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; 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 /*! and 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 #define XXH_STATIC_LINKING_ONLY 108 #include "xxhash.h" 109 110 111 /* ************************************* 112 * Compiler Specific Options 113 ***************************************/ 114 #ifdef _MSC_VER /* Visual Studio */ 115 # pragma warning(disable : 4127) /* disable: C4127: conditional expression is constant */ 116 # define FORCE_INLINE static __forceinline 117 #else 118 # if defined (__cplusplus) || defined (__STDC_VERSION__) && __STDC_VERSION__ >= 199901L /* C99 */ 119 # ifdef __GNUC__ 120 # define FORCE_INLINE static inline __attribute__((always_inline)) 121 # else 122 # define FORCE_INLINE static inline 123 # endif 124 # else 125 # define FORCE_INLINE static 126 # endif /* __STDC_VERSION__ */ 127 #endif 128 129 130 /* ************************************* 131 * Basic Types 132 ***************************************/ 133 #ifndef MEM_MODULE 134 # if !defined (__VMS) && (defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) ) 135 # include <stdint.h> 136 typedef uint8_t BYTE; 137 typedef uint16_t U16; 138 typedef uint32_t U32; 139 typedef int32_t S32; 140 # else 141 typedef unsigned char BYTE; 142 typedef unsigned short U16; 143 typedef unsigned int U32; 144 typedef signed int S32; 145 # endif 146 #endif 147 148 #if (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==2)) 149 150 /* Force direct memory access. Only works on CPU which support unaligned memory access in hardware */ 151 static U32 XXH_read32(const void* memPtr) { return *(const U32*) memPtr; } 152 153 #elif (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==1)) 154 155 /* __pack instructions are safer, but compiler specific, hence potentially problematic for some compilers */ 156 /* currently only defined for gcc and icc */ 157 typedef union { U32 u32; } __attribute__((packed)) unalign; 158 static U32 XXH_read32(const void* ptr) { return ((const unalign*)ptr)->u32; } 159 160 #else 161 162 /* portable and safe solution. Generally efficient. 163 * see : http://stackoverflow.com/a/32095106/646947 164 */ 165 static U32 XXH_read32(const void* memPtr) 166 { 167 U32 val; 168 memcpy(&val, memPtr, sizeof(val)); 169 return val; 170 } 171 172 #endif /* XXH_FORCE_DIRECT_MEMORY_ACCESS */ 173 174 175 /* **************************************** 176 * Compiler-specific Functions and Macros 177 ******************************************/ 178 #define XXH_GCC_VERSION (__GNUC__ * 100 + __GNUC_MINOR__) 179 180 /* Note : although _rotl exists for minGW (GCC under windows), performance seems poor */ 181 #if defined(_MSC_VER) 182 # define XXH_rotl32(x,r) _rotl(x,r) 183 # define XXH_rotl64(x,r) _rotl64(x,r) 184 #else 185 # define XXH_rotl32(x,r) ((x << r) | (x >> (32 - r))) 186 # define XXH_rotl64(x,r) ((x << r) | (x >> (64 - r))) 187 #endif 188 189 #if defined(_MSC_VER) /* Visual Studio */ 190 # define XXH_swap32 _byteswap_ulong 191 #elif XXH_GCC_VERSION >= 403 192 # define XXH_swap32 __builtin_bswap32 193 #else 194 static U32 XXH_swap32 (U32 x) 195 { 196 return ((x << 24) & 0xff000000 ) | 197 ((x << 8) & 0x00ff0000 ) | 198 ((x >> 8) & 0x0000ff00 ) | 199 ((x >> 24) & 0x000000ff ); 200 } 201 #endif 202 203 204 /* ************************************* 205 * Architecture Macros 206 ***************************************/ 207 typedef enum { XXH_bigEndian=0, XXH_littleEndian=1 } XXH_endianess; 208 209 /* XXH_CPU_LITTLE_ENDIAN can be defined externally, for example on the compiler command line */ 210 #ifndef XXH_CPU_LITTLE_ENDIAN 211 static const int g_one = 1; 212 # define XXH_CPU_LITTLE_ENDIAN (*(const char*)(&g_one)) 213 #endif 214 215 216 /* *************************** 217 * Memory reads 218 *****************************/ 219 typedef enum { XXH_aligned, XXH_unaligned } XXH_alignment; 220 221 FORCE_INLINE U32 XXH_readLE32_align(const void* ptr, XXH_endianess endian, XXH_alignment align) 222 { 223 if (align==XXH_unaligned) 224 return endian==XXH_littleEndian ? XXH_read32(ptr) : XXH_swap32(XXH_read32(ptr)); 225 else 226 return endian==XXH_littleEndian ? *(const U32*)ptr : XXH_swap32(*(const U32*)ptr); 227 } 228 229 FORCE_INLINE U32 XXH_readLE32(const void* ptr, XXH_endianess endian) 230 { 231 return XXH_readLE32_align(ptr, endian, XXH_unaligned); 232 } 233 234 static U32 XXH_readBE32(const void* ptr) 235 { 236 return XXH_CPU_LITTLE_ENDIAN ? XXH_swap32(XXH_read32(ptr)) : XXH_read32(ptr); 237 } 238 239 240 /* ************************************* 241 * Macros 242 ***************************************/ 243 #define XXH_STATIC_ASSERT(c) { enum { XXH_static_assert = 1/(int)(!!(c)) }; } /* use only *after* variable declarations */ 244 XXH_PUBLIC_API unsigned XXH_versionNumber (void) { return XXH_VERSION_NUMBER; } 245 246 247 /* ******************************************************************* 248 * 32-bits hash functions 249 *********************************************************************/ 250 static const U32 PRIME32_1 = 2654435761U; 251 static const U32 PRIME32_2 = 2246822519U; 252 static const U32 PRIME32_3 = 3266489917U; 253 static const U32 PRIME32_4 = 668265263U; 254 static const U32 PRIME32_5 = 374761393U; 255 256 static U32 XXH32_round(U32 seed, U32 input) 257 { 258 seed += input * PRIME32_2; 259 seed = XXH_rotl32(seed, 13); 260 seed *= PRIME32_1; 261 return seed; 262 } 263 264 FORCE_INLINE U32 XXH32_endian_align(const void* input, size_t len, U32 seed, XXH_endianess endian, XXH_alignment align) 265 { 266 const BYTE* p = (const BYTE*)input; 267 const BYTE* bEnd = p + len; 268 U32 h32; 269 #define XXH_get32bits(p) XXH_readLE32_align(p, endian, align) 270 271 #ifdef XXH_ACCEPT_NULL_INPUT_POINTER 272 if (p==NULL) { 273 len=0; 274 bEnd=p=(const BYTE*)(size_t)16; 275 } 276 #endif 277 278 if (len>=16) { 279 const BYTE* const limit = bEnd - 16; 280 U32 v1 = seed + PRIME32_1 + PRIME32_2; 281 U32 v2 = seed + PRIME32_2; 282 U32 v3 = seed + 0; 283 U32 v4 = seed - PRIME32_1; 284 285 do { 286 v1 = XXH32_round(v1, XXH_get32bits(p)); p+=4; 287 v2 = XXH32_round(v2, XXH_get32bits(p)); p+=4; 288 v3 = XXH32_round(v3, XXH_get32bits(p)); p+=4; 289 v4 = XXH32_round(v4, XXH_get32bits(p)); p+=4; 290 } while (p<=limit); 291 292 h32 = XXH_rotl32(v1, 1) + XXH_rotl32(v2, 7) + XXH_rotl32(v3, 12) + XXH_rotl32(v4, 18); 293 } else { 294 h32 = seed + PRIME32_5; 295 } 296 297 h32 += (U32) len; 298 299 while (p+4<=bEnd) { 300 h32 += XXH_get32bits(p) * PRIME32_3; 301 h32 = XXH_rotl32(h32, 17) * PRIME32_4 ; 302 p+=4; 303 } 304 305 while (p<bEnd) { 306 h32 += (*p) * PRIME32_5; 307 h32 = XXH_rotl32(h32, 11) * PRIME32_1 ; 308 p++; 309 } 310 311 h32 ^= h32 >> 15; 312 h32 *= PRIME32_2; 313 h32 ^= h32 >> 13; 314 h32 *= PRIME32_3; 315 h32 ^= h32 >> 16; 316 317 return h32; 318 } 319 320 321 XXH_PUBLIC_API unsigned int XXH32 (const void* input, size_t len, unsigned int seed) 322 { 323 #if 0 324 /* Simple version, good for code maintenance, but unfortunately slow for small inputs */ 325 XXH32_state_t state; 326 XXH32_reset(&state, seed); 327 XXH32_update(&state, input, len); 328 return XXH32_digest(&state); 329 #else 330 XXH_endianess endian_detected = (XXH_endianess)XXH_CPU_LITTLE_ENDIAN; 331 332 if (XXH_FORCE_ALIGN_CHECK) { 333 if ((((size_t)input) & 3) == 0) { /* Input is 4-bytes aligned, leverage the speed benefit */ 334 if ((endian_detected==XXH_littleEndian) || XXH_FORCE_NATIVE_FORMAT) 335 return XXH32_endian_align(input, len, seed, XXH_littleEndian, XXH_aligned); 336 else 337 return XXH32_endian_align(input, len, seed, XXH_bigEndian, XXH_aligned); 338 } } 339 340 if ((endian_detected==XXH_littleEndian) || XXH_FORCE_NATIVE_FORMAT) 341 return XXH32_endian_align(input, len, seed, XXH_littleEndian, XXH_unaligned); 342 else 343 return XXH32_endian_align(input, len, seed, XXH_bigEndian, XXH_unaligned); 344 #endif 345 } 346 347 348 349 /*====== Hash streaming ======*/ 350 351 XXH_PUBLIC_API XXH32_state_t* XXH32_createState(void) 352 { 353 return (XXH32_state_t*)XXH_malloc(sizeof(XXH32_state_t)); 354 } 355 XXH_PUBLIC_API XXH_errorcode XXH32_freeState(XXH32_state_t* statePtr) 356 { 357 XXH_free(statePtr); 358 return XXH_OK; 359 } 360 361 XXH_PUBLIC_API void XXH32_copyState(XXH32_state_t* dstState, const XXH32_state_t* srcState) 362 { 363 memcpy(dstState, srcState, sizeof(*dstState)); 364 } 365 366 XXH_PUBLIC_API XXH_errorcode XXH32_reset(XXH32_state_t* statePtr, unsigned int seed) 367 { 368 XXH32_state_t state; /* using a local state to memcpy() in order to avoid strict-aliasing warnings */ 369 memset(&state, 0, sizeof(state)-4); /* do not write into reserved, for future removal */ 370 state.v1 = seed + PRIME32_1 + PRIME32_2; 371 state.v2 = seed + PRIME32_2; 372 state.v3 = seed + 0; 373 state.v4 = seed - PRIME32_1; 374 memcpy(statePtr, &state, sizeof(state)); 375 return XXH_OK; 376 } 377 378 379 FORCE_INLINE XXH_errorcode XXH32_update_endian (XXH32_state_t* state, const void* input, size_t len, XXH_endianess endian) 380 { 381 const BYTE* p = (const BYTE*)input; 382 const BYTE* const bEnd = p + len; 383 384 #ifdef XXH_ACCEPT_NULL_INPUT_POINTER 385 if (input==NULL) return XXH_ERROR; 386 #endif 387 388 state->total_len_32 += (unsigned)len; 389 state->large_len |= (len>=16) | (state->total_len_32>=16); 390 391 if (state->memsize + len < 16) { /* fill in tmp buffer */ 392 XXH_memcpy((BYTE*)(state->mem32) + state->memsize, input, len); 393 state->memsize += (unsigned)len; 394 return XXH_OK; 395 } 396 397 if (state->memsize) { /* some data left from previous update */ 398 XXH_memcpy((BYTE*)(state->mem32) + state->memsize, input, 16-state->memsize); 399 { const U32* p32 = state->mem32; 400 state->v1 = XXH32_round(state->v1, XXH_readLE32(p32, endian)); p32++; 401 state->v2 = XXH32_round(state->v2, XXH_readLE32(p32, endian)); p32++; 402 state->v3 = XXH32_round(state->v3, XXH_readLE32(p32, endian)); p32++; 403 state->v4 = XXH32_round(state->v4, XXH_readLE32(p32, endian)); p32++; 404 } 405 p += 16-state->memsize; 406 state->memsize = 0; 407 } 408 409 if (p <= bEnd-16) { 410 const BYTE* const limit = bEnd - 16; 411 U32 v1 = state->v1; 412 U32 v2 = state->v2; 413 U32 v3 = state->v3; 414 U32 v4 = state->v4; 415 416 do { 417 v1 = XXH32_round(v1, XXH_readLE32(p, endian)); p+=4; 418 v2 = XXH32_round(v2, XXH_readLE32(p, endian)); p+=4; 419 v3 = XXH32_round(v3, XXH_readLE32(p, endian)); p+=4; 420 v4 = XXH32_round(v4, XXH_readLE32(p, endian)); p+=4; 421 } while (p<=limit); 422 423 state->v1 = v1; 424 state->v2 = v2; 425 state->v3 = v3; 426 state->v4 = v4; 427 } 428 429 if (p < bEnd) { 430 XXH_memcpy(state->mem32, p, (size_t)(bEnd-p)); 431 state->memsize = (unsigned)(bEnd-p); 432 } 433 434 return XXH_OK; 435 } 436 437 XXH_PUBLIC_API XXH_errorcode XXH32_update (XXH32_state_t* state_in, const void* input, size_t len) 438 { 439 XXH_endianess endian_detected = (XXH_endianess)XXH_CPU_LITTLE_ENDIAN; 440 441 if ((endian_detected==XXH_littleEndian) || XXH_FORCE_NATIVE_FORMAT) 442 return XXH32_update_endian(state_in, input, len, XXH_littleEndian); 443 else 444 return XXH32_update_endian(state_in, input, len, XXH_bigEndian); 445 } 446 447 448 449 FORCE_INLINE U32 XXH32_digest_endian (const XXH32_state_t* state, XXH_endianess endian) 450 { 451 const BYTE * p = (const BYTE*)state->mem32; 452 const BYTE* const bEnd = (const BYTE*)(state->mem32) + state->memsize; 453 U32 h32; 454 455 if (state->large_len) { 456 h32 = XXH_rotl32(state->v1, 1) + XXH_rotl32(state->v2, 7) + XXH_rotl32(state->v3, 12) + XXH_rotl32(state->v4, 18); 457 } else { 458 h32 = state->v3 /* == seed */ + PRIME32_5; 459 } 460 461 h32 += state->total_len_32; 462 463 while (p+4<=bEnd) { 464 h32 += XXH_readLE32(p, endian) * PRIME32_3; 465 h32 = XXH_rotl32(h32, 17) * PRIME32_4; 466 p+=4; 467 } 468 469 while (p<bEnd) { 470 h32 += (*p) * PRIME32_5; 471 h32 = XXH_rotl32(h32, 11) * PRIME32_1; 472 p++; 473 } 474 475 h32 ^= h32 >> 15; 476 h32 *= PRIME32_2; 477 h32 ^= h32 >> 13; 478 h32 *= PRIME32_3; 479 h32 ^= h32 >> 16; 480 481 return h32; 482 } 483 484 485 XXH_PUBLIC_API unsigned int XXH32_digest (const XXH32_state_t* state_in) 486 { 487 XXH_endianess endian_detected = (XXH_endianess)XXH_CPU_LITTLE_ENDIAN; 488 489 if ((endian_detected==XXH_littleEndian) || XXH_FORCE_NATIVE_FORMAT) 490 return XXH32_digest_endian(state_in, XXH_littleEndian); 491 else 492 return XXH32_digest_endian(state_in, XXH_bigEndian); 493 } 494 495 496 /*====== Canonical representation ======*/ 497 498 /*! Default XXH result types are basic unsigned 32 and 64 bits. 499 * The canonical representation follows human-readable write convention, aka big-endian (large digits first). 500 * These functions allow transformation of hash result into and from its canonical format. 501 * This way, hash values can be written into a file or buffer, and remain comparable across different systems and programs. 502 */ 503 504 XXH_PUBLIC_API void XXH32_canonicalFromHash(XXH32_canonical_t* dst, XXH32_hash_t hash) 505 { 506 XXH_STATIC_ASSERT(sizeof(XXH32_canonical_t) == sizeof(XXH32_hash_t)); 507 if (XXH_CPU_LITTLE_ENDIAN) hash = XXH_swap32(hash); 508 memcpy(dst, &hash, sizeof(*dst)); 509 } 510 511 XXH_PUBLIC_API XXH32_hash_t XXH32_hashFromCanonical(const XXH32_canonical_t* src) 512 { 513 return XXH_readBE32(src); 514 } 515 516 517 #ifndef XXH_NO_LONG_LONG 518 519 /* ******************************************************************* 520 * 64-bits hash functions 521 *********************************************************************/ 522 523 /*====== Memory access ======*/ 524 525 #ifndef MEM_MODULE 526 # define MEM_MODULE 527 # if !defined (__VMS) && (defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) ) 528 # include <stdint.h> 529 typedef uint64_t U64; 530 # else 531 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. */ 532 # endif 533 #endif 534 535 536 #if (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==2)) 537 538 /* Force direct memory access. Only works on CPU which support unaligned memory access in hardware */ 539 static U64 XXH_read64(const void* memPtr) { return *(const U64*) memPtr; } 540 541 #elif (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==1)) 542 543 /* __pack instructions are safer, but compiler specific, hence potentially problematic for some compilers */ 544 /* currently only defined for gcc and icc */ 545 typedef union { U32 u32; U64 u64; } __attribute__((packed)) unalign64; 546 static U64 XXH_read64(const void* ptr) { return ((const unalign64*)ptr)->u64; } 547 548 #else 549 550 /* portable and safe solution. Generally efficient. 551 * see : http://stackoverflow.com/a/32095106/646947 552 */ 553 554 static U64 XXH_read64(const void* memPtr) 555 { 556 U64 val; 557 memcpy(&val, memPtr, sizeof(val)); 558 return val; 559 } 560 561 #endif /* XXH_FORCE_DIRECT_MEMORY_ACCESS */ 562 563 #if defined(_MSC_VER) /* Visual Studio */ 564 # define XXH_swap64 _byteswap_uint64 565 #elif XXH_GCC_VERSION >= 403 566 # define XXH_swap64 __builtin_bswap64 567 #else 568 static U64 XXH_swap64 (U64 x) 569 { 570 return ((x << 56) & 0xff00000000000000ULL) | 571 ((x << 40) & 0x00ff000000000000ULL) | 572 ((x << 24) & 0x0000ff0000000000ULL) | 573 ((x << 8) & 0x000000ff00000000ULL) | 574 ((x >> 8) & 0x00000000ff000000ULL) | 575 ((x >> 24) & 0x0000000000ff0000ULL) | 576 ((x >> 40) & 0x000000000000ff00ULL) | 577 ((x >> 56) & 0x00000000000000ffULL); 578 } 579 #endif 580 581 FORCE_INLINE U64 XXH_readLE64_align(const void* ptr, XXH_endianess endian, XXH_alignment align) 582 { 583 if (align==XXH_unaligned) 584 return endian==XXH_littleEndian ? XXH_read64(ptr) : XXH_swap64(XXH_read64(ptr)); 585 else 586 return endian==XXH_littleEndian ? *(const U64*)ptr : XXH_swap64(*(const U64*)ptr); 587 } 588 589 FORCE_INLINE U64 XXH_readLE64(const void* ptr, XXH_endianess endian) 590 { 591 return XXH_readLE64_align(ptr, endian, XXH_unaligned); 592 } 593 594 static U64 XXH_readBE64(const void* ptr) 595 { 596 return XXH_CPU_LITTLE_ENDIAN ? XXH_swap64(XXH_read64(ptr)) : XXH_read64(ptr); 597 } 598 599 600 /*====== xxh64 ======*/ 601 602 static const U64 PRIME64_1 = 11400714785074694791ULL; 603 static const U64 PRIME64_2 = 14029467366897019727ULL; 604 static const U64 PRIME64_3 = 1609587929392839161ULL; 605 static const U64 PRIME64_4 = 9650029242287828579ULL; 606 static const U64 PRIME64_5 = 2870177450012600261ULL; 607 608 static U64 XXH64_round(U64 acc, U64 input) 609 { 610 acc += input * PRIME64_2; 611 acc = XXH_rotl64(acc, 31); 612 acc *= PRIME64_1; 613 return acc; 614 } 615 616 static U64 XXH64_mergeRound(U64 acc, U64 val) 617 { 618 val = XXH64_round(0, val); 619 acc ^= val; 620 acc = acc * PRIME64_1 + PRIME64_4; 621 return acc; 622 } 623 624 FORCE_INLINE U64 XXH64_endian_align(const void* input, size_t len, U64 seed, XXH_endianess endian, XXH_alignment align) 625 { 626 const BYTE* p = (const BYTE*)input; 627 const BYTE* const bEnd = p + len; 628 U64 h64; 629 #define XXH_get64bits(p) XXH_readLE64_align(p, endian, align) 630 631 #ifdef XXH_ACCEPT_NULL_INPUT_POINTER 632 if (p==NULL) { 633 len=0; 634 bEnd=p=(const BYTE*)(size_t)32; 635 } 636 #endif 637 638 if (len>=32) { 639 const BYTE* const limit = bEnd - 32; 640 U64 v1 = seed + PRIME64_1 + PRIME64_2; 641 U64 v2 = seed + PRIME64_2; 642 U64 v3 = seed + 0; 643 U64 v4 = seed - PRIME64_1; 644 645 do { 646 v1 = XXH64_round(v1, XXH_get64bits(p)); p+=8; 647 v2 = XXH64_round(v2, XXH_get64bits(p)); p+=8; 648 v3 = XXH64_round(v3, XXH_get64bits(p)); p+=8; 649 v4 = XXH64_round(v4, XXH_get64bits(p)); p+=8; 650 } while (p<=limit); 651 652 h64 = XXH_rotl64(v1, 1) + XXH_rotl64(v2, 7) + XXH_rotl64(v3, 12) + XXH_rotl64(v4, 18); 653 h64 = XXH64_mergeRound(h64, v1); 654 h64 = XXH64_mergeRound(h64, v2); 655 h64 = XXH64_mergeRound(h64, v3); 656 h64 = XXH64_mergeRound(h64, v4); 657 658 } else { 659 h64 = seed + PRIME64_5; 660 } 661 662 h64 += (U64) len; 663 664 while (p+8<=bEnd) { 665 U64 const k1 = XXH64_round(0, XXH_get64bits(p)); 666 h64 ^= k1; 667 h64 = XXH_rotl64(h64,27) * PRIME64_1 + PRIME64_4; 668 p+=8; 669 } 670 671 if (p+4<=bEnd) { 672 h64 ^= (U64)(XXH_get32bits(p)) * PRIME64_1; 673 h64 = XXH_rotl64(h64, 23) * PRIME64_2 + PRIME64_3; 674 p+=4; 675 } 676 677 while (p<bEnd) { 678 h64 ^= (*p) * PRIME64_5; 679 h64 = XXH_rotl64(h64, 11) * PRIME64_1; 680 p++; 681 } 682 683 h64 ^= h64 >> 33; 684 h64 *= PRIME64_2; 685 h64 ^= h64 >> 29; 686 h64 *= PRIME64_3; 687 h64 ^= h64 >> 32; 688 689 return h64; 690 } 691 692 693 XXH_PUBLIC_API unsigned long long XXH64 (const void* input, size_t len, unsigned long long seed) 694 { 695 #if 0 696 /* Simple version, good for code maintenance, but unfortunately slow for small inputs */ 697 XXH64_state_t state; 698 XXH64_reset(&state, seed); 699 XXH64_update(&state, input, len); 700 return XXH64_digest(&state); 701 #else 702 XXH_endianess endian_detected = (XXH_endianess)XXH_CPU_LITTLE_ENDIAN; 703 704 if (XXH_FORCE_ALIGN_CHECK) { 705 if ((((size_t)input) & 7)==0) { /* Input is aligned, let's leverage the speed advantage */ 706 if ((endian_detected==XXH_littleEndian) || XXH_FORCE_NATIVE_FORMAT) 707 return XXH64_endian_align(input, len, seed, XXH_littleEndian, XXH_aligned); 708 else 709 return XXH64_endian_align(input, len, seed, XXH_bigEndian, XXH_aligned); 710 } } 711 712 if ((endian_detected==XXH_littleEndian) || XXH_FORCE_NATIVE_FORMAT) 713 return XXH64_endian_align(input, len, seed, XXH_littleEndian, XXH_unaligned); 714 else 715 return XXH64_endian_align(input, len, seed, XXH_bigEndian, XXH_unaligned); 716 #endif 717 } 718 719 /*====== Hash Streaming ======*/ 720 721 XXH_PUBLIC_API XXH64_state_t* XXH64_createState(void) 722 { 723 return (XXH64_state_t*)XXH_malloc(sizeof(XXH64_state_t)); 724 } 725 XXH_PUBLIC_API XXH_errorcode XXH64_freeState(XXH64_state_t* statePtr) 726 { 727 XXH_free(statePtr); 728 return XXH_OK; 729 } 730 731 XXH_PUBLIC_API void XXH64_copyState(XXH64_state_t* dstState, const XXH64_state_t* srcState) 732 { 733 memcpy(dstState, srcState, sizeof(*dstState)); 734 } 735 736 XXH_PUBLIC_API XXH_errorcode XXH64_reset(XXH64_state_t* statePtr, unsigned long long seed) 737 { 738 XXH64_state_t state; /* using a local state to memcpy() in order to avoid strict-aliasing warnings */ 739 memset(&state, 0, sizeof(state)-8); /* do not write into reserved, for future removal */ 740 state.v1 = seed + PRIME64_1 + PRIME64_2; 741 state.v2 = seed + PRIME64_2; 742 state.v3 = seed + 0; 743 state.v4 = seed - PRIME64_1; 744 memcpy(statePtr, &state, sizeof(state)); 745 return XXH_OK; 746 } 747 748 FORCE_INLINE XXH_errorcode XXH64_update_endian (XXH64_state_t* state, const void* input, size_t len, XXH_endianess endian) 749 { 750 const BYTE* p = (const BYTE*)input; 751 const BYTE* const bEnd = p + len; 752 753 #ifdef XXH_ACCEPT_NULL_INPUT_POINTER 754 if (input==NULL) return XXH_ERROR; 755 #endif 756 757 state->total_len += len; 758 759 if (state->memsize + len < 32) { /* fill in tmp buffer */ 760 XXH_memcpy(((BYTE*)state->mem64) + state->memsize, input, len); 761 state->memsize += (U32)len; 762 return XXH_OK; 763 } 764 765 if (state->memsize) { /* tmp buffer is full */ 766 XXH_memcpy(((BYTE*)state->mem64) + state->memsize, input, 32-state->memsize); 767 state->v1 = XXH64_round(state->v1, XXH_readLE64(state->mem64+0, endian)); 768 state->v2 = XXH64_round(state->v2, XXH_readLE64(state->mem64+1, endian)); 769 state->v3 = XXH64_round(state->v3, XXH_readLE64(state->mem64+2, endian)); 770 state->v4 = XXH64_round(state->v4, XXH_readLE64(state->mem64+3, endian)); 771 p += 32-state->memsize; 772 state->memsize = 0; 773 } 774 775 if (p+32 <= bEnd) { 776 const BYTE* const limit = bEnd - 32; 777 U64 v1 = state->v1; 778 U64 v2 = state->v2; 779 U64 v3 = state->v3; 780 U64 v4 = state->v4; 781 782 do { 783 v1 = XXH64_round(v1, XXH_readLE64(p, endian)); p+=8; 784 v2 = XXH64_round(v2, XXH_readLE64(p, endian)); p+=8; 785 v3 = XXH64_round(v3, XXH_readLE64(p, endian)); p+=8; 786 v4 = XXH64_round(v4, XXH_readLE64(p, endian)); p+=8; 787 } while (p<=limit); 788 789 state->v1 = v1; 790 state->v2 = v2; 791 state->v3 = v3; 792 state->v4 = v4; 793 } 794 795 if (p < bEnd) { 796 XXH_memcpy(state->mem64, p, (size_t)(bEnd-p)); 797 state->memsize = (unsigned)(bEnd-p); 798 } 799 800 return XXH_OK; 801 } 802 803 XXH_PUBLIC_API XXH_errorcode XXH64_update (XXH64_state_t* state_in, const void* input, size_t len) 804 { 805 XXH_endianess endian_detected = (XXH_endianess)XXH_CPU_LITTLE_ENDIAN; 806 807 if ((endian_detected==XXH_littleEndian) || XXH_FORCE_NATIVE_FORMAT) 808 return XXH64_update_endian(state_in, input, len, XXH_littleEndian); 809 else 810 return XXH64_update_endian(state_in, input, len, XXH_bigEndian); 811 } 812 813 FORCE_INLINE U64 XXH64_digest_endian (const XXH64_state_t* state, XXH_endianess endian) 814 { 815 const BYTE * p = (const BYTE*)state->mem64; 816 const BYTE* const bEnd = (const BYTE*)state->mem64 + state->memsize; 817 U64 h64; 818 819 if (state->total_len >= 32) { 820 U64 const v1 = state->v1; 821 U64 const v2 = state->v2; 822 U64 const v3 = state->v3; 823 U64 const v4 = state->v4; 824 825 h64 = XXH_rotl64(v1, 1) + XXH_rotl64(v2, 7) + XXH_rotl64(v3, 12) + XXH_rotl64(v4, 18); 826 h64 = XXH64_mergeRound(h64, v1); 827 h64 = XXH64_mergeRound(h64, v2); 828 h64 = XXH64_mergeRound(h64, v3); 829 h64 = XXH64_mergeRound(h64, v4); 830 } else { 831 h64 = state->v3 + PRIME64_5; 832 } 833 834 h64 += (U64) state->total_len; 835 836 while (p+8<=bEnd) { 837 U64 const k1 = XXH64_round(0, XXH_readLE64(p, endian)); 838 h64 ^= k1; 839 h64 = XXH_rotl64(h64,27) * PRIME64_1 + PRIME64_4; 840 p+=8; 841 } 842 843 if (p+4<=bEnd) { 844 h64 ^= (U64)(XXH_readLE32(p, endian)) * PRIME64_1; 845 h64 = XXH_rotl64(h64, 23) * PRIME64_2 + PRIME64_3; 846 p+=4; 847 } 848 849 while (p<bEnd) { 850 h64 ^= (*p) * PRIME64_5; 851 h64 = XXH_rotl64(h64, 11) * PRIME64_1; 852 p++; 853 } 854 855 h64 ^= h64 >> 33; 856 h64 *= PRIME64_2; 857 h64 ^= h64 >> 29; 858 h64 *= PRIME64_3; 859 h64 ^= h64 >> 32; 860 861 return h64; 862 } 863 864 XXH_PUBLIC_API unsigned long long XXH64_digest (const XXH64_state_t* state_in) 865 { 866 XXH_endianess endian_detected = (XXH_endianess)XXH_CPU_LITTLE_ENDIAN; 867 868 if ((endian_detected==XXH_littleEndian) || XXH_FORCE_NATIVE_FORMAT) 869 return XXH64_digest_endian(state_in, XXH_littleEndian); 870 else 871 return XXH64_digest_endian(state_in, XXH_bigEndian); 872 } 873 874 875 /*====== Canonical representation ======*/ 876 877 XXH_PUBLIC_API void XXH64_canonicalFromHash(XXH64_canonical_t* dst, XXH64_hash_t hash) 878 { 879 XXH_STATIC_ASSERT(sizeof(XXH64_canonical_t) == sizeof(XXH64_hash_t)); 880 if (XXH_CPU_LITTLE_ENDIAN) hash = XXH_swap64(hash); 881 memcpy(dst, &hash, sizeof(*dst)); 882 } 883 884 XXH_PUBLIC_API XXH64_hash_t XXH64_hashFromCanonical(const XXH64_canonical_t* src) 885 { 886 return XXH_readBE64(src); 887 } 888 889 #endif /* XXH_NO_LONG_LONG */ 890
