xxHash/xxhash.c

   1 /*
   2 *  xxHash - Fast Hash algorithm
   3 *  Copyright (C) 2012-2016, Yann Collet
   4 *
   5 *  BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
   6 *
   7 *  Redistribution and use in source and binary forms, with or without
   8 *  modification, are permitted provided that the following conditions are
   9 *  met:
  10 *
  11 *  * Redistributions of source code must retain the above copyright
  12 *  notice, this list of conditions and the following disclaimer.
  13 *  * Redistributions in binary form must reproduce the above
  14 *  copyright notice, this list of conditions and the following disclaimer
  15 *  in the documentation and/or other materials provided with the
  16 *  distribution.
  17 *
  18 *  THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  19 *  "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  20 *  LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  21 *  A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  22 *  OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  23 *  SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  24 *  LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  25 *  DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  26 *  THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  27 *  (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  28 *  OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  29 *
  30 *  You can contact the author at :
  31 *  - xxHash homepage: http://www.xxhash.com
  32 *  - xxHash source repository : https://github.com/Cyan4973/xxHash
  33 */
  34
  35
  36 /* *************************************
  37 *  Tuning parameters
  38 ***************************************/
  39 /*!XXH_FORCE_MEMORY_ACCESS :
  40  * By default, access to unaligned memory is controlled by `memcpy()`, which is safe and portable.
  41  * Unfortunately, on some target/compiler combinations, the generated assembly is sub-optimal.
  42  * The below switch allow to select different access method for improved performance.
  43  * Method 0 (default) : use `memcpy()`. Safe and portable.
  44  * Method 1 : `__packed` statement. It depends on compiler extension (ie, not portable).
  45  *            This method is safe if your compiler supports it, and *generally* as fast or faster than `memcpy`.
  46  * Method 2 : direct access. This method doesn't depend on compiler but violate C standard.
  47  *            It can generate buggy code on targets which do not support unaligned memory accesses.
  48  *            But in some circumstances, it's the only known way to get the most performance (ie GCC + ARMv6)
  49  * See http://stackoverflow.com/a/32095106/646947 for details.
  50  * Prefer these methods in priority order (0 > 1 > 2)
  51  */
  52 #ifndef XXH_FORCE_MEMORY_ACCESS   /* can be defined externally, on command line for example */
  53 #  if defined(__GNUC__) && ( defined(__ARM_ARCH_6__) || defined(__ARM_ARCH_6J__) || defined(__ARM_ARCH_6K__) || defined(__ARM_ARCH_6Z__) || defined(__ARM_ARCH_6ZK__) || defined(__ARM_ARCH_6T2__) )
  54 #    define XXH_FORCE_MEMORY_ACCESS 2
  55 #  elif defined(__INTEL_COMPILER) || \
  56   (defined(__GNUC__) && ( defined(__ARM_ARCH_7__) || defined(__ARM_ARCH_7A__) || defined(__ARM_ARCH_7R__) || defined(__ARM_ARCH_7M__) || defined(__ARM_ARCH_7S__) ))
  57 #    define XXH_FORCE_MEMORY_ACCESS 1
  58 #  endif
  59 #endif
  60
  61 /*!XXH_ACCEPT_NULL_INPUT_POINTER :
  62  * If the input pointer is a null pointer, xxHash default behavior is to trigger a memory access error, since it is a bad pointer.
  63  * When this option is enabled, xxHash output for null input pointers will be the same as a null-length input.
  64  * By default, this option is disabled. To enable it, uncomment below define :
  65  */
  66 /* #define XXH_ACCEPT_NULL_INPUT_POINTER 1 */
  67
  68 /*!XXH_FORCE_NATIVE_FORMAT :
  69  * By default, xxHash library provides endian-independent Hash values, based on little-endian convention.
  70  * Results are therefore identical for little-endian and big-endian CPU.
  71  * This comes at a performance cost for big-endian CPU, since some swapping is required to emulate little-endian format.
  72  * Should endian-independence be of no importance for your application, you may set the #define below to 1,
  73  * to improve speed for Big-endian CPU.
  74  * This option has no impact on Little_Endian CPU.
  75  */
  76 #ifndef XXH_FORCE_NATIVE_FORMAT   /* can be defined externally */
  77 #  define XXH_FORCE_NATIVE_FORMAT 0
  78 #endif
  79
  80 /*!XXH_FORCE_ALIGN_CHECK :
  81  * This is a minor performance trick, only useful with lots of very small keys.
  82  * It means : check for aligned/unaligned input.
  83  * The check costs one initial branch per hash;
  84  * set it to 0 when the input is guaranteed to be aligned,
  85  * or when alignment doesn't matter for performance.
  86  */
  87 #ifndef XXH_FORCE_ALIGN_CHECK /* can be defined externally */
  88 #  if defined(__i386) || defined(_M_IX86) || defined(__x86_64__) || defined(_M_X64)
  89 #    define XXH_FORCE_ALIGN_CHECK 0
  90 #  else
  91 #    define XXH_FORCE_ALIGN_CHECK 1
  92 #  endif
  93 #endif
  94
  95
  96 /* *************************************
  97 *  Includes & Memory related functions
  98 ***************************************/
  99 /*! Modify the local functions below should you wish to use some other memory routines
 100 *   for malloc(), free() */
 101 #include <stdlib.h>
 102 static void* XXH_malloc(size_t s) { return malloc(s); }
 103 static void  XXH_free  (void* p)  { free(p); }
 104 /*! and for memcpy() */
 105 #include <string.h>
 106 static void* XXH_memcpy(void* dest, const void* src, size_t size) { return memcpy(dest,src,size); }
 107
 108 #define XXH_STATIC_LINKING_ONLY
 109 #include "xxhash.h"
 110
 111
 112 /* *************************************
 113 *  Compiler Specific Options
 114 ***************************************/
 115 #ifdef _MSC_VER    /* Visual Studio */
 116 #  pragma warning(disable : 4127)      /* disable: C4127: conditional expression is constant */
 117 #  define FORCE_INLINE static __forceinline
 118 #else
 119 #  if defined (__cplusplus) || defined (__STDC_VERSION__) && __STDC_VERSION__ >= 199901L   /* C99 */
 120 #    ifdef __GNUC__
 121 #      define FORCE_INLINE static inline __attribute__((always_inline))
 122 #    else
 123 #      define FORCE_INLINE static inline
 124 #    endif
 125 #  else
 126 #    define FORCE_INLINE static
 127 #  endif /* __STDC_VERSION__ */
 128 #endif
 129
 130
 131 /* *************************************
 132 *  Basic Types
 133 ***************************************/
 134 #ifndef MEM_MODULE
 135 # if !defined (__VMS) && (defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) )
 136 #   include <stdint.h>
 137     typedef uint8_t  BYTE;
 138     typedef uint16_t U16;
 139     typedef uint32_t U32;
 140 # else
 141     typedef unsigned char      BYTE;
 142     typedef unsigned short     U16;
 143     typedef unsigned int       U32;
 144 # endif
 145 #endif
 146
 147 #if (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==2))
 148
 149 /* Force direct memory access. Only works on CPU which support unaligned memory access in hardware */
 150 static U32 XXH_read32(const void* memPtr) { return *(const U32*) memPtr; }
 151
 152 #elif (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==1))
 153
 154 /* __pack instructions are safer, but compiler specific, hence potentially problematic for some compilers */
 155 /* currently only defined for gcc and icc */
 156 typedef union { U32 u32; } __attribute__((packed)) unalign;
 157 static U32 XXH_read32(const void* ptr) { return ((const unalign*)ptr)->u32; }
 158
 159 #else
 160
 161 /* portable and safe solution. Generally efficient.
 162  * see : http://stackoverflow.com/a/32095106/646947
 163  */
 164 static U32 XXH_read32(const void* memPtr)
 165 {
 166     U32 val;
 167     memcpy(&val, memPtr, sizeof(val));
 168     return val;
 169 }
 170
 171 #endif   /* XXH_FORCE_DIRECT_MEMORY_ACCESS */
 172
 173
 174 /* ****************************************
 175 *  Compiler-specific Functions and Macros
 176 ******************************************/
 177 #define XXH_GCC_VERSION (__GNUC__ * 100 + __GNUC_MINOR__)
 178
 179 /* Note : although _rotl exists for minGW (GCC under windows), performance seems poor */
 180 #if defined(_MSC_VER)
 181 #  define XXH_rotl32(x,r) _rotl(x,r)
 182 #  define XXH_rotl64(x,r) _rotl64(x,r)
 183 #else
 184 #  define XXH_rotl32(x,r) ((x << r) | (x >> (32 - r)))
 185 #  define XXH_rotl64(x,r) ((x << r) | (x >> (64 - r)))
 186 #endif
 187
 188 #if defined(_MSC_VER)     /* Visual Studio */
 189 #  define XXH_swap32 _byteswap_ulong
 190 #elif XXH_GCC_VERSION >= 403
 191 #  define XXH_swap32 __builtin_bswap32
 192 #else
 193 static U32 XXH_swap32 (U32 x)
 194 {
 195     return  ((x << 24) & 0xff000000 ) |
 196             ((x <<  8) & 0x00ff0000 ) |
 197             ((x >>  8) & 0x0000ff00 ) |
 198             ((x >> 24) & 0x000000ff );
 199 }
 200 #endif
 201
 202
 203 /* *************************************
 204 *  Architecture Macros
 205 ***************************************/
 206 typedef enum { XXH_bigEndian=0, XXH_littleEndian=1 } XXH_endianess;
 207
 208 /* XXH_CPU_LITTLE_ENDIAN can be defined externally, for example on the compiler command line */
 209 #ifndef XXH_CPU_LITTLE_ENDIAN
 210     static const int g_one = 1;
 211 #   define XXH_CPU_LITTLE_ENDIAN   (*(const char*)(&g_one))
 212 #endif
 213
 214
 215 /* ***************************
 216 *  Memory reads
 217 *****************************/
 218 typedef enum { XXH_aligned, XXH_unaligned } XXH_alignment;
 219
 220 FORCE_INLINE U32 XXH_readLE32_align(const void* ptr, XXH_endianess endian, XXH_alignment align)
 221 {
 222     if (align==XXH_unaligned)
 223         return endian==XXH_littleEndian ? XXH_read32(ptr) : XXH_swap32(XXH_read32(ptr));
 224     else
 225         return endian==XXH_littleEndian ? *(const U32*)ptr : XXH_swap32(*(const U32*)ptr);
 226 }
 227
 228 FORCE_INLINE U32 XXH_readLE32(const void* ptr, XXH_endianess endian)
 229 {
 230     return XXH_readLE32_align(ptr, endian, XXH_unaligned);
 231 }
 232
 233 static U32 XXH_readBE32(const void* ptr)
 234 {
 235     return XXH_CPU_LITTLE_ENDIAN ? XXH_swap32(XXH_read32(ptr)) : XXH_read32(ptr);
 236 }
 237
 238
 239 /* *************************************
 240 *  Macros
 241 ***************************************/
 242 #define XXH_STATIC_ASSERT(c)   { enum { XXH_static_assert = 1/(int)(!!(c)) }; }    /* use only *after* variable declarations */
 243 XXH_PUBLIC_API unsigned XXH_versionNumber (void) { return XXH_VERSION_NUMBER; }
 244
 245
 246 /* *******************************************************************
 247 *  32-bits hash functions
 248 *********************************************************************/
 249 static const U32 PRIME32_1 = 2654435761U;
 250 static const U32 PRIME32_2 = 2246822519U;
 251 static const U32 PRIME32_3 = 3266489917U;
 252 static const U32 PRIME32_4 =  668265263U;
 253 static const U32 PRIME32_5 =  374761393U;
 254
 255 static U32 XXH32_round(U32 seed, U32 input)
 256 {
 257     seed += input * PRIME32_2;
 258     seed  = XXH_rotl32(seed, 13);
 259     seed *= PRIME32_1;
 260     return seed;
 261 }
 262
 263 FORCE_INLINE U32 XXH32_endian_align(const void* input, size_t len, U32 seed, XXH_endianess endian, XXH_alignment align)
 264 {
 265     const BYTE* p = (const BYTE*)input;
 266     const BYTE* bEnd = p + len;
 267     U32 h32;
 268 #define XXH_get32bits(p) XXH_readLE32_align(p, endian, align)
 269
 270 #ifdef XXH_ACCEPT_NULL_INPUT_POINTER
 271     if (p==NULL) {
 272         len=0;
 273         bEnd=p=(const BYTE*)(size_t)16;
 274     }
 275 #endif
 276
 277     if (len>=16) {
 278         const BYTE* const limit = bEnd - 16;
 279         U32 v1 = seed + PRIME32_1 + PRIME32_2;
 280         U32 v2 = seed + PRIME32_2;
 281         U32 v3 = seed + 0;
 282         U32 v4 = seed - PRIME32_1;
 283
 284         do {
 285             v1 = XXH32_round(v1, XXH_get32bits(p)); p+=4;
 286             v2 = XXH32_round(v2, XXH_get32bits(p)); p+=4;
 287             v3 = XXH32_round(v3, XXH_get32bits(p)); p+=4;
 288             v4 = XXH32_round(v4, XXH_get32bits(p)); p+=4;
 289         } while (p<=limit);
 290
 291         h32 = XXH_rotl32(v1, 1) + XXH_rotl32(v2, 7) + XXH_rotl32(v3, 12) + XXH_rotl32(v4, 18);
 292     } else {
 293         h32  = seed + PRIME32_5;
 294     }
 295
 296     h32 += (U32) len;
 297
 298     while (p+4<=bEnd) {
 299         h32 += XXH_get32bits(p) * PRIME32_3;
 300         h32  = XXH_rotl32(h32, 17) * PRIME32_4 ;
 301         p+=4;
 302     }
 303
 304     while (p<bEnd) {
 305         h32 += (*p) * PRIME32_5;
 306         h32 = XXH_rotl32(h32, 11) * PRIME32_1 ;
 307         p++;
 308     }
 309
 310     h32 ^= h32 >> 15;
 311     h32 *= PRIME32_2;
 312     h32 ^= h32 >> 13;
 313     h32 *= PRIME32_3;
 314     h32 ^= h32 >> 16;
 315
 316     return h32;
 317 }
 318
 319
 320 XXH_PUBLIC_API unsigned int XXH32 (const void* input, size_t len, unsigned int seed)
 321 {
 322 #if 0
 323     /* Simple version, good for code maintenance, but unfortunately slow for small inputs */
 324     XXH32_state_t state;
 325     XXH32_reset(&state, seed);
 326     XXH32_update(&state, input, len);
 327     return XXH32_digest(&state);
 328 #else
 329     XXH_endianess endian_detected = (XXH_endianess)XXH_CPU_LITTLE_ENDIAN;
 330
 331     if (XXH_FORCE_ALIGN_CHECK) {
 332         if ((((size_t)input) & 3) == 0) {   /* Input is 4-bytes aligned, leverage the speed benefit */
 333             if ((endian_detected==XXH_littleEndian) || XXH_FORCE_NATIVE_FORMAT)
 334                 return XXH32_endian_align(input, len, seed, XXH_littleEndian, XXH_aligned);
 335             else
 336                 return XXH32_endian_align(input, len, seed, XXH_bigEndian, XXH_aligned);
 337     }   }
 338
 339     if ((endian_detected==XXH_littleEndian) || XXH_FORCE_NATIVE_FORMAT)
 340         return XXH32_endian_align(input, len, seed, XXH_littleEndian, XXH_unaligned);
 341     else
 342         return XXH32_endian_align(input, len, seed, XXH_bigEndian, XXH_unaligned);
 343 #endif
 344 }
 345
 346
 347
 348 /*======   Hash streaming   ======*/
 349
 350 XXH_PUBLIC_API XXH32_state_t* XXH32_createState(void)
 351 {
 352     return (XXH32_state_t*)XXH_malloc(sizeof(XXH32_state_t));
 353 }
 354 XXH_PUBLIC_API XXH_errorcode XXH32_freeState(XXH32_state_t* statePtr)
 355 {
 356     XXH_free(statePtr);
 357     return XXH_OK;
 358 }
 359
 360 XXH_PUBLIC_API void XXH32_copyState(XXH32_state_t* dstState, const XXH32_state_t* srcState)
 361 {
 362     memcpy(dstState, srcState, sizeof(*dstState));
 363 }
 364
 365 XXH_PUBLIC_API XXH_errorcode XXH32_reset(XXH32_state_t* statePtr, unsigned int seed)
 366 {
 367     XXH32_state_t state;   /* using a local state to memcpy() in order to avoid strict-aliasing warnings */
 368     memset(&state, 0, sizeof(state)-4);   /* do not write into reserved, for future removal */
 369     state.v1 = seed + PRIME32_1 + PRIME32_2;
 370     state.v2 = seed + PRIME32_2;
 371     state.v3 = seed + 0;
 372     state.v4 = seed - PRIME32_1;
 373     memcpy(statePtr, &state, sizeof(state));
 374     return XXH_OK;
 375 }
 376
 377
 378 FORCE_INLINE XXH_errorcode XXH32_update_endian (XXH32_state_t* state, const void* input, size_t len, XXH_endianess endian)
 379 {
 380     const BYTE* p = (const BYTE*)input;
 381     const BYTE* const bEnd = p + len;
 382
 383 #ifdef XXH_ACCEPT_NULL_INPUT_POINTER
 384     if (input==NULL) return XXH_ERROR;
 385 #endif
 386
 387     state->total_len_32 += (unsigned)len;
 388     state->large_len |= (len>=16) | (state->total_len_32>=16);
 389
 390     if (state->memsize + len < 16)  {   /* fill in tmp buffer */
 391         XXH_memcpy((BYTE*)(state->mem32) + state->memsize, input, len);
 392         state->memsize += (unsigned)len;
 393         return XXH_OK;
 394     }
 395
 396     if (state->memsize) {   /* some data left from previous update */
 397         XXH_memcpy((BYTE*)(state->mem32) + state->memsize, input, 16-state->memsize);
 398         {   const U32* p32 = state->mem32;
 399             state->v1 = XXH32_round(state->v1, XXH_readLE32(p32, endian)); p32++;
 400             state->v2 = XXH32_round(state->v2, XXH_readLE32(p32, endian)); p32++;
 401             state->v3 = XXH32_round(state->v3, XXH_readLE32(p32, endian)); p32++;
 402             state->v4 = XXH32_round(state->v4, XXH_readLE32(p32, endian));
 403         }
 404         p += 16-state->memsize;
 405         state->memsize = 0;
 406     }
 407
 408     if (p <= bEnd-16) {
 409         const BYTE* const limit = bEnd - 16;
 410         U32 v1 = state->v1;
 411         U32 v2 = state->v2;
 412         U32 v3 = state->v3;
 413         U32 v4 = state->v4;
 414
 415         do {
 416             v1 = XXH32_round(v1, XXH_readLE32(p, endian)); p+=4;
 417             v2 = XXH32_round(v2, XXH_readLE32(p, endian)); p+=4;
 418             v3 = XXH32_round(v3, XXH_readLE32(p, endian)); p+=4;
 419             v4 = XXH32_round(v4, XXH_readLE32(p, endian)); p+=4;
 420         } while (p<=limit);
 421
 422         state->v1 = v1;
 423         state->v2 = v2;
 424         state->v3 = v3;
 425         state->v4 = v4;
 426     }
 427
 428     if (p < bEnd) {
 429         XXH_memcpy(state->mem32, p, (size_t)(bEnd-p));
 430         state->memsize = (unsigned)(bEnd-p);
 431     }
 432
 433     return XXH_OK;
 434 }
 435
 436 XXH_PUBLIC_API XXH_errorcode XXH32_update (XXH32_state_t* state_in, const void* input, size_t len)
 437 {
 438     XXH_endianess endian_detected = (XXH_endianess)XXH_CPU_LITTLE_ENDIAN;
 439
 440     if ((endian_detected==XXH_littleEndian) || XXH_FORCE_NATIVE_FORMAT)
 441         return XXH32_update_endian(state_in, input, len, XXH_littleEndian);
 442     else
 443         return XXH32_update_endian(state_in, input, len, XXH_bigEndian);
 444 }
 445
 446
 447
 448 FORCE_INLINE U32 XXH32_digest_endian (const XXH32_state_t* state, XXH_endianess endian)
 449 {
 450     const BYTE * p = (const BYTE*)state->mem32;
 451     const BYTE* const bEnd = (const BYTE*)(state->mem32) + state->memsize;
 452     U32 h32;
 453
 454     if (state->large_len) {
 455         h32 = XXH_rotl32(state->v1, 1) + XXH_rotl32(state->v2, 7) + XXH_rotl32(state->v3, 12) + XXH_rotl32(state->v4, 18);
 456     } else {
 457         h32 = state->v3 /* == seed */ + PRIME32_5;
 458     }
 459
 460     h32 += state->total_len_32;
 461
 462     while (p+4<=bEnd) {
 463         h32 += XXH_readLE32(p, endian) * PRIME32_3;
 464         h32  = XXH_rotl32(h32, 17) * PRIME32_4;
 465         p+=4;
 466     }
 467
 468     while (p<bEnd) {
 469         h32 += (*p) * PRIME32_5;
 470         h32  = XXH_rotl32(h32, 11) * PRIME32_1;
 471         p++;
 472     }
 473
 474     h32 ^= h32 >> 15;
 475     h32 *= PRIME32_2;
 476     h32 ^= h32 >> 13;
 477     h32 *= PRIME32_3;
 478     h32 ^= h32 >> 16;
 479
 480     return h32;
 481 }
 482
 483
 484 XXH_PUBLIC_API unsigned int XXH32_digest (const XXH32_state_t* state_in)
 485 {
 486     XXH_endianess endian_detected = (XXH_endianess)XXH_CPU_LITTLE_ENDIAN;
 487
 488     if ((endian_detected==XXH_littleEndian) || XXH_FORCE_NATIVE_FORMAT)
 489         return XXH32_digest_endian(state_in, XXH_littleEndian);
 490     else
 491         return XXH32_digest_endian(state_in, XXH_bigEndian);
 492 }
 493
 494
 495 /*======   Canonical representation   ======*/
 496
 497 /*! Default XXH result types are basic unsigned 32 and 64 bits.
 498 *   The canonical representation follows human-readable write convention, aka big-endian (large digits first).
 499 *   These functions allow transformation of hash result into and from its canonical format.
 500 *   This way, hash values can be written into a file or buffer, and remain comparable across different systems and programs.
 501 */
 502
 503 XXH_PUBLIC_API void XXH32_canonicalFromHash(XXH32_canonical_t* dst, XXH32_hash_t hash)
 504 {
 505     XXH_STATIC_ASSERT(sizeof(XXH32_canonical_t) == sizeof(XXH32_hash_t));
 506     if (XXH_CPU_LITTLE_ENDIAN) hash = XXH_swap32(hash);
 507     memcpy(dst, &hash, sizeof(*dst));
 508 }
 509
 510 XXH_PUBLIC_API XXH32_hash_t XXH32_hashFromCanonical(const XXH32_canonical_t* src)
 511 {
 512     return XXH_readBE32(src);
 513 }
 514
 515
 516 #ifndef XXH_NO_LONG_LONG
 517
 518 /* *******************************************************************
 519 *  64-bits hash functions
 520 *********************************************************************/
 521
 522 /*======   Memory access   ======*/
 523
 524 #ifndef MEM_MODULE
 525 # define MEM_MODULE
 526 # if !defined (__VMS) && (defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) )
 527 #   include <stdint.h>
 528     typedef uint64_t U64;
 529 # else
 530     typedef unsigned long long U64;   /* if your compiler doesn't support unsigned long long, replace by another 64-bit type here. Note that xxhash.h will also need to be updated. */
 531 # endif
 532 #endif
 533
 534
 535 #if (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==2))
 536
 537 /* Force direct memory access. Only works on CPU which support unaligned memory access in hardware */
 538 static U64 XXH_read64(const void* memPtr) { return *(const U64*) memPtr; }
 539
 540 #elif (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==1))
 541
 542 /* __pack instructions are safer, but compiler specific, hence potentially problematic for some compilers */
 543 /* currently only defined for gcc and icc */
 544 typedef union { U32 u32; U64 u64; } __attribute__((packed)) unalign64;
 545 static U64 XXH_read64(const void* ptr) { return ((const unalign64*)ptr)->u64; }
 546
 547 #else
 548
 549 /* portable and safe solution. Generally efficient.
 550  * see : http://stackoverflow.com/a/32095106/646947
 551  */
 552
 553 static U64 XXH_read64(const void* memPtr)
 554 {
 555     U64 val;
 556     memcpy(&val, memPtr, sizeof(val));
 557     return val;
 558 }
 559
 560 #endif   /* XXH_FORCE_DIRECT_MEMORY_ACCESS */
 561
 562 #if defined(_MSC_VER)     /* Visual Studio */
 563 #  define XXH_swap64 _byteswap_uint64
 564 #elif XXH_GCC_VERSION >= 403
 565 #  define XXH_swap64 __builtin_bswap64
 566 #else
 567 static U64 XXH_swap64 (U64 x)
 568 {
 569     return  ((x << 56) & 0xff00000000000000ULL) |
 570             ((x << 40) & 0x00ff000000000000ULL) |
 571             ((x << 24) & 0x0000ff0000000000ULL) |
 572             ((x << 8)  & 0x000000ff00000000ULL) |
 573             ((x >> 8)  & 0x00000000ff000000ULL) |
 574             ((x >> 24) & 0x0000000000ff0000ULL) |
 575             ((x >> 40) & 0x000000000000ff00ULL) |
 576             ((x >> 56) & 0x00000000000000ffULL);
 577 }
 578 #endif
 579
 580 FORCE_INLINE U64 XXH_readLE64_align(const void* ptr, XXH_endianess endian, XXH_alignment align)
 581 {
 582     if (align==XXH_unaligned)
 583         return endian==XXH_littleEndian ? XXH_read64(ptr) : XXH_swap64(XXH_read64(ptr));
 584     else
 585         return endian==XXH_littleEndian ? *(const U64*)ptr : XXH_swap64(*(const U64*)ptr);
 586 }
 587
 588 FORCE_INLINE U64 XXH_readLE64(const void* ptr, XXH_endianess endian)
 589 {
 590     return XXH_readLE64_align(ptr, endian, XXH_unaligned);
 591 }
 592
 593 static U64 XXH_readBE64(const void* ptr)
 594 {
 595     return XXH_CPU_LITTLE_ENDIAN ? XXH_swap64(XXH_read64(ptr)) : XXH_read64(ptr);
 596 }
 597
 598
 599 /*======   xxh64   ======*/
 600
 601 static const U64 PRIME64_1 = 11400714785074694791ULL;
 602 static const U64 PRIME64_2 = 14029467366897019727ULL;
 603 static const U64 PRIME64_3 =  1609587929392839161ULL;
 604 static const U64 PRIME64_4 =  9650029242287828579ULL;
 605 static const U64 PRIME64_5 =  2870177450012600261ULL;
 606
 607 static U64 XXH64_round(U64 acc, U64 input)
 608 {
 609     acc += input * PRIME64_2;
 610     acc  = XXH_rotl64(acc, 31);
 611     acc *= PRIME64_1;
 612     return acc;
 613 }
 614
 615 static U64 XXH64_mergeRound(U64 acc, U64 val)
 616 {
 617     val  = XXH64_round(0, val);
 618     acc ^= val;
 619     acc  = acc * PRIME64_1 + PRIME64_4;
 620     return acc;
 621 }
 622
 623 FORCE_INLINE U64 XXH64_endian_align(const void* input, size_t len, U64 seed, XXH_endianess endian, XXH_alignment align)
 624 {
 625     const BYTE* p = (const BYTE*)input;
 626     const BYTE* bEnd = p + len;
 627     U64 h64;
 628 #define XXH_get64bits(p) XXH_readLE64_align(p, endian, align)
 629
 630 #ifdef XXH_ACCEPT_NULL_INPUT_POINTER
 631     if (p==NULL) {
 632         len=0;
 633         bEnd=p=(const BYTE*)(size_t)32;
 634     }
 635 #endif
 636
 637     if (len>=32) {
 638         const BYTE* const limit = bEnd - 32;
 639         U64 v1 = seed + PRIME64_1 + PRIME64_2;
 640         U64 v2 = seed + PRIME64_2;
 641         U64 v3 = seed + 0;
 642         U64 v4 = seed - PRIME64_1;
 643
 644         do {
 645             v1 = XXH64_round(v1, XXH_get64bits(p)); p+=8;
 646             v2 = XXH64_round(v2, XXH_get64bits(p)); p+=8;
 647             v3 = XXH64_round(v3, XXH_get64bits(p)); p+=8;
 648             v4 = XXH64_round(v4, XXH_get64bits(p)); p+=8;
 649         } while (p<=limit);
 650
 651         h64 = XXH_rotl64(v1, 1) + XXH_rotl64(v2, 7) + XXH_rotl64(v3, 12) + XXH_rotl64(v4, 18);
 652         h64 = XXH64_mergeRound(h64, v1);
 653         h64 = XXH64_mergeRound(h64, v2);
 654         h64 = XXH64_mergeRound(h64, v3);
 655         h64 = XXH64_mergeRound(h64, v4);
 656
 657     } else {
 658         h64  = seed + PRIME64_5;
 659     }
 660
 661     h64 += (U64) len;
 662
 663     while (p+8<=bEnd) {
 664         U64 const k1 = XXH64_round(0, XXH_get64bits(p));
 665         h64 ^= k1;
 666         h64  = XXH_rotl64(h64,27) * PRIME64_1 + PRIME64_4;
 667         p+=8;
 668     }
 669
 670     if (p+4<=bEnd) {
 671         h64 ^= (U64)(XXH_get32bits(p)) * PRIME64_1;
 672         h64 = XXH_rotl64(h64, 23) * PRIME64_2 + PRIME64_3;
 673         p+=4;
 674     }
 675
 676     while (p<bEnd) {
 677         h64 ^= (*p) * PRIME64_5;
 678         h64 = XXH_rotl64(h64, 11) * PRIME64_1;
 679         p++;
 680     }
 681
 682     h64 ^= h64 >> 33;
 683     h64 *= PRIME64_2;
 684     h64 ^= h64 >> 29;
 685     h64 *= PRIME64_3;
 686     h64 ^= h64 >> 32;
 687
 688     return h64;
 689 }
 690
 691
 692 XXH_PUBLIC_API unsigned long long XXH64 (const void* input, size_t len, unsigned long long seed)
 693 {
 694 #if 0
 695     /* Simple version, good for code maintenance, but unfortunately slow for small inputs */
 696     XXH64_state_t state;
 697     XXH64_reset(&state, seed);
 698     XXH64_update(&state, input, len);
 699     return XXH64_digest(&state);
 700 #else
 701     XXH_endianess endian_detected = (XXH_endianess)XXH_CPU_LITTLE_ENDIAN;
 702
 703     if (XXH_FORCE_ALIGN_CHECK) {
 704         if ((((size_t)input) & 7)==0) {  /* Input is aligned, let's leverage the speed advantage */
 705             if ((endian_detected==XXH_littleEndian) || XXH_FORCE_NATIVE_FORMAT)
 706                 return XXH64_endian_align(input, len, seed, XXH_littleEndian, XXH_aligned);
 707             else
 708                 return XXH64_endian_align(input, len, seed, XXH_bigEndian, XXH_aligned);
 709     }   }
 710
 711     if ((endian_detected==XXH_littleEndian) || XXH_FORCE_NATIVE_FORMAT)
 712         return XXH64_endian_align(input, len, seed, XXH_littleEndian, XXH_unaligned);
 713     else
 714         return XXH64_endian_align(input, len, seed, XXH_bigEndian, XXH_unaligned);
 715 #endif
 716 }
 717
 718 /*======   Hash Streaming   ======*/
 719
 720 XXH_PUBLIC_API XXH64_state_t* XXH64_createState(void)
 721 {
 722     return (XXH64_state_t*)XXH_malloc(sizeof(XXH64_state_t));
 723 }
 724 XXH_PUBLIC_API XXH_errorcode XXH64_freeState(XXH64_state_t* statePtr)
 725 {
 726     XXH_free(statePtr);
 727     return XXH_OK;
 728 }
 729
 730 XXH_PUBLIC_API void XXH64_copyState(XXH64_state_t* dstState, const XXH64_state_t* srcState)
 731 {
 732     memcpy(dstState, srcState, sizeof(*dstState));
 733 }
 734
 735 XXH_PUBLIC_API XXH_errorcode XXH64_reset(XXH64_state_t* statePtr, unsigned long long seed)
 736 {
 737     XXH64_state_t state;   /* using a local state to memcpy() in order to avoid strict-aliasing warnings */
 738     memset(&state, 0, sizeof(state)-8);   /* do not write into reserved, for future removal */
 739     state.v1 = seed + PRIME64_1 + PRIME64_2;
 740     state.v2 = seed + PRIME64_2;
 741     state.v3 = seed + 0;
 742     state.v4 = seed - PRIME64_1;
 743     memcpy(statePtr, &state, sizeof(state));
 744     return XXH_OK;
 745 }
 746
 747 FORCE_INLINE XXH_errorcode XXH64_update_endian (XXH64_state_t* state, const void* input, size_t len, XXH_endianess endian)
 748 {
 749     const BYTE* p = (const BYTE*)input;
 750     const BYTE* const bEnd = p + len;
 751
 752 #ifdef XXH_ACCEPT_NULL_INPUT_POINTER
 753     if (input==NULL) return XXH_ERROR;
 754 #endif
 755
 756     state->total_len += len;
 757
 758     if (state->memsize + len < 32) {  /* fill in tmp buffer */
 759         XXH_memcpy(((BYTE*)state->mem64) + state->memsize, input, len);
 760         state->memsize += (U32)len;
 761         return XXH_OK;
 762     }
 763
 764     if (state->memsize) {   /* tmp buffer is full */
 765         XXH_memcpy(((BYTE*)state->mem64) + state->memsize, input, 32-state->memsize);
 766         state->v1 = XXH64_round(state->v1, XXH_readLE64(state->mem64+0, endian));
 767         state->v2 = XXH64_round(state->v2, XXH_readLE64(state->mem64+1, endian));
 768         state->v3 = XXH64_round(state->v3, XXH_readLE64(state->mem64+2, endian));
 769         state->v4 = XXH64_round(state->v4, XXH_readLE64(state->mem64+3, endian));
 770         p += 32-state->memsize;
 771         state->memsize = 0;
 772     }
 773
 774     if (p+32 <= bEnd) {
 775         const BYTE* const limit = bEnd - 32;
 776         U64 v1 = state->v1;
 777         U64 v2 = state->v2;
 778         U64 v3 = state->v3;
 779         U64 v4 = state->v4;
 780
 781         do {
 782             v1 = XXH64_round(v1, XXH_readLE64(p, endian)); p+=8;
 783             v2 = XXH64_round(v2, XXH_readLE64(p, endian)); p+=8;
 784             v3 = XXH64_round(v3, XXH_readLE64(p, endian)); p+=8;
 785             v4 = XXH64_round(v4, XXH_readLE64(p, endian)); p+=8;
 786         } while (p<=limit);
 787
 788         state->v1 = v1;
 789         state->v2 = v2;
 790         state->v3 = v3;
 791         state->v4 = v4;
 792     }
 793
 794     if (p < bEnd) {
 795         XXH_memcpy(state->mem64, p, (size_t)(bEnd-p));
 796         state->memsize = (unsigned)(bEnd-p);
 797     }
 798
 799     return XXH_OK;
 800 }
 801
 802 XXH_PUBLIC_API XXH_errorcode XXH64_update (XXH64_state_t* state_in, const void* input, size_t len)
 803 {
 804     XXH_endianess endian_detected = (XXH_endianess)XXH_CPU_LITTLE_ENDIAN;
 805
 806     if ((endian_detected==XXH_littleEndian) || XXH_FORCE_NATIVE_FORMAT)
 807         return XXH64_update_endian(state_in, input, len, XXH_littleEndian);
 808     else
 809         return XXH64_update_endian(state_in, input, len, XXH_bigEndian);
 810 }
 811
 812 FORCE_INLINE U64 XXH64_digest_endian (const XXH64_state_t* state, XXH_endianess endian)
 813 {
 814     const BYTE * p = (const BYTE*)state->mem64;
 815     const BYTE* const bEnd = (const BYTE*)state->mem64 + state->memsize;
 816     U64 h64;
 817
 818     if (state->total_len >= 32) {
 819         U64 const v1 = state->v1;
 820         U64 const v2 = state->v2;
 821         U64 const v3 = state->v3;
 822         U64 const v4 = state->v4;
 823
 824         h64 = XXH_rotl64(v1, 1) + XXH_rotl64(v2, 7) + XXH_rotl64(v3, 12) + XXH_rotl64(v4, 18);
 825         h64 = XXH64_mergeRound(h64, v1);
 826         h64 = XXH64_mergeRound(h64, v2);
 827         h64 = XXH64_mergeRound(h64, v3);
 828         h64 = XXH64_mergeRound(h64, v4);
 829     } else {
 830         h64  = state->v3 + PRIME64_5;
 831     }
 832
 833     h64 += (U64) state->total_len;
 834
 835     while (p+8<=bEnd) {
 836         U64 const k1 = XXH64_round(0, XXH_readLE64(p, endian));
 837         h64 ^= k1;
 838         h64  = XXH_rotl64(h64,27) * PRIME64_1 + PRIME64_4;
 839         p+=8;
 840     }
 841
 842     if (p+4<=bEnd) {
 843         h64 ^= (U64)(XXH_readLE32(p, endian)) * PRIME64_1;
 844         h64  = XXH_rotl64(h64, 23) * PRIME64_2 + PRIME64_3;
 845         p+=4;
 846     }
 847
 848     while (p<bEnd) {
 849         h64 ^= (*p) * PRIME64_5;
 850         h64  = XXH_rotl64(h64, 11) * PRIME64_1;
 851         p++;
 852     }
 853
 854     h64 ^= h64 >> 33;
 855     h64 *= PRIME64_2;
 856     h64 ^= h64 >> 29;
 857     h64 *= PRIME64_3;
 858     h64 ^= h64 >> 32;
 859
 860     return h64;
 861 }
 862
 863 XXH_PUBLIC_API unsigned long long XXH64_digest (const XXH64_state_t* state_in)
 864 {
 865     XXH_endianess endian_detected = (XXH_endianess)XXH_CPU_LITTLE_ENDIAN;
 866
 867     if ((endian_detected==XXH_littleEndian) || XXH_FORCE_NATIVE_FORMAT)
 868         return XXH64_digest_endian(state_in, XXH_littleEndian);
 869     else
 870         return XXH64_digest_endian(state_in, XXH_bigEndian);
 871 }
 872
 873
 874 /*====== Canonical representation   ======*/
 875
 876 XXH_PUBLIC_API void XXH64_canonicalFromHash(XXH64_canonical_t* dst, XXH64_hash_t hash)
 877 {
 878     XXH_STATIC_ASSERT(sizeof(XXH64_canonical_t) == sizeof(XXH64_hash_t));
 879     if (XXH_CPU_LITTLE_ENDIAN) hash = XXH_swap64(hash);
 880     memcpy(dst, &hash, sizeof(*dst));
 881 }
 882
 883 XXH_PUBLIC_API XXH64_hash_t XXH64_hashFromCanonical(const XXH64_canonical_t* src)
 884 {
 885     return XXH_readBE64(src);
 886 }
 887
 888 #endif  /* XXH_NO_LONG_LONG */