git.m6w6.name Git - awesomized/libmemcached/blob - libhashkit/murmur3.cc

   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 "libhashkit/hashkitcon.h"
  11
  12 #include "libhashkit/murmur3.h"
  13
  14 //-----------------------------------------------------------------------------
  15 // Platform-specific functions and macros
  16
  17 #ifdef __GNUC__
  18 #define FORCE_INLINE __attribute__((always_inline)) inline
  19 #else
  20 #define FORCE_INLINE inline
  21 #endif
  22
  23 static FORCE_INLINE uint32_t rotl32 ( uint32_t x, int8_t r )
  24 {
  25   return (x << r) | (x >> (32 - r));
  26 }
  27
  28 static FORCE_INLINE uint64_t rotl64 ( uint64_t x, int8_t r )
  29 {
  30   return (x << r) | (x >> (64 - r));
  31 }
  32
  33 #define ROTL32(x,y)     rotl32(x,y)
  34 #define ROTL64(x,y)     rotl64(x,y)
  35
  36 #define BIG_CONSTANT(x) (x##LLU)
  37
  38 //-----------------------------------------------------------------------------
  39 // Block read - if your platform needs to do endian-swapping or can only
  40 // handle aligned reads, do the conversion here
  41
  42 #define getblock(p, i) (p[i])
  43
  44 //-----------------------------------------------------------------------------
  45 // Finalization mix - force all bits of a hash block to avalanche
  46
  47 static FORCE_INLINE uint32_t fmix32 ( uint32_t h )
  48 {
  49   h ^= h >> 16;
  50   h *= 0x85ebca6b;
  51   h ^= h >> 13;
  52   h *= 0xc2b2ae35;
  53   h ^= h >> 16;
  54
  55   return h;
  56 }
  57
  58 //----------
  59
  60 static FORCE_INLINE uint64_t fmix64 ( uint64_t k )
  61 {
  62   k ^= k >> 33;
  63   k *= BIG_CONSTANT(0xff51afd7ed558ccd);
  64   k ^= k >> 33;
  65   k *= BIG_CONSTANT(0xc4ceb9fe1a85ec53);
  66   k ^= k >> 33;
  67
  68   return k;
  69 }
  70
  71 //-----------------------------------------------------------------------------
  72
  73 void MurmurHash3_x86_32 ( const void * key, int len,
  74                           uint32_t seed, void * out )
  75 {
  76   const uint8_t * data = (const uint8_t*)key;
  77   const int nblocks = len / 4;
  78   int i;
  79
  80   uint32_t h1 = seed;
  81
  82   uint32_t c1 = 0xcc9e2d51;
  83   uint32_t c2 = 0x1b873593;
  84
  85   //----------
  86   // body
  87
  88   const uint32_t * blocks = (const uint32_t *)(data + nblocks*4);
  89
  90   for(i = -nblocks; i; i++)
  91   {
  92     uint32_t k1 = getblock(blocks,i);
  93
  94     k1 *= c1;
  95     k1 = ROTL32(k1,15);
  96     k1 *= c2;
  97
  98     h1 ^= k1;
  99     h1 = ROTL32(h1,13);
 100     h1 = h1*5+0xe6546b64;
 101   }
 102
 103   //----------
 104   // tail
 105
 106   const uint8_t * tail = (const uint8_t*)(data + nblocks*4);
 107
 108   uint32_t k1 = 0;
 109
 110   switch(len & 3)
 111   {
 112   case 3: k1 ^= tail[2] << 16;  /* fall through */
 113   case 2: k1 ^= tail[1] << 8;   /* fall through */
 114   case 1: k1 ^= tail[0];
 115           k1 *= c1; k1 = ROTL32(k1,15); k1 *= c2; h1 ^= k1;
 116   };
 117
 118   //----------
 119   // finalization
 120
 121   h1 ^= len;
 122
 123   h1 = fmix32(h1);
 124
 125   *(uint32_t*)out = h1;
 126 }
 127
 128 //-----------------------------------------------------------------------------
 129
 130 void MurmurHash3_x86_128 ( const void * key, const int len,
 131                            uint32_t seed, void * out )
 132 {
 133   const uint8_t * data = (const uint8_t*)key;
 134   const int nblocks = len / 16;
 135   int i;
 136
 137   uint32_t h1 = seed;
 138   uint32_t h2 = seed;
 139   uint32_t h3 = seed;
 140   uint32_t h4 = seed;
 141
 142   uint32_t c1 = 0x239b961b;
 143   uint32_t c2 = 0xab0e9789;
 144   uint32_t c3 = 0x38b34ae5;
 145   uint32_t c4 = 0xa1e38b93;
 146
 147   //----------
 148   // body
 149
 150   const uint32_t * blocks = (const uint32_t *)(data + nblocks*16);
 151
 152   for(i = -nblocks; i; i++)
 153   {
 154     uint32_t k1 = getblock(blocks,i*4+0);
 155     uint32_t k2 = getblock(blocks,i*4+1);
 156     uint32_t k3 = getblock(blocks,i*4+2);
 157     uint32_t k4 = getblock(blocks,i*4+3);
 158
 159     k1 *= c1; k1  = ROTL32(k1,15); k1 *= c2; h1 ^= k1;
 160
 161     h1 = ROTL32(h1,19); h1 += h2; h1 = h1*5+0x561ccd1b;
 162
 163     k2 *= c2; k2  = ROTL32(k2,16); k2 *= c3; h2 ^= k2;
 164
 165     h2 = ROTL32(h2,17); h2 += h3; h2 = h2*5+0x0bcaa747;
 166
 167     k3 *= c3; k3  = ROTL32(k3,17); k3 *= c4; h3 ^= k3;
 168
 169     h3 = ROTL32(h3,15); h3 += h4; h3 = h3*5+0x96cd1c35;
 170
 171     k4 *= c4; k4  = ROTL32(k4,18); k4 *= c1; h4 ^= k4;
 172
 173     h4 = ROTL32(h4,13); h4 += h1; h4 = h4*5+0x32ac3b17;
 174   }
 175
 176   //----------
 177   // tail
 178
 179   const uint8_t * tail = (const uint8_t*)(data + nblocks*16);
 180
 181   uint32_t k1 = 0;
 182   uint32_t k2 = 0;
 183   uint32_t k3 = 0;
 184   uint32_t k4 = 0;
 185
 186   switch(len & 15)
 187   {
 188   case 15: k4 ^= tail[14] << 16;  /* fall through */
 189   case 14: k4 ^= tail[13] << 8;   /* fall through */
 190   case 13: k4 ^= tail[12] << 0;
 191            k4 *= c4; k4  = ROTL32(k4,18); k4 *= c1; h4 ^= k4;
 192                                   /* fall through */
 193   case 12: k3 ^= tail[11] << 24;  /* fall through */
 194   case 11: k3 ^= tail[10] << 16;  /* fall through */
 195   case 10: k3 ^= tail[ 9] << 8;   /* fall through */
 196   case  9: k3 ^= tail[ 8] << 0;   /* fall through */
 197            k3 *= c3; k3  = ROTL32(k3,17); k3 *= c4; h3 ^= k3;
 198                                   /* fall through */
 199   case  8: k2 ^= tail[ 7] << 24;  /* fall through */
 200   case  7: k2 ^= tail[ 6] << 16;  /* fall through */
 201   case  6: k2 ^= tail[ 5] << 8;   /* fall through */
 202   case  5: k2 ^= tail[ 4] << 0;   /* fall through */
 203            k2 *= c2; k2  = ROTL32(k2,16); k2 *= c3; h2 ^= k2;
 204                                   /* fall through */
 205   case  4: k1 ^= tail[ 3] << 24;  /* fall through */
 206   case  3: k1 ^= tail[ 2] << 16;  /* fall through */
 207   case  2: k1 ^= tail[ 1] << 8;   /* fall through */
 208   case  1: k1 ^= tail[ 0] << 0;   /* fall through */
 209            k1 *= c1; k1  = ROTL32(k1,15); k1 *= c2; h1 ^= k1;
 210   };
 211
 212   //----------
 213   // finalization
 214
 215   h1 ^= len; h2 ^= len; h3 ^= len; h4 ^= len;
 216
 217   h1 += h2; h1 += h3; h1 += h4;
 218   h2 += h1; h3 += h1; h4 += h1;
 219
 220   h1 = fmix32(h1);
 221   h2 = fmix32(h2);
 222   h3 = fmix32(h3);
 223   h4 = fmix32(h4);
 224
 225   h1 += h2; h1 += h3; h1 += h4;
 226   h2 += h1; h3 += h1; h4 += h1;
 227
 228   ((uint32_t*)out)[0] = h1;
 229   ((uint32_t*)out)[1] = h2;
 230   ((uint32_t*)out)[2] = h3;
 231   ((uint32_t*)out)[3] = h4;
 232 }
 233
 234 //-----------------------------------------------------------------------------
 235
 236 void MurmurHash3_x64_128 ( const void * key, const int len,
 237                            const uint32_t seed, void * out )
 238 {
 239   const uint8_t * data = (const uint8_t*)key;
 240   const int nblocks = len / 16;
 241   int i;
 242
 243   uint64_t h1 = seed;
 244   uint64_t h2 = seed;
 245
 246   uint64_t c1 = BIG_CONSTANT(0x87c37b91114253d5);
 247   uint64_t c2 = BIG_CONSTANT(0x4cf5ad432745937f);
 248
 249   //----------
 250   // body
 251
 252   const uint64_t * blocks = (const uint64_t *)(data);
 253
 254   for(i = 0; i < nblocks; i++)
 255   {
 256     uint64_t k1 = getblock(blocks,i*2+0);
 257     uint64_t k2 = getblock(blocks,i*2+1);
 258
 259     k1 *= c1; k1  = ROTL64(k1,31); k1 *= c2; h1 ^= k1;
 260
 261     h1 = ROTL64(h1,27); h1 += h2; h1 = h1*5+0x52dce729;
 262
 263     k2 *= c2; k2  = ROTL64(k2,33); k2 *= c1; h2 ^= k2;
 264
 265     h2 = ROTL64(h2,31); h2 += h1; h2 = h2*5+0x38495ab5;
 266   }
 267
 268   //----------
 269   // tail
 270
 271   const uint8_t * tail = (const uint8_t*)(data + nblocks*16);
 272
 273   uint64_t k1 = 0;
 274   uint64_t k2 = 0;
 275
 276   switch(len & 15)
 277   {
 278   case 15: k2 ^= (uint64_t)(tail[14]) << 48;    /* fall through */
 279   case 14: k2 ^= (uint64_t)(tail[13]) << 40;    /* fall through */
 280   case 13: k2 ^= (uint64_t)(tail[12]) << 32;    /* fall through */
 281   case 12: k2 ^= (uint64_t)(tail[11]) << 24;    /* fall through */
 282   case 11: k2 ^= (uint64_t)(tail[10]) << 16;    /* fall through */
 283   case 10: k2 ^= (uint64_t)(tail[ 9]) << 8;     /* fall through */
 284   case  9: k2 ^= (uint64_t)(tail[ 8]) << 0;     /* fall through */
 285            k2 *= c2; k2  = ROTL64(k2,33); k2 *= c1; h2 ^= k2;
 286                                                /* fall through */
 287   case  8: k1 ^= (uint64_t)(tail[ 7]) << 56;   /* fall through */
 288   case  7: k1 ^= (uint64_t)(tail[ 6]) << 48;   /* fall through */
 289   case  6: k1 ^= (uint64_t)(tail[ 5]) << 40;   /* fall through */
 290   case  5: k1 ^= (uint64_t)(tail[ 4]) << 32;   /* fall through */
 291   case  4: k1 ^= (uint64_t)(tail[ 3]) << 24;   /* fall through */
 292   case  3: k1 ^= (uint64_t)(tail[ 2]) << 16;   /* fall through */
 293   case  2: k1 ^= (uint64_t)(tail[ 1]) << 8;    /* fall through */
 294   case  1: k1 ^= (uint64_t)(tail[ 0]) << 0;    /* fall through */
 295            k1 *= c1; k1  = ROTL64(k1,31); k1 *= c2; h1 ^= k1;
 296   };
 297
 298   //----------
 299   // finalization
 300
 301   h1 ^= len; h2 ^= len;
 302
 303   h1 += h2;
 304   h2 += h1;
 305
 306   h1 = fmix64(h1);
 307   h2 = fmix64(h2);
 308
 309   h1 += h2;
 310   h2 += h1;
 311
 312   ((uint64_t*)out)[0] = h1;
 313   ((uint64_t*)out)[1] = h2;
 314 }
 315
 316 //-----------------------------------------------------------------------------
 317