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 #include <cstring>
  43 template <typename T>
  44 static inline T getblock(const T *blocks, int i) {
  45   T b;
  46   memcpy(&b, ((const uint8_t *) blocks) + i * sizeof(T), sizeof(T));
  47   return b;
  48 }
  49
  50 //-----------------------------------------------------------------------------
  51 // Finalization mix - force all bits of a hash block to avalanche
  52
  53 static FORCE_INLINE uint32_t fmix32 ( uint32_t h )
  54 {
  55   h ^= h >> 16;
  56   h *= 0x85ebca6b;
  57   h ^= h >> 13;
  58   h *= 0xc2b2ae35;
  59   h ^= h >> 16;
  60
  61   return h;
  62 }
  63
  64 //----------
  65
  66 static FORCE_INLINE uint64_t fmix64 ( uint64_t k )
  67 {
  68   k ^= k >> 33;
  69   k *= BIG_CONSTANT(0xff51afd7ed558ccd);
  70   k ^= k >> 33;
  71   k *= BIG_CONSTANT(0xc4ceb9fe1a85ec53);
  72   k ^= k >> 33;
  73
  74   return k;
  75 }
  76
  77 //-----------------------------------------------------------------------------
  78
  79 void MurmurHash3_x86_32 ( const void * key, int len,
  80                           uint32_t seed, void * out )
  81 {
  82   const uint8_t * data = (const uint8_t*)key;
  83   const int nblocks = len / 4;
  84   int i;
  85
  86   uint32_t h1 = seed;
  87
  88   uint32_t c1 = 0xcc9e2d51;
  89   uint32_t c2 = 0x1b873593;
  90
  91   //----------
  92   // body
  93
  94   const uint32_t * blocks = (const uint32_t *)(data + nblocks*4);
  95
  96   for(i = -nblocks; i; i++)
  97   {
  98     uint32_t k1 = getblock(blocks,i);
  99
 100     k1 *= c1;
 101     k1 = ROTL32(k1,15);
 102     k1 *= c2;
 103
 104     h1 ^= k1;
 105     h1 = ROTL32(h1,13);
 106     h1 = h1*5+0xe6546b64;
 107   }
 108
 109   //----------
 110   // tail
 111
 112   const uint8_t * tail = (const uint8_t*)(data + nblocks*4);
 113
 114   uint32_t k1 = 0;
 115
 116   switch(len & 3)
 117   {
 118   case 3: k1 ^= tail[2] << 16;
 119           /* fall through */
 120   case 2: k1 ^= tail[1] << 8;
 121           /* fall through */
 122   case 1: k1 ^= tail[0];
 123           k1 *= c1; k1 = ROTL32(k1,15); k1 *= c2; h1 ^= k1;
 124   };
 125
 126   //----------
 127   // finalization
 128
 129   h1 ^= len;
 130
 131   h1 = fmix32(h1);
 132
 133   *(uint32_t*)out = h1;
 134 }
 135
 136 //-----------------------------------------------------------------------------
 137
 138 void MurmurHash3_x86_128 ( const void * key, const int len,
 139                            uint32_t seed, void * out )
 140 {
 141   const uint8_t * data = (const uint8_t*)key;
 142   const int nblocks = len / 16;
 143   int i;
 144
 145   uint32_t h1 = seed;
 146   uint32_t h2 = seed;
 147   uint32_t h3 = seed;
 148   uint32_t h4 = seed;
 149
 150   uint32_t c1 = 0x239b961b;
 151   uint32_t c2 = 0xab0e9789;
 152   uint32_t c3 = 0x38b34ae5;
 153   uint32_t c4 = 0xa1e38b93;
 154
 155   //----------
 156   // body
 157
 158   const uint32_t * blocks = (const uint32_t *)(data + nblocks*16);
 159
 160   for(i = -nblocks; i; i++)
 161   {
 162     uint32_t k1 = getblock(blocks,i*4+0);
 163     uint32_t k2 = getblock(blocks,i*4+1);
 164     uint32_t k3 = getblock(blocks,i*4+2);
 165     uint32_t k4 = getblock(blocks,i*4+3);
 166
 167     k1 *= c1; k1  = ROTL32(k1,15); k1 *= c2; h1 ^= k1;
 168
 169     h1 = ROTL32(h1,19); h1 += h2; h1 = h1*5+0x561ccd1b;
 170
 171     k2 *= c2; k2  = ROTL32(k2,16); k2 *= c3; h2 ^= k2;
 172
 173     h2 = ROTL32(h2,17); h2 += h3; h2 = h2*5+0x0bcaa747;
 174
 175     k3 *= c3; k3  = ROTL32(k3,17); k3 *= c4; h3 ^= k3;
 176
 177     h3 = ROTL32(h3,15); h3 += h4; h3 = h3*5+0x96cd1c35;
 178
 179     k4 *= c4; k4  = ROTL32(k4,18); k4 *= c1; h4 ^= k4;
 180
 181     h4 = ROTL32(h4,13); h4 += h1; h4 = h4*5+0x32ac3b17;
 182   }
 183
 184   //----------
 185   // tail
 186
 187   const uint8_t * tail = (const uint8_t*)(data + nblocks*16);
 188
 189   uint32_t k1 = 0;
 190   uint32_t k2 = 0;
 191   uint32_t k3 = 0;
 192   uint32_t k4 = 0;
 193
 194   switch(len & 15)
 195   {
 196   case 15: k4 ^= tail[14] << 16;
 197            /* fall through */
 198   case 14: k4 ^= tail[13] << 8;
 199            /* fall through */
 200   case 13: k4 ^= tail[12] << 0;
 201            k4 *= c4; k4  = ROTL32(k4,18); k4 *= c1; h4 ^= k4;
 202            /* fall through */
 203   case 12: k3 ^= tail[11] << 24;
 204            /* fall through */
 205   case 11: k3 ^= tail[10] << 16;
 206            /* fall through */
 207   case 10: k3 ^= tail[ 9] << 8;
 208            /* fall through */
 209   case  9: k3 ^= tail[ 8] << 0;
 210            k3 *= c3; k3  = ROTL32(k3,17); k3 *= c4; h3 ^= k3;
 211            /* fall through */
 212   case  8: k2 ^= tail[ 7] << 24;
 213            /* fall through */
 214   case  7: k2 ^= tail[ 6] << 16;
 215            /* fall through */
 216   case  6: k2 ^= tail[ 5] << 8;
 217            /* fall through */
 218   case  5: k2 ^= tail[ 4] << 0;
 219            k2 *= c2; k2  = ROTL32(k2,16); k2 *= c3; h2 ^= k2;
 220            /* fall through */
 221   case  4: k1 ^= tail[ 3] << 24;
 222            /* fall through */
 223   case  3: k1 ^= tail[ 2] << 16;
 224            /* fall through */
 225   case  2: k1 ^= tail[ 1] << 8;
 226            /* fall through */
 227   case  1: k1 ^= tail[ 0] << 0;
 228            k1 *= c1; k1  = ROTL32(k1,15); k1 *= c2; h1 ^= k1;
 229   };
 230
 231   //----------
 232   // finalization
 233
 234   h1 ^= len; h2 ^= len; h3 ^= len; h4 ^= len;
 235
 236   h1 += h2; h1 += h3; h1 += h4;
 237   h2 += h1; h3 += h1; h4 += h1;
 238
 239   h1 = fmix32(h1);
 240   h2 = fmix32(h2);
 241   h3 = fmix32(h3);
 242   h4 = fmix32(h4);
 243
 244   h1 += h2; h1 += h3; h1 += h4;
 245   h2 += h1; h3 += h1; h4 += h1;
 246
 247   ((uint32_t*)out)[0] = h1;
 248   ((uint32_t*)out)[1] = h2;
 249   ((uint32_t*)out)[2] = h3;
 250   ((uint32_t*)out)[3] = h4;
 251 }
 252
 253 //-----------------------------------------------------------------------------
 254
 255 void MurmurHash3_x64_128 ( const void * key, const int len,
 256                            const uint32_t seed, void * out )
 257 {
 258   const uint8_t * data = (const uint8_t*)key;
 259   const int nblocks = len / 16;
 260   int i;
 261
 262   uint64_t h1 = seed;
 263   uint64_t h2 = seed;
 264
 265   uint64_t c1 = BIG_CONSTANT(0x87c37b91114253d5);
 266   uint64_t c2 = BIG_CONSTANT(0x4cf5ad432745937f);
 267
 268   //----------
 269   // body
 270
 271   const uint64_t * blocks = (const uint64_t *)(data);
 272
 273   for(i = 0; i < nblocks; i++)
 274   {
 275     uint64_t k1 = getblock(blocks,i*2+0);
 276     uint64_t k2 = getblock(blocks,i*2+1);
 277
 278     k1 *= c1; k1  = ROTL64(k1,31); k1 *= c2; h1 ^= k1;
 279
 280     h1 = ROTL64(h1,27); h1 += h2; h1 = h1*5+0x52dce729;
 281
 282     k2 *= c2; k2  = ROTL64(k2,33); k2 *= c1; h2 ^= k2;
 283
 284     h2 = ROTL64(h2,31); h2 += h1; h2 = h2*5+0x38495ab5;
 285   }
 286
 287   //----------
 288   // tail
 289
 290   const uint8_t * tail = (const uint8_t*)(data + nblocks*16);
 291
 292   uint64_t k1 = 0;
 293   uint64_t k2 = 0;
 294
 295   switch(len & 15)
 296   {
 297   case 15: k2 ^= (uint64_t)(tail[14]) << 48;
 298            /* fall through */
 299   case 14: k2 ^= (uint64_t)(tail[13]) << 40;
 300            /* fall through */
 301   case 13: k2 ^= (uint64_t)(tail[12]) << 32;
 302            /* fall through */
 303   case 12: k2 ^= (uint64_t)(tail[11]) << 24;
 304            /* fall through */
 305   case 11: k2 ^= (uint64_t)(tail[10]) << 16;
 306            /* fall through */
 307   case 10: k2 ^= (uint64_t)(tail[ 9]) << 8;
 308            /* fall through */
 309   case  9: k2 ^= (uint64_t)(tail[ 8]) << 0;
 310            k2 *= c2; k2  = ROTL64(k2,33); k2 *= c1; h2 ^= k2;
 311            /* fall through */
 312   case  8: k1 ^= (uint64_t)(tail[ 7]) << 56;
 313            /* fall through */
 314   case  7: k1 ^= (uint64_t)(tail[ 6]) << 48;
 315            /* fall through */
 316   case  6: k1 ^= (uint64_t)(tail[ 5]) << 40;
 317            /* fall through */
 318   case  5: k1 ^= (uint64_t)(tail[ 4]) << 32;
 319            /* fall through */
 320   case  4: k1 ^= (uint64_t)(tail[ 3]) << 24;
 321            /* fall through */
 322   case  3: k1 ^= (uint64_t)(tail[ 2]) << 16;
 323            /* fall through */
 324   case  2: k1 ^= (uint64_t)(tail[ 1]) << 8;
 325            /* fall through */
 326   case  1: k1 ^= (uint64_t)(tail[ 0]) << 0;
 327            k1 *= c1; k1  = ROTL64(k1,31); k1 *= c2; h1 ^= k1;
 328   };
 329
 330   //----------
 331   // finalization
 332
 333   h1 ^= len; h2 ^= len;
 334
 335   h1 += h2;
 336   h2 += h1;
 337
 338   h1 = fmix64(h1);
 339   h2 = fmix64(h2);
 340
 341   h1 += h2;
 342   h2 += h1;
 343
 344   ((uint64_t*)out)[0] = h1;
 345   ((uint64_t*)out)[1] = h2;
 346 }
 347
 348 //-----------------------------------------------------------------------------
 349