ceph/src/zstd/lib/common/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) && !defined(WIN32)) || \
  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-independant 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-independance 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; set to 0 when the input data
  84  * is guaranteed to be aligned.
  85  */
  86 #ifndef XXH_FORCE_ALIGN_CHECK /* can be defined externally */
  87 #  if defined(__i386) || defined(_M_IX86) || defined(__x86_64__) || defined(_M_X64)
  88 #    define XXH_FORCE_ALIGN_CHECK 0
  89 #  else
  90 #    define XXH_FORCE_ALIGN_CHECK 1
  91 #  endif
  92 #endif
  93
  94
  95 /* *************************************
  96 *  Includes & Memory related functions
  97 ***************************************/
  98 /* Modify the local functions below should you wish to use some other memory routines */
  99 /* for malloc(), free() */
 100 #include <stdlib.h>
 101 static void* XXH_malloc(size_t s) { return malloc(s); }
 102 static void  XXH_free  (void* p)  { free(p); }
 103 /* for memcpy() */
 104 #include <string.h>
 105 static void* XXH_memcpy(void* dest, const void* src, size_t size) { return memcpy(dest,src,size); }
 106
 107 #ifndef XXH_STATIC_LINKING_ONLY
 108 #  define XXH_STATIC_LINKING_ONLY
 109 #endif
 110 #include "xxhash.h"
 111
 112
 113 /* *************************************
 114 *  Compiler Specific Options
 115 ***************************************/
 116 #if defined (__GNUC__) || defined(__cplusplus) || defined(__STDC_VERSION__) && __STDC_VERSION__ >= 199901L   /* C99 */
 117 #  define INLINE_KEYWORD inline
 118 #else
 119 #  define INLINE_KEYWORD
 120 #endif
 121
 122 #if defined(__GNUC__)
 123 #  define FORCE_INLINE_ATTR __attribute__((always_inline))
 124 #elif defined(_MSC_VER)
 125 #  define FORCE_INLINE_ATTR __forceinline
 126 #else
 127 #  define FORCE_INLINE_ATTR
 128 #endif
 129
 130 #define FORCE_INLINE_TEMPLATE static INLINE_KEYWORD FORCE_INLINE_ATTR
 131
 132
 133 #ifdef _MSC_VER
 134 #  pragma warning(disable : 4127)      /* disable: C4127: conditional expression is constant */
 135 #endif
 136
 137
 138 /* *************************************
 139 *  Basic Types
 140 ***************************************/
 141 #ifndef MEM_MODULE
 142 # define MEM_MODULE
 143 # if !defined (__VMS) && (defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) )
 144 #   include <stdint.h>
 145     typedef uint8_t  BYTE;
 146     typedef uint16_t U16;
 147     typedef uint32_t U32;
 148     typedef  int32_t S32;
 149     typedef uint64_t U64;
 150 #  else
 151     typedef unsigned char      BYTE;
 152     typedef unsigned short     U16;
 153     typedef unsigned int       U32;
 154     typedef   signed int       S32;
 155     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. */
 156 #  endif
 157 #endif
 158
 159
 160 #if (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==2))
 161
 162 /* Force direct memory access. Only works on CPU which support unaligned memory access in hardware */
 163 static U32 XXH_read32(const void* memPtr) { return *(const U32*) memPtr; }
 164 static U64 XXH_read64(const void* memPtr) { return *(const U64*) memPtr; }
 165
 166 #elif (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==1))
 167
 168 /* __pack instructions are safer, but compiler specific, hence potentially problematic for some compilers */
 169 /* currently only defined for gcc and icc */
 170 typedef union { U32 u32; U64 u64; } __attribute__((packed)) unalign;
 171
 172 static U32 XXH_read32(const void* ptr) { return ((const unalign*)ptr)->u32; }
 173 static U64 XXH_read64(const void* ptr) { return ((const unalign*)ptr)->u64; }
 174
 175 #else
 176
 177 /* portable and safe solution. Generally efficient.
 178  * see : http://stackoverflow.com/a/32095106/646947
 179  */
 180
 181 static U32 XXH_read32(const void* memPtr)
 182 {
 183     U32 val;
 184     memcpy(&val, memPtr, sizeof(val));
 185     return val;
 186 }
 187
 188 static U64 XXH_read64(const void* memPtr)
 189 {
 190     U64 val;
 191     memcpy(&val, memPtr, sizeof(val));
 192     return val;
 193 }
 194
 195 #endif   /* XXH_FORCE_DIRECT_MEMORY_ACCESS */
 196
 197
 198 /* ****************************************
 199 *  Compiler-specific Functions and Macros
 200 ******************************************/
 201 #define GCC_VERSION (__GNUC__ * 100 + __GNUC_MINOR__)
 202
 203 /* Note : although _rotl exists for minGW (GCC under windows), performance seems poor */
 204 #if defined(_MSC_VER)
 205 #  define XXH_rotl32(x,r) _rotl(x,r)
 206 #  define XXH_rotl64(x,r) _rotl64(x,r)
 207 #else
 208 #  define XXH_rotl32(x,r) ((x << r) | (x >> (32 - r)))
 209 #  define XXH_rotl64(x,r) ((x << r) | (x >> (64 - r)))
 210 #endif
 211
 212 #if defined(_MSC_VER)     /* Visual Studio */
 213 #  define XXH_swap32 _byteswap_ulong
 214 #  define XXH_swap64 _byteswap_uint64
 215 #elif GCC_VERSION >= 403
 216 #  define XXH_swap32 __builtin_bswap32
 217 #  define XXH_swap64 __builtin_bswap64
 218 #else
 219 static U32 XXH_swap32 (U32 x)
 220 {
 221     return  ((x << 24) & 0xff000000 ) |
 222             ((x <<  8) & 0x00ff0000 ) |
 223             ((x >>  8) & 0x0000ff00 ) |
 224             ((x >> 24) & 0x000000ff );
 225 }
 226 static U64 XXH_swap64 (U64 x)
 227 {
 228     return  ((x << 56) & 0xff00000000000000ULL) |
 229             ((x << 40) & 0x00ff000000000000ULL) |
 230             ((x << 24) & 0x0000ff0000000000ULL) |
 231             ((x << 8)  & 0x000000ff00000000ULL) |
 232             ((x >> 8)  & 0x00000000ff000000ULL) |
 233             ((x >> 24) & 0x0000000000ff0000ULL) |
 234             ((x >> 40) & 0x000000000000ff00ULL) |
 235             ((x >> 56) & 0x00000000000000ffULL);
 236 }
 237 #endif
 238
 239
 240 /* *************************************
 241 *  Architecture Macros
 242 ***************************************/
 243 typedef enum { XXH_bigEndian=0, XXH_littleEndian=1 } XXH_endianess;
 244
 245 /* XXH_CPU_LITTLE_ENDIAN can be defined externally, for example on the compiler command line */
 246 #ifndef XXH_CPU_LITTLE_ENDIAN
 247     static const int g_one = 1;
 248 #   define XXH_CPU_LITTLE_ENDIAN   (*(const char*)(&g_one))
 249 #endif
 250
 251
 252 /* ***************************
 253 *  Memory reads
 254 *****************************/
 255 typedef enum { XXH_aligned, XXH_unaligned } XXH_alignment;
 256
 257 FORCE_INLINE_TEMPLATE U32 XXH_readLE32_align(const void* ptr, XXH_endianess endian, XXH_alignment align)
 258 {
 259     if (align==XXH_unaligned)
 260         return endian==XXH_littleEndian ? XXH_read32(ptr) : XXH_swap32(XXH_read32(ptr));
 261     else
 262         return endian==XXH_littleEndian ? *(const U32*)ptr : XXH_swap32(*(const U32*)ptr);
 263 }
 264
 265 FORCE_INLINE_TEMPLATE U32 XXH_readLE32(const void* ptr, XXH_endianess endian)
 266 {
 267     return XXH_readLE32_align(ptr, endian, XXH_unaligned);
 268 }
 269
 270 static U32 XXH_readBE32(const void* ptr)
 271 {
 272     return XXH_CPU_LITTLE_ENDIAN ? XXH_swap32(XXH_read32(ptr)) : XXH_read32(ptr);
 273 }
 274
 275 FORCE_INLINE_TEMPLATE U64 XXH_readLE64_align(const void* ptr, XXH_endianess endian, XXH_alignment align)
 276 {
 277     if (align==XXH_unaligned)
 278         return endian==XXH_littleEndian ? XXH_read64(ptr) : XXH_swap64(XXH_read64(ptr));
 279     else
 280         return endian==XXH_littleEndian ? *(const U64*)ptr : XXH_swap64(*(const U64*)ptr);
 281 }
 282
 283 FORCE_INLINE_TEMPLATE U64 XXH_readLE64(const void* ptr, XXH_endianess endian)
 284 {
 285     return XXH_readLE64_align(ptr, endian, XXH_unaligned);
 286 }
 287
 288 static U64 XXH_readBE64(const void* ptr)
 289 {
 290     return XXH_CPU_LITTLE_ENDIAN ? XXH_swap64(XXH_read64(ptr)) : XXH_read64(ptr);
 291 }
 292
 293
 294 /* *************************************
 295 *  Macros
 296 ***************************************/
 297 #define XXH_STATIC_ASSERT(c)   { enum { XXH_static_assert = 1/(int)(!!(c)) }; }    /* use only *after* variable declarations */
 298
 299
 300 /* *************************************
 301 *  Constants
 302 ***************************************/
 303 static const U32 PRIME32_1 = 2654435761U;
 304 static const U32 PRIME32_2 = 2246822519U;
 305 static const U32 PRIME32_3 = 3266489917U;
 306 static const U32 PRIME32_4 =  668265263U;
 307 static const U32 PRIME32_5 =  374761393U;
 308
 309 static const U64 PRIME64_1 = 11400714785074694791ULL;
 310 static const U64 PRIME64_2 = 14029467366897019727ULL;
 311 static const U64 PRIME64_3 =  1609587929392839161ULL;
 312 static const U64 PRIME64_4 =  9650029242287828579ULL;
 313 static const U64 PRIME64_5 =  2870177450012600261ULL;
 314
 315 XXH_PUBLIC_API unsigned XXH_versionNumber (void) { return XXH_VERSION_NUMBER; }
 316
 317
 318 /* **************************
 319 *  Utils
 320 ****************************/
 321 XXH_PUBLIC_API void XXH32_copyState(XXH32_state_t* restrict dstState, const XXH32_state_t* restrict srcState)
 322 {
 323     memcpy(dstState, srcState, sizeof(*dstState));
 324 }
 325
 326 XXH_PUBLIC_API void XXH64_copyState(XXH64_state_t* restrict dstState, const XXH64_state_t* restrict srcState)
 327 {
 328     memcpy(dstState, srcState, sizeof(*dstState));
 329 }
 330
 331
 332 /* ***************************
 333 *  Simple Hash Functions
 334 *****************************/
 335
 336 static U32 XXH32_round(U32 seed, U32 input)
 337 {
 338     seed += input * PRIME32_2;
 339     seed  = XXH_rotl32(seed, 13);
 340     seed *= PRIME32_1;
 341     return seed;
 342 }
 343
 344 FORCE_INLINE_TEMPLATE U32 XXH32_endian_align(const void* input, size_t len, U32 seed, XXH_endianess endian, XXH_alignment align)
 345 {
 346     const BYTE* p = (const BYTE*)input;
 347     const BYTE* bEnd = p + len;
 348     U32 h32;
 349 #define XXH_get32bits(p) XXH_readLE32_align(p, endian, align)
 350
 351 #ifdef XXH_ACCEPT_NULL_INPUT_POINTER
 352     if (p==NULL) {
 353         len=0;
 354         bEnd=p=(const BYTE*)(size_t)16;
 355     }
 356 #endif
 357
 358     if (len>=16) {
 359         const BYTE* const limit = bEnd - 16;
 360         U32 v1 = seed + PRIME32_1 + PRIME32_2;
 361         U32 v2 = seed + PRIME32_2;
 362         U32 v3 = seed + 0;
 363         U32 v4 = seed - PRIME32_1;
 364
 365         do {
 366             v1 = XXH32_round(v1, XXH_get32bits(p)); p+=4;
 367             v2 = XXH32_round(v2, XXH_get32bits(p)); p+=4;
 368             v3 = XXH32_round(v3, XXH_get32bits(p)); p+=4;
 369             v4 = XXH32_round(v4, XXH_get32bits(p)); p+=4;
 370         } while (p<=limit);
 371
 372         h32 = XXH_rotl32(v1, 1) + XXH_rotl32(v2, 7) + XXH_rotl32(v3, 12) + XXH_rotl32(v4, 18);
 373     } else {
 374         h32  = seed + PRIME32_5;
 375     }
 376
 377     h32 += (U32) len;
 378
 379     while (p+4<=bEnd) {
 380         h32 += XXH_get32bits(p) * PRIME32_3;
 381         h32  = XXH_rotl32(h32, 17) * PRIME32_4 ;
 382         p+=4;
 383     }
 384
 385     while (p<bEnd) {
 386         h32 += (*p) * PRIME32_5;
 387         h32 = XXH_rotl32(h32, 11) * PRIME32_1 ;
 388         p++;
 389     }
 390
 391     h32 ^= h32 >> 15;
 392     h32 *= PRIME32_2;
 393     h32 ^= h32 >> 13;
 394     h32 *= PRIME32_3;
 395     h32 ^= h32 >> 16;
 396
 397     return h32;
 398 }
 399
 400
 401 XXH_PUBLIC_API unsigned int XXH32 (const void* input, size_t len, unsigned int seed)
 402 {
 403 #if 0
 404     /* Simple version, good for code maintenance, but unfortunately slow for small inputs */
 405     XXH32_CREATESTATE_STATIC(state);
 406     XXH32_reset(state, seed);
 407     XXH32_update(state, input, len);
 408     return XXH32_digest(state);
 409 #else
 410     XXH_endianess endian_detected = (XXH_endianess)XXH_CPU_LITTLE_ENDIAN;
 411
 412     if (XXH_FORCE_ALIGN_CHECK) {
 413         if ((((size_t)input) & 3) == 0) {   /* Input is 4-bytes aligned, leverage the speed benefit */
 414             if ((endian_detected==XXH_littleEndian) || XXH_FORCE_NATIVE_FORMAT)
 415                 return XXH32_endian_align(input, len, seed, XXH_littleEndian, XXH_aligned);
 416             else
 417                 return XXH32_endian_align(input, len, seed, XXH_bigEndian, XXH_aligned);
 418     }   }
 419
 420     if ((endian_detected==XXH_littleEndian) || XXH_FORCE_NATIVE_FORMAT)
 421         return XXH32_endian_align(input, len, seed, XXH_littleEndian, XXH_unaligned);
 422     else
 423         return XXH32_endian_align(input, len, seed, XXH_bigEndian, XXH_unaligned);
 424 #endif
 425 }
 426
 427
 428 static U64 XXH64_round(U64 acc, U64 input)
 429 {
 430     acc += input * PRIME64_2;
 431     acc  = XXH_rotl64(acc, 31);
 432     acc *= PRIME64_1;
 433     return acc;
 434 }
 435
 436 static U64 XXH64_mergeRound(U64 acc, U64 val)
 437 {
 438     val  = XXH64_round(0, val);
 439     acc ^= val;
 440     acc  = acc * PRIME64_1 + PRIME64_4;
 441     return acc;
 442 }
 443
 444 FORCE_INLINE_TEMPLATE U64 XXH64_endian_align(const void* input, size_t len, U64 seed, XXH_endianess endian, XXH_alignment align)
 445 {
 446     const BYTE* p = (const BYTE*)input;
 447     const BYTE* const bEnd = p + len;
 448     U64 h64;
 449 #define XXH_get64bits(p) XXH_readLE64_align(p, endian, align)
 450
 451 #ifdef XXH_ACCEPT_NULL_INPUT_POINTER
 452     if (p==NULL) {
 453         len=0;
 454         bEnd=p=(const BYTE*)(size_t)32;
 455     }
 456 #endif
 457
 458     if (len>=32) {
 459         const BYTE* const limit = bEnd - 32;
 460         U64 v1 = seed + PRIME64_1 + PRIME64_2;
 461         U64 v2 = seed + PRIME64_2;
 462         U64 v3 = seed + 0;
 463         U64 v4 = seed - PRIME64_1;
 464
 465         do {
 466             v1 = XXH64_round(v1, XXH_get64bits(p)); p+=8;
 467             v2 = XXH64_round(v2, XXH_get64bits(p)); p+=8;
 468             v3 = XXH64_round(v3, XXH_get64bits(p)); p+=8;
 469             v4 = XXH64_round(v4, XXH_get64bits(p)); p+=8;
 470         } while (p<=limit);
 471
 472         h64 = XXH_rotl64(v1, 1) + XXH_rotl64(v2, 7) + XXH_rotl64(v3, 12) + XXH_rotl64(v4, 18);
 473         h64 = XXH64_mergeRound(h64, v1);
 474         h64 = XXH64_mergeRound(h64, v2);
 475         h64 = XXH64_mergeRound(h64, v3);
 476         h64 = XXH64_mergeRound(h64, v4);
 477
 478     } else {
 479         h64  = seed + PRIME64_5;
 480     }
 481
 482     h64 += (U64) len;
 483
 484     while (p+8<=bEnd) {
 485         U64 const k1 = XXH64_round(0, XXH_get64bits(p));
 486         h64 ^= k1;
 487         h64  = XXH_rotl64(h64,27) * PRIME64_1 + PRIME64_4;
 488         p+=8;
 489     }
 490
 491     if (p+4<=bEnd) {
 492         h64 ^= (U64)(XXH_get32bits(p)) * PRIME64_1;
 493         h64 = XXH_rotl64(h64, 23) * PRIME64_2 + PRIME64_3;
 494         p+=4;
 495     }
 496
 497     while (p<bEnd) {
 498         h64 ^= (*p) * PRIME64_5;
 499         h64 = XXH_rotl64(h64, 11) * PRIME64_1;
 500         p++;
 501     }
 502
 503     h64 ^= h64 >> 33;
 504     h64 *= PRIME64_2;
 505     h64 ^= h64 >> 29;
 506     h64 *= PRIME64_3;
 507     h64 ^= h64 >> 32;
 508
 509     return h64;
 510 }
 511
 512
 513 XXH_PUBLIC_API unsigned long long XXH64 (const void* input, size_t len, unsigned long long seed)
 514 {
 515 #if 0
 516     /* Simple version, good for code maintenance, but unfortunately slow for small inputs */
 517     XXH64_CREATESTATE_STATIC(state);
 518     XXH64_reset(state, seed);
 519     XXH64_update(state, input, len);
 520     return XXH64_digest(state);
 521 #else
 522     XXH_endianess endian_detected = (XXH_endianess)XXH_CPU_LITTLE_ENDIAN;
 523
 524     if (XXH_FORCE_ALIGN_CHECK) {
 525         if ((((size_t)input) & 7)==0) {  /* Input is aligned, let's leverage the speed advantage */
 526             if ((endian_detected==XXH_littleEndian) || XXH_FORCE_NATIVE_FORMAT)
 527                 return XXH64_endian_align(input, len, seed, XXH_littleEndian, XXH_aligned);
 528             else
 529                 return XXH64_endian_align(input, len, seed, XXH_bigEndian, XXH_aligned);
 530     }   }
 531
 532     if ((endian_detected==XXH_littleEndian) || XXH_FORCE_NATIVE_FORMAT)
 533         return XXH64_endian_align(input, len, seed, XXH_littleEndian, XXH_unaligned);
 534     else
 535         return XXH64_endian_align(input, len, seed, XXH_bigEndian, XXH_unaligned);
 536 #endif
 537 }
 538
 539
 540 /* **************************************************
 541 *  Advanced Hash Functions
 542 ****************************************************/
 543
 544 XXH_PUBLIC_API XXH32_state_t* XXH32_createState(void)
 545 {
 546     return (XXH32_state_t*)XXH_malloc(sizeof(XXH32_state_t));
 547 }
 548 XXH_PUBLIC_API XXH_errorcode XXH32_freeState(XXH32_state_t* statePtr)
 549 {
 550     XXH_free(statePtr);
 551     return XXH_OK;
 552 }
 553
 554 XXH_PUBLIC_API XXH64_state_t* XXH64_createState(void)
 555 {
 556     return (XXH64_state_t*)XXH_malloc(sizeof(XXH64_state_t));
 557 }
 558 XXH_PUBLIC_API XXH_errorcode XXH64_freeState(XXH64_state_t* statePtr)
 559 {
 560     XXH_free(statePtr);
 561     return XXH_OK;
 562 }
 563
 564
 565 /*** Hash feed ***/
 566
 567 XXH_PUBLIC_API XXH_errorcode XXH32_reset(XXH32_state_t* statePtr, unsigned int seed)
 568 {
 569     XXH32_state_t state;   /* using a local state to memcpy() in order to avoid strict-aliasing warnings */
 570     memset(&state, 0, sizeof(state)-4);   /* do not write into reserved, for future removal */
 571     state.v1 = seed + PRIME32_1 + PRIME32_2;
 572     state.v2 = seed + PRIME32_2;
 573     state.v3 = seed + 0;
 574     state.v4 = seed - PRIME32_1;
 575     memcpy(statePtr, &state, sizeof(state));
 576     return XXH_OK;
 577 }
 578
 579
 580 XXH_PUBLIC_API XXH_errorcode XXH64_reset(XXH64_state_t* statePtr, unsigned long long seed)
 581 {
 582     XXH64_state_t state;   /* using a local state to memcpy() in order to avoid strict-aliasing warnings */
 583     memset(&state, 0, sizeof(state)-8);   /* do not write into reserved, for future removal */
 584     state.v1 = seed + PRIME64_1 + PRIME64_2;
 585     state.v2 = seed + PRIME64_2;
 586     state.v3 = seed + 0;
 587     state.v4 = seed - PRIME64_1;
 588     memcpy(statePtr, &state, sizeof(state));
 589     return XXH_OK;
 590 }
 591
 592
 593 FORCE_INLINE_TEMPLATE XXH_errorcode XXH32_update_endian (XXH32_state_t* state, const void* input, size_t len, XXH_endianess endian)
 594 {
 595     const BYTE* p = (const BYTE*)input;
 596     const BYTE* const bEnd = p + len;
 597
 598 #ifdef XXH_ACCEPT_NULL_INPUT_POINTER
 599     if (input==NULL) return XXH_ERROR;
 600 #endif
 601
 602     state->total_len_32 += (unsigned)len;
 603     state->large_len |= (len>=16) | (state->total_len_32>=16);
 604
 605     if (state->memsize + len < 16)  {   /* fill in tmp buffer */
 606         XXH_memcpy((BYTE*)(state->mem32) + state->memsize, input, len);
 607         state->memsize += (unsigned)len;
 608         return XXH_OK;
 609     }
 610
 611     if (state->memsize) {   /* some data left from previous update */
 612         XXH_memcpy((BYTE*)(state->mem32) + state->memsize, input, 16-state->memsize);
 613         {   const U32* p32 = state->mem32;
 614             state->v1 = XXH32_round(state->v1, XXH_readLE32(p32, endian)); p32++;
 615             state->v2 = XXH32_round(state->v2, XXH_readLE32(p32, endian)); p32++;
 616             state->v3 = XXH32_round(state->v3, XXH_readLE32(p32, endian)); p32++;
 617             state->v4 = XXH32_round(state->v4, XXH_readLE32(p32, endian)); p32++;
 618         }
 619         p += 16-state->memsize;
 620         state->memsize = 0;
 621     }
 622
 623     if (p <= bEnd-16) {
 624         const BYTE* const limit = bEnd - 16;
 625         U32 v1 = state->v1;
 626         U32 v2 = state->v2;
 627         U32 v3 = state->v3;
 628         U32 v4 = state->v4;
 629
 630         do {
 631             v1 = XXH32_round(v1, XXH_readLE32(p, endian)); p+=4;
 632             v2 = XXH32_round(v2, XXH_readLE32(p, endian)); p+=4;
 633             v3 = XXH32_round(v3, XXH_readLE32(p, endian)); p+=4;
 634             v4 = XXH32_round(v4, XXH_readLE32(p, endian)); p+=4;
 635         } while (p<=limit);
 636
 637         state->v1 = v1;
 638         state->v2 = v2;
 639         state->v3 = v3;
 640         state->v4 = v4;
 641     }
 642
 643     if (p < bEnd) {
 644         XXH_memcpy(state->mem32, p, (size_t)(bEnd-p));
 645         state->memsize = (unsigned)(bEnd-p);
 646     }
 647
 648     return XXH_OK;
 649 }
 650
 651 XXH_PUBLIC_API XXH_errorcode XXH32_update (XXH32_state_t* state_in, const void* input, size_t len)
 652 {
 653     XXH_endianess endian_detected = (XXH_endianess)XXH_CPU_LITTLE_ENDIAN;
 654
 655     if ((endian_detected==XXH_littleEndian) || XXH_FORCE_NATIVE_FORMAT)
 656         return XXH32_update_endian(state_in, input, len, XXH_littleEndian);
 657     else
 658         return XXH32_update_endian(state_in, input, len, XXH_bigEndian);
 659 }
 660
 661
 662
 663 FORCE_INLINE_TEMPLATE U32 XXH32_digest_endian (const XXH32_state_t* state, XXH_endianess endian)
 664 {
 665     const BYTE * p = (const BYTE*)state->mem32;
 666     const BYTE* const bEnd = (const BYTE*)(state->mem32) + state->memsize;
 667     U32 h32;
 668
 669     if (state->large_len) {
 670         h32 = XXH_rotl32(state->v1, 1) + XXH_rotl32(state->v2, 7) + XXH_rotl32(state->v3, 12) + XXH_rotl32(state->v4, 18);
 671     } else {
 672         h32 = state->v3 /* == seed */ + PRIME32_5;
 673     }
 674
 675     h32 += state->total_len_32;
 676
 677     while (p+4<=bEnd) {
 678         h32 += XXH_readLE32(p, endian) * PRIME32_3;
 679         h32  = XXH_rotl32(h32, 17) * PRIME32_4;
 680         p+=4;
 681     }
 682
 683     while (p<bEnd) {
 684         h32 += (*p) * PRIME32_5;
 685         h32  = XXH_rotl32(h32, 11) * PRIME32_1;
 686         p++;
 687     }
 688
 689     h32 ^= h32 >> 15;
 690     h32 *= PRIME32_2;
 691     h32 ^= h32 >> 13;
 692     h32 *= PRIME32_3;
 693     h32 ^= h32 >> 16;
 694
 695     return h32;
 696 }
 697
 698
 699 XXH_PUBLIC_API unsigned int XXH32_digest (const XXH32_state_t* state_in)
 700 {
 701     XXH_endianess endian_detected = (XXH_endianess)XXH_CPU_LITTLE_ENDIAN;
 702
 703     if ((endian_detected==XXH_littleEndian) || XXH_FORCE_NATIVE_FORMAT)
 704         return XXH32_digest_endian(state_in, XXH_littleEndian);
 705     else
 706         return XXH32_digest_endian(state_in, XXH_bigEndian);
 707 }
 708
 709
 710
 711 /* **** XXH64 **** */
 712
 713 FORCE_INLINE_TEMPLATE XXH_errorcode XXH64_update_endian (XXH64_state_t* state, const void* input, size_t len, XXH_endianess endian)
 714 {
 715     const BYTE* p = (const BYTE*)input;
 716     const BYTE* const bEnd = p + len;
 717
 718 #ifdef XXH_ACCEPT_NULL_INPUT_POINTER
 719     if (input==NULL) return XXH_ERROR;
 720 #endif
 721
 722     state->total_len += len;
 723
 724     if (state->memsize + len < 32) {  /* fill in tmp buffer */
 725         XXH_memcpy(((BYTE*)state->mem64) + state->memsize, input, len);
 726         state->memsize += (U32)len;
 727         return XXH_OK;
 728     }
 729
 730     if (state->memsize) {   /* tmp buffer is full */
 731         XXH_memcpy(((BYTE*)state->mem64) + state->memsize, input, 32-state->memsize);
 732         state->v1 = XXH64_round(state->v1, XXH_readLE64(state->mem64+0, endian));
 733         state->v2 = XXH64_round(state->v2, XXH_readLE64(state->mem64+1, endian));
 734         state->v3 = XXH64_round(state->v3, XXH_readLE64(state->mem64+2, endian));
 735         state->v4 = XXH64_round(state->v4, XXH_readLE64(state->mem64+3, endian));
 736         p += 32-state->memsize;
 737         state->memsize = 0;
 738     }
 739
 740     if (p+32 <= bEnd) {
 741         const BYTE* const limit = bEnd - 32;
 742         U64 v1 = state->v1;
 743         U64 v2 = state->v2;
 744         U64 v3 = state->v3;
 745         U64 v4 = state->v4;
 746
 747         do {
 748             v1 = XXH64_round(v1, XXH_readLE64(p, endian)); p+=8;
 749             v2 = XXH64_round(v2, XXH_readLE64(p, endian)); p+=8;
 750             v3 = XXH64_round(v3, XXH_readLE64(p, endian)); p+=8;
 751             v4 = XXH64_round(v4, XXH_readLE64(p, endian)); p+=8;
 752         } while (p<=limit);
 753
 754         state->v1 = v1;
 755         state->v2 = v2;
 756         state->v3 = v3;
 757         state->v4 = v4;
 758     }
 759
 760     if (p < bEnd) {
 761         XXH_memcpy(state->mem64, p, (size_t)(bEnd-p));
 762         state->memsize = (unsigned)(bEnd-p);
 763     }
 764
 765     return XXH_OK;
 766 }
 767
 768 XXH_PUBLIC_API XXH_errorcode XXH64_update (XXH64_state_t* state_in, const void* input, size_t len)
 769 {
 770     XXH_endianess endian_detected = (XXH_endianess)XXH_CPU_LITTLE_ENDIAN;
 771
 772     if ((endian_detected==XXH_littleEndian) || XXH_FORCE_NATIVE_FORMAT)
 773         return XXH64_update_endian(state_in, input, len, XXH_littleEndian);
 774     else
 775         return XXH64_update_endian(state_in, input, len, XXH_bigEndian);
 776 }
 777
 778
 779
 780 FORCE_INLINE_TEMPLATE U64 XXH64_digest_endian (const XXH64_state_t* state, XXH_endianess endian)
 781 {
 782     const BYTE * p = (const BYTE*)state->mem64;
 783     const BYTE* const bEnd = (const BYTE*)state->mem64 + state->memsize;
 784     U64 h64;
 785
 786     if (state->total_len >= 32) {
 787         U64 const v1 = state->v1;
 788         U64 const v2 = state->v2;
 789         U64 const v3 = state->v3;
 790         U64 const v4 = state->v4;
 791
 792         h64 = XXH_rotl64(v1, 1) + XXH_rotl64(v2, 7) + XXH_rotl64(v3, 12) + XXH_rotl64(v4, 18);
 793         h64 = XXH64_mergeRound(h64, v1);
 794         h64 = XXH64_mergeRound(h64, v2);
 795         h64 = XXH64_mergeRound(h64, v3);
 796         h64 = XXH64_mergeRound(h64, v4);
 797     } else {
 798         h64  = state->v3 + PRIME64_5;
 799     }
 800
 801     h64 += (U64) state->total_len;
 802
 803     while (p+8<=bEnd) {
 804         U64 const k1 = XXH64_round(0, XXH_readLE64(p, endian));
 805         h64 ^= k1;
 806         h64  = XXH_rotl64(h64,27) * PRIME64_1 + PRIME64_4;
 807         p+=8;
 808     }
 809
 810     if (p+4<=bEnd) {
 811         h64 ^= (U64)(XXH_readLE32(p, endian)) * PRIME64_1;
 812         h64  = XXH_rotl64(h64, 23) * PRIME64_2 + PRIME64_3;
 813         p+=4;
 814     }
 815
 816     while (p<bEnd) {
 817         h64 ^= (*p) * PRIME64_5;
 818         h64  = XXH_rotl64(h64, 11) * PRIME64_1;
 819         p++;
 820     }
 821
 822     h64 ^= h64 >> 33;
 823     h64 *= PRIME64_2;
 824     h64 ^= h64 >> 29;
 825     h64 *= PRIME64_3;
 826     h64 ^= h64 >> 32;
 827
 828     return h64;
 829 }
 830
 831
 832 XXH_PUBLIC_API unsigned long long XXH64_digest (const XXH64_state_t* state_in)
 833 {
 834     XXH_endianess endian_detected = (XXH_endianess)XXH_CPU_LITTLE_ENDIAN;
 835
 836     if ((endian_detected==XXH_littleEndian) || XXH_FORCE_NATIVE_FORMAT)
 837         return XXH64_digest_endian(state_in, XXH_littleEndian);
 838     else
 839         return XXH64_digest_endian(state_in, XXH_bigEndian);
 840 }
 841
 842
 843 /* **************************
 844 *  Canonical representation
 845 ****************************/
 846
 847 /*! Default XXH result types are basic unsigned 32 and 64 bits.
 848 *   The canonical representation follows human-readable write convention, aka big-endian (large digits first).
 849 *   These functions allow transformation of hash result into and from its canonical format.
 850 *   This way, hash values can be written into a file or buffer, and remain comparable across different systems and programs.
 851 */
 852
 853 XXH_PUBLIC_API void XXH32_canonicalFromHash(XXH32_canonical_t* dst, XXH32_hash_t hash)
 854 {
 855     XXH_STATIC_ASSERT(sizeof(XXH32_canonical_t) == sizeof(XXH32_hash_t));
 856     if (XXH_CPU_LITTLE_ENDIAN) hash = XXH_swap32(hash);
 857     memcpy(dst, &hash, sizeof(*dst));
 858 }
 859
 860 XXH_PUBLIC_API void XXH64_canonicalFromHash(XXH64_canonical_t* dst, XXH64_hash_t hash)
 861 {
 862     XXH_STATIC_ASSERT(sizeof(XXH64_canonical_t) == sizeof(XXH64_hash_t));
 863     if (XXH_CPU_LITTLE_ENDIAN) hash = XXH_swap64(hash);
 864     memcpy(dst, &hash, sizeof(*dst));
 865 }
 866
 867 XXH_PUBLIC_API XXH32_hash_t XXH32_hashFromCanonical(const XXH32_canonical_t* src)
 868 {
 869     return XXH_readBE32(src);
 870 }
 871
 872 XXH_PUBLIC_API XXH64_hash_t XXH64_hashFromCanonical(const XXH64_canonical_t* src)
 873 {
 874     return XXH_readBE64(src);
 875 }