail[11]) << 24;
 315   case 11: k2 ^= (uint64_t)(tail[10]) << 16;
 316   case 10: k2 ^= (uint64_t)(tail[ 9]) << 8;
 317   case  9: k2 ^= (uint64_t)(tail[ 8]) << 0;
 318            k2 *= c2; k2  = ROTL64(k2,33); k2 *= c1; h2 ^= k2;
 319
 320   case  8: k1 ^= (uint64_t)(tail[ 7]) << 56;
 321   case  7: k1 ^= (uint64_t)(tail[ 6]) << 48;
 322   case  6: k1 ^= (uint64_t)(tail[ 5]) << 40;
 323   case  5: k1 ^= (uint64_t)(tail[ 4]) << 32;
 324   case  4: k1 ^= (uint64_t)(tail[ 3]) << 24;
 325   case  3: k1 ^= (uint64_t)(tail[ 2]) << 16;
 326   case  2: k1 ^= (uint64_t)(tail[ 1]) << 8;
 327   case  1: k1 ^= (uint64_t)(tail[ 0]) << 0;
 328            k1 *= c1; k1  = ROTL64(k1,31); k1 *= c2; h1 ^= k1;
 329   };
 330
 331   //----------
 332   // finalization
 333
 334   h1 ^= len; h2 ^= len;
 335
 336   h1 += h2;
 337   h2 += h1;
 338
 339   h1 = fmix64(h1);
 340   h2 = fmix64(h2);
 341
 342   h1 += h2;
 343   h2 += h1;
 344
 345   ((uint64_t*)out)[0] = h1;
 346   ((uint64_t*)out)[1] = h2;
 347 }
 348
 349 //-----------------------------------------------------------------------------