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-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>
XXH_malloc(size_t s)101 static void* XXH_malloc(size_t s) { return malloc(s); }
XXH_free(void * p)102 static void XXH_free (void* p) { free(p); }
103 /*! and for memcpy() */
104 #include <string.h>
XXH_memcpy(void * dest,const void * src,size_t size)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 #endif
117
118 #ifndef XXH_FORCE_INLINE
119 # ifdef _MSC_VER /* Visual Studio */
120 # define XXH_FORCE_INLINE static __forceinline
121 # else
122 # if defined (__cplusplus) || defined (__STDC_VERSION__) && __STDC_VERSION__ >= 199901L /* C99 */
123 # ifdef __GNUC__
124 # define XXH_FORCE_INLINE static inline __attribute__((always_inline))
125 # else
126 # define XXH_FORCE_INLINE static inline
127 # endif
128 # else
129 # define XXH_FORCE_INLINE static
130 # endif /* __STDC_VERSION__ */
131 # endif /* _MSC_VER */
132 #endif /* XXH_FORCE_INLINE */
133
134
135 /* *************************************
136 * Basic Types
137 ***************************************/
138 #ifndef MEM_MODULE
139 # if !defined (__VMS) && (defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) )
140 # include <stdint.h>
141 typedef uint8_t BYTE;
142 typedef uint16_t U16;
143 typedef uint32_t U32;
144 typedef int32_t S32;
145 # else
146 typedef unsigned char BYTE;
147 typedef unsigned short U16;
148 typedef unsigned int U32;
149 typedef signed int S32;
150 # endif
151 #endif
152
153 #if (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==2))
154
155 /* Force direct memory access. Only works on CPU which support unaligned memory access in hardware */
XXH_read32(const void * memPtr)156 static U32 XXH_read32(const void* memPtr) { return *(const U32*) memPtr; }
157
158 #elif (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==1))
159
160 /* __pack instructions are safer, but compiler specific, hence potentially problematic for some compilers */
161 /* currently only defined for gcc and icc */
162 typedef union { U32 u32; } __attribute__((packed)) unalign;
XXH_read32(const void * ptr)163 static U32 XXH_read32(const void* ptr) { return ((const unalign*)ptr)->u32; }
164
165 #else
166
167 /* portable and safe solution. Generally efficient.
168 * see : http://stackoverflow.com/a/32095106/646947
169 */
XXH_read32(const void * memPtr)170 static U32 XXH_read32(const void* memPtr)
171 {
172 U32 val;
173 memcpy(&val, memPtr, sizeof(val));
174 return val;
175 }
176
177 #endif /* XXH_FORCE_DIRECT_MEMORY_ACCESS */
178
179
180 /* ****************************************
181 * Compiler-specific Functions and Macros
182 ******************************************/
183 #define XXH_GCC_VERSION (__GNUC__ * 100 + __GNUC_MINOR__)
184
185 /* Note : although _rotl exists for minGW (GCC under windows), performance seems poor */
186 #if defined(_MSC_VER)
187 # define XXH_rotl32(x,r) _rotl(x,r)
188 # define XXH_rotl64(x,r) _rotl64(x,r)
189 #else
190 # define XXH_rotl32(x,r) ((x << r) | (x >> (32 - r)))
191 # define XXH_rotl64(x,r) ((x << r) | (x >> (64 - r)))
192 #endif
193
194 #if defined(_MSC_VER) /* Visual Studio */
195 # define XXH_swap32 _byteswap_ulong
196 #elif XXH_GCC_VERSION >= 403
197 # define XXH_swap32 __builtin_bswap32
198 #else
XXH_swap32(U32 x)199 static U32 XXH_swap32 (U32 x)
200 {
201 return ((x << 24) & 0xff000000 ) |
202 ((x << 8) & 0x00ff0000 ) |
203 ((x >> 8) & 0x0000ff00 ) |
204 ((x >> 24) & 0x000000ff );
205 }
206 #endif
207
208
209 /* *************************************
210 * Architecture Macros
211 ***************************************/
212 typedef enum { XXH_bigEndian=0, XXH_littleEndian=1 } XXH_endianess;
213
214 /* XXH_CPU_LITTLE_ENDIAN can be defined externally, for example on the compiler command line */
215 #ifndef XXH_CPU_LITTLE_ENDIAN
216 static const int g_one = 1;
217 # define XXH_CPU_LITTLE_ENDIAN (*(const char*)(&g_one))
218 #endif
219
220
221 /* ***************************
222 * Memory reads
223 *****************************/
224 typedef enum { XXH_aligned, XXH_unaligned } XXH_alignment;
225
XXH_readLE32_align(const void * ptr,XXH_endianess endian,XXH_alignment align)226 XXH_FORCE_INLINE U32 XXH_readLE32_align(const void* ptr, XXH_endianess endian, XXH_alignment align)
227 {
228 if (align==XXH_unaligned)
229 return endian==XXH_littleEndian ? XXH_read32(ptr) : XXH_swap32(XXH_read32(ptr));
230 else
231 return endian==XXH_littleEndian ? *(const U32*)ptr : XXH_swap32(*(const U32*)ptr);
232 }
233
XXH_readLE32(const void * ptr,XXH_endianess endian)234 XXH_FORCE_INLINE U32 XXH_readLE32(const void* ptr, XXH_endianess endian)
235 {
236 return XXH_readLE32_align(ptr, endian, XXH_unaligned);
237 }
238
XXH_readBE32(const void * ptr)239 static U32 XXH_readBE32(const void* ptr)
240 {
241 return XXH_CPU_LITTLE_ENDIAN ? XXH_swap32(XXH_read32(ptr)) : XXH_read32(ptr);
242 }
243
244
245 /* *************************************
246 * Macros
247 ***************************************/
248 #define XXH_STATIC_ASSERT(c) { enum { XXH_static_assert = 1/(int)(!!(c)) }; } /* use only *after* variable declarations */
XXH_versionNumber(void)249 XXH_PUBLIC_API unsigned XXH_versionNumber (void) { return XXH_VERSION_NUMBER; }
250
251
252 /* *******************************************************************
253 * 32-bits hash functions
254 *********************************************************************/
255 static const U32 PRIME32_1 = 2654435761U;
256 static const U32 PRIME32_2 = 2246822519U;
257 static const U32 PRIME32_3 = 3266489917U;
258 static const U32 PRIME32_4 = 668265263U;
259 static const U32 PRIME32_5 = 374761393U;
260
XXH32_round(U32 seed,U32 input)261 static U32 XXH32_round(U32 seed, U32 input)
262 {
263 seed += input * PRIME32_2;
264 seed = XXH_rotl32(seed, 13);
265 seed *= PRIME32_1;
266 return seed;
267 }
268
XXH32_endian_align(const void * input,size_t len,U32 seed,XXH_endianess endian,XXH_alignment align)269 XXH_FORCE_INLINE U32 XXH32_endian_align(const void* input, size_t len, U32 seed, XXH_endianess endian, XXH_alignment align)
270 {
271 const BYTE* p = (const BYTE*)input;
272 const BYTE* bEnd = p + len;
273 U32 h32;
274 #define XXH_get32bits(p) XXH_readLE32_align(p, endian, align)
275
276 #ifdef XXH_ACCEPT_NULL_INPUT_POINTER
277 if (p==NULL) {
278 len=0;
279 bEnd=p=(const BYTE*)(size_t)16;
280 }
281 #endif
282
283 if (len>=16) {
284 const BYTE* const limit = bEnd - 16;
285 U32 v1 = seed + PRIME32_1 + PRIME32_2;
286 U32 v2 = seed + PRIME32_2;
287 U32 v3 = seed + 0;
288 U32 v4 = seed - PRIME32_1;
289
290 do {
291 v1 = XXH32_round(v1, XXH_get32bits(p)); p+=4;
292 v2 = XXH32_round(v2, XXH_get32bits(p)); p+=4;
293 v3 = XXH32_round(v3, XXH_get32bits(p)); p+=4;
294 v4 = XXH32_round(v4, XXH_get32bits(p)); p+=4;
295 } while (p<=limit);
296
297 h32 = XXH_rotl32(v1, 1) + XXH_rotl32(v2, 7) + XXH_rotl32(v3, 12) + XXH_rotl32(v4, 18);
298 } else {
299 h32 = seed + PRIME32_5;
300 }
301
302 h32 += (U32) len;
303
304 while (p+4<=bEnd) {
305 h32 += XXH_get32bits(p) * PRIME32_3;
306 h32 = XXH_rotl32(h32, 17) * PRIME32_4 ;
307 p+=4;
308 }
309
310 while (p<bEnd) {
311 h32 += (*p) * PRIME32_5;
312 h32 = XXH_rotl32(h32, 11) * PRIME32_1 ;
313 p++;
314 }
315
316 h32 ^= h32 >> 15;
317 h32 *= PRIME32_2;
318 h32 ^= h32 >> 13;
319 h32 *= PRIME32_3;
320 h32 ^= h32 >> 16;
321
322 return h32;
323 }
324
325
XXH32(const void * input,size_t len,unsigned int seed)326 XXH_PUBLIC_API unsigned int XXH32 (const void* input, size_t len, unsigned int seed)
327 {
328 #if 0
329 /* Simple version, good for code maintenance, but unfortunately slow for small inputs */
330 XXH32_state_t state;
331 XXH32_reset(&state, seed);
332 XXH32_update(&state, input, len);
333 return XXH32_digest(&state);
334 #else
335 XXH_endianess endian_detected = (XXH_endianess)XXH_CPU_LITTLE_ENDIAN;
336
337 if (XXH_FORCE_ALIGN_CHECK) {
338 if ((((size_t)input) & 3) == 0) { /* Input is 4-bytes aligned, leverage the speed benefit */
339 if ((endian_detected==XXH_littleEndian) || XXH_FORCE_NATIVE_FORMAT)
340 return XXH32_endian_align(input, len, seed, XXH_littleEndian, XXH_aligned);
341 else
342 return XXH32_endian_align(input, len, seed, XXH_bigEndian, XXH_aligned);
343 } }
344
345 if ((endian_detected==XXH_littleEndian) || XXH_FORCE_NATIVE_FORMAT)
346 return XXH32_endian_align(input, len, seed, XXH_littleEndian, XXH_unaligned);
347 else
348 return XXH32_endian_align(input, len, seed, XXH_bigEndian, XXH_unaligned);
349 #endif
350 }
351
352
353
354 /*====== Hash streaming ======*/
355
XXH32_createState(void)356 XXH_PUBLIC_API XXH32_state_t* XXH32_createState(void)
357 {
358 return (XXH32_state_t*)XXH_malloc(sizeof(XXH32_state_t));
359 }
XXH32_freeState(XXH32_state_t * statePtr)360 XXH_PUBLIC_API XXH_errorcode XXH32_freeState(XXH32_state_t* statePtr)
361 {
362 XXH_free(statePtr);
363 return XXH_OK;
364 }
365
XXH32_copyState(XXH32_state_t * dstState,const XXH32_state_t * srcState)366 XXH_PUBLIC_API void XXH32_copyState(XXH32_state_t* dstState, const XXH32_state_t* srcState)
367 {
368 memcpy(dstState, srcState, sizeof(*dstState));
369 }
370
XXH32_reset(XXH32_state_t * statePtr,unsigned int seed)371 XXH_PUBLIC_API XXH_errorcode XXH32_reset(XXH32_state_t* statePtr, unsigned int seed)
372 {
373 XXH32_state_t state; /* using a local state to memcpy() in order to avoid strict-aliasing warnings */
374 memset(&state, 0, sizeof(state)-4); /* do not write into reserved, for future removal */
375 state.v1 = seed + PRIME32_1 + PRIME32_2;
376 state.v2 = seed + PRIME32_2;
377 state.v3 = seed + 0;
378 state.v4 = seed - PRIME32_1;
379 memcpy(statePtr, &state, sizeof(state));
380 return XXH_OK;
381 }
382
383
XXH32_update_endian(XXH32_state_t * state,const void * input,size_t len,XXH_endianess endian)384 XXH_FORCE_INLINE XXH_errorcode XXH32_update_endian (XXH32_state_t* state, const void* input, size_t len, XXH_endianess endian)
385 {
386 const BYTE* p = (const BYTE*)input;
387 const BYTE* const bEnd = p + len;
388
389 #ifdef XXH_ACCEPT_NULL_INPUT_POINTER
390 if (input==NULL) return XXH_ERROR;
391 #endif
392
393 state->total_len_32 += (unsigned)len;
394 state->large_len |= (len>=16) | (state->total_len_32>=16);
395
396 if (state->memsize + len < 16) { /* fill in tmp buffer */
397 XXH_memcpy((BYTE*)(state->mem32) + state->memsize, input, len);
398 state->memsize += (unsigned)len;
399 return XXH_OK;
400 }
401
402 if (state->memsize) { /* some data left from previous update */
403 XXH_memcpy((BYTE*)(state->mem32) + state->memsize, input, 16-state->memsize);
404 { const U32* p32 = state->mem32;
405 state->v1 = XXH32_round(state->v1, XXH_readLE32(p32, endian)); p32++;
406 state->v2 = XXH32_round(state->v2, XXH_readLE32(p32, endian)); p32++;
407 state->v3 = XXH32_round(state->v3, XXH_readLE32(p32, endian)); p32++;
408 state->v4 = XXH32_round(state->v4, XXH_readLE32(p32, endian)); p32++;
409 }
410 p += 16-state->memsize;
411 state->memsize = 0;
412 }
413
414 if (p <= bEnd-16) {
415 const BYTE* const limit = bEnd - 16;
416 U32 v1 = state->v1;
417 U32 v2 = state->v2;
418 U32 v3 = state->v3;
419 U32 v4 = state->v4;
420
421 do {
422 v1 = XXH32_round(v1, XXH_readLE32(p, endian)); p+=4;
423 v2 = XXH32_round(v2, XXH_readLE32(p, endian)); p+=4;
424 v3 = XXH32_round(v3, XXH_readLE32(p, endian)); p+=4;
425 v4 = XXH32_round(v4, XXH_readLE32(p, endian)); p+=4;
426 } while (p<=limit);
427
428 state->v1 = v1;
429 state->v2 = v2;
430 state->v3 = v3;
431 state->v4 = v4;
432 }
433
434 if (p < bEnd) {
435 XXH_memcpy(state->mem32, p, (size_t)(bEnd-p));
436 state->memsize = (unsigned)(bEnd-p);
437 }
438
439 return XXH_OK;
440 }
441
XXH32_update(XXH32_state_t * state_in,const void * input,size_t len)442 XXH_PUBLIC_API XXH_errorcode XXH32_update (XXH32_state_t* state_in, const void* input, size_t len)
443 {
444 XXH_endianess endian_detected = (XXH_endianess)XXH_CPU_LITTLE_ENDIAN;
445
446 if ((endian_detected==XXH_littleEndian) || XXH_FORCE_NATIVE_FORMAT)
447 return XXH32_update_endian(state_in, input, len, XXH_littleEndian);
448 else
449 return XXH32_update_endian(state_in, input, len, XXH_bigEndian);
450 }
451
452
453
XXH32_digest_endian(const XXH32_state_t * state,XXH_endianess endian)454 XXH_FORCE_INLINE U32 XXH32_digest_endian (const XXH32_state_t* state, XXH_endianess endian)
455 {
456 const BYTE * p = (const BYTE*)state->mem32;
457 const BYTE* const bEnd = (const BYTE*)(state->mem32) + state->memsize;
458 U32 h32;
459
460 if (state->large_len) {
461 h32 = XXH_rotl32(state->v1, 1) + XXH_rotl32(state->v2, 7) + XXH_rotl32(state->v3, 12) + XXH_rotl32(state->v4, 18);
462 } else {
463 h32 = state->v3 /* == seed */ + PRIME32_5;
464 }
465
466 h32 += state->total_len_32;
467
468 while (p+4<=bEnd) {
469 h32 += XXH_readLE32(p, endian) * PRIME32_3;
470 h32 = XXH_rotl32(h32, 17) * PRIME32_4;
471 p+=4;
472 }
473
474 while (p<bEnd) {
475 h32 += (*p) * PRIME32_5;
476 h32 = XXH_rotl32(h32, 11) * PRIME32_1;
477 p++;
478 }
479
480 h32 ^= h32 >> 15;
481 h32 *= PRIME32_2;
482 h32 ^= h32 >> 13;
483 h32 *= PRIME32_3;
484 h32 ^= h32 >> 16;
485
486 return h32;
487 }
488
489
XXH32_digest(const XXH32_state_t * state_in)490 XXH_PUBLIC_API unsigned int XXH32_digest (const XXH32_state_t* state_in)
491 {
492 XXH_endianess endian_detected = (XXH_endianess)XXH_CPU_LITTLE_ENDIAN;
493
494 if ((endian_detected==XXH_littleEndian) || XXH_FORCE_NATIVE_FORMAT)
495 return XXH32_digest_endian(state_in, XXH_littleEndian);
496 else
497 return XXH32_digest_endian(state_in, XXH_bigEndian);
498 }
499
500
501 /*====== Canonical representation ======*/
502
503 /*! Default XXH result types are basic unsigned 32 and 64 bits.
504 * The canonical representation follows human-readable write convention, aka big-endian (large digits first).
505 * These functions allow transformation of hash result into and from its canonical format.
506 * This way, hash values can be written into a file or buffer, and remain comparable across different systems and programs.
507 */
508
XXH32_canonicalFromHash(XXH32_canonical_t * dst,XXH32_hash_t hash)509 XXH_PUBLIC_API void XXH32_canonicalFromHash(XXH32_canonical_t* dst, XXH32_hash_t hash)
510 {
511 XXH_STATIC_ASSERT(sizeof(XXH32_canonical_t) == sizeof(XXH32_hash_t));
512 if (XXH_CPU_LITTLE_ENDIAN) hash = XXH_swap32(hash);
513 memcpy(dst, &hash, sizeof(*dst));
514 }
515
XXH32_hashFromCanonical(const XXH32_canonical_t * src)516 XXH_PUBLIC_API XXH32_hash_t XXH32_hashFromCanonical(const XXH32_canonical_t* src)
517 {
518 return XXH_readBE32(src);
519 }
520
521
522 #ifndef XXH_NO_LONG_LONG
523
524 /* *******************************************************************
525 * 64-bits hash functions
526 *********************************************************************/
527
528 /*====== Memory access ======*/
529
530 #ifndef MEM_MODULE
531 # define MEM_MODULE
532 # if !defined (__VMS) && (defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) )
533 # include <stdint.h>
534 typedef uint64_t U64;
535 # else
536 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. */
537 # endif
538 #endif
539
540
541 #if (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==2))
542
543 /* Force direct memory access. Only works on CPU which support unaligned memory access in hardware */
XXH_read64(const void * memPtr)544 static U64 XXH_read64(const void* memPtr) { return *(const U64*) memPtr; }
545
546 #elif (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==1))
547
548 /* __pack instructions are safer, but compiler specific, hence potentially problematic for some compilers */
549 /* currently only defined for gcc and icc */
550 typedef union { U32 u32; U64 u64; } __attribute__((packed)) unalign64;
XXH_read64(const void * ptr)551 static U64 XXH_read64(const void* ptr) { return ((const unalign64*)ptr)->u64; }
552
553 #else
554
555 /* portable and safe solution. Generally efficient.
556 * see : http://stackoverflow.com/a/32095106/646947
557 */
558
XXH_read64(const void * memPtr)559 static U64 XXH_read64(const void* memPtr)
560 {
561 U64 val;
562 memcpy(&val, memPtr, sizeof(val));
563 return val;
564 }
565
566 #endif /* XXH_FORCE_DIRECT_MEMORY_ACCESS */
567
568 #if defined(_MSC_VER) /* Visual Studio */
569 # define XXH_swap64 _byteswap_uint64
570 #elif XXH_GCC_VERSION >= 403
571 # define XXH_swap64 __builtin_bswap64
572 #else
XXH_swap64(U64 x)573 static U64 XXH_swap64 (U64 x)
574 {
575 return ((x << 56) & 0xff00000000000000ULL) |
576 ((x << 40) & 0x00ff000000000000ULL) |
577 ((x << 24) & 0x0000ff0000000000ULL) |
578 ((x << 8) & 0x000000ff00000000ULL) |
579 ((x >> 8) & 0x00000000ff000000ULL) |
580 ((x >> 24) & 0x0000000000ff0000ULL) |
581 ((x >> 40) & 0x000000000000ff00ULL) |
582 ((x >> 56) & 0x00000000000000ffULL);
583 }
584 #endif
585
XXH_readLE64_align(const void * ptr,XXH_endianess endian,XXH_alignment align)586 XXH_FORCE_INLINE U64 XXH_readLE64_align(const void* ptr, XXH_endianess endian, XXH_alignment align)
587 {
588 if (align==XXH_unaligned)
589 return endian==XXH_littleEndian ? XXH_read64(ptr) : XXH_swap64(XXH_read64(ptr));
590 else
591 return endian==XXH_littleEndian ? *(const U64*)ptr : XXH_swap64(*(const U64*)ptr);
592 }
593
XXH_readLE64(const void * ptr,XXH_endianess endian)594 XXH_FORCE_INLINE U64 XXH_readLE64(const void* ptr, XXH_endianess endian)
595 {
596 return XXH_readLE64_align(ptr, endian, XXH_unaligned);
597 }
598
XXH_readBE64(const void * ptr)599 static U64 XXH_readBE64(const void* ptr)
600 {
601 return XXH_CPU_LITTLE_ENDIAN ? XXH_swap64(XXH_read64(ptr)) : XXH_read64(ptr);
602 }
603
604
605 /*====== xxh64 ======*/
606
607 static const U64 PRIME64_1 = 11400714785074694791ULL;
608 static const U64 PRIME64_2 = 14029467366897019727ULL;
609 static const U64 PRIME64_3 = 1609587929392839161ULL;
610 static const U64 PRIME64_4 = 9650029242287828579ULL;
611 static const U64 PRIME64_5 = 2870177450012600261ULL;
612
XXH64_round(U64 acc,U64 input)613 static U64 XXH64_round(U64 acc, U64 input)
614 {
615 acc += input * PRIME64_2;
616 acc = XXH_rotl64(acc, 31);
617 acc *= PRIME64_1;
618 return acc;
619 }
620
XXH64_mergeRound(U64 acc,U64 val)621 static U64 XXH64_mergeRound(U64 acc, U64 val)
622 {
623 val = XXH64_round(0, val);
624 acc ^= val;
625 acc = acc * PRIME64_1 + PRIME64_4;
626 return acc;
627 }
628
XXH64_endian_align(const void * input,size_t len,U64 seed,XXH_endianess endian,XXH_alignment align)629 XXH_FORCE_INLINE U64 XXH64_endian_align(const void* input, size_t len, U64 seed, XXH_endianess endian, XXH_alignment align)
630 {
631 const BYTE* p = (const BYTE*)input;
632 const BYTE* const bEnd = p + len;
633 U64 h64;
634 #define XXH_get64bits(p) XXH_readLE64_align(p, endian, align)
635
636 #ifdef XXH_ACCEPT_NULL_INPUT_POINTER
637 if (p==NULL) {
638 len=0;
639 bEnd=p=(const BYTE*)(size_t)32;
640 }
641 #endif
642
643 if (len>=32) {
644 const BYTE* const limit = bEnd - 32;
645 U64 v1 = seed + PRIME64_1 + PRIME64_2;
646 U64 v2 = seed + PRIME64_2;
647 U64 v3 = seed + 0;
648 U64 v4 = seed - PRIME64_1;
649
650 do {
651 v1 = XXH64_round(v1, XXH_get64bits(p)); p+=8;
652 v2 = XXH64_round(v2, XXH_get64bits(p)); p+=8;
653 v3 = XXH64_round(v3, XXH_get64bits(p)); p+=8;
654 v4 = XXH64_round(v4, XXH_get64bits(p)); p+=8;
655 } while (p<=limit);
656
657 h64 = XXH_rotl64(v1, 1) + XXH_rotl64(v2, 7) + XXH_rotl64(v3, 12) + XXH_rotl64(v4, 18);
658 h64 = XXH64_mergeRound(h64, v1);
659 h64 = XXH64_mergeRound(h64, v2);
660 h64 = XXH64_mergeRound(h64, v3);
661 h64 = XXH64_mergeRound(h64, v4);
662
663 } else {
664 h64 = seed + PRIME64_5;
665 }
666
667 h64 += (U64) len;
668
669 while (p+8<=bEnd) {
670 U64 const k1 = XXH64_round(0, XXH_get64bits(p));
671 h64 ^= k1;
672 h64 = XXH_rotl64(h64,27) * PRIME64_1 + PRIME64_4;
673 p+=8;
674 }
675
676 if (p+4<=bEnd) {
677 h64 ^= (U64)(XXH_get32bits(p)) * PRIME64_1;
678 h64 = XXH_rotl64(h64, 23) * PRIME64_2 + PRIME64_3;
679 p+=4;
680 }
681
682 while (p<bEnd) {
683 h64 ^= (*p) * PRIME64_5;
684 h64 = XXH_rotl64(h64, 11) * PRIME64_1;
685 p++;
686 }
687
688 h64 ^= h64 >> 33;
689 h64 *= PRIME64_2;
690 h64 ^= h64 >> 29;
691 h64 *= PRIME64_3;
692 h64 ^= h64 >> 32;
693
694 return h64;
695 }
696
697
XXH64(const void * input,size_t len,unsigned long long seed)698 XXH_PUBLIC_API unsigned long long XXH64 (const void* input, size_t len, unsigned long long seed)
699 {
700 #if 0
701 /* Simple version, good for code maintenance, but unfortunately slow for small inputs */
702 XXH64_state_t state;
703 XXH64_reset(&state, seed);
704 XXH64_update(&state, input, len);
705 return XXH64_digest(&state);
706 #else
707 XXH_endianess endian_detected = (XXH_endianess)XXH_CPU_LITTLE_ENDIAN;
708
709 if (XXH_FORCE_ALIGN_CHECK) {
710 if ((((size_t)input) & 7)==0) { /* Input is aligned, let's leverage the speed advantage */
711 if ((endian_detected==XXH_littleEndian) || XXH_FORCE_NATIVE_FORMAT)
712 return XXH64_endian_align(input, len, seed, XXH_littleEndian, XXH_aligned);
713 else
714 return XXH64_endian_align(input, len, seed, XXH_bigEndian, XXH_aligned);
715 } }
716
717 if ((endian_detected==XXH_littleEndian) || XXH_FORCE_NATIVE_FORMAT)
718 return XXH64_endian_align(input, len, seed, XXH_littleEndian, XXH_unaligned);
719 else
720 return XXH64_endian_align(input, len, seed, XXH_bigEndian, XXH_unaligned);
721 #endif
722 }
723
724 /*====== Hash Streaming ======*/
725
XXH64_createState(void)726 XXH_PUBLIC_API XXH64_state_t* XXH64_createState(void)
727 {
728 return (XXH64_state_t*)XXH_malloc(sizeof(XXH64_state_t));
729 }
XXH64_freeState(XXH64_state_t * statePtr)730 XXH_PUBLIC_API XXH_errorcode XXH64_freeState(XXH64_state_t* statePtr)
731 {
732 XXH_free(statePtr);
733 return XXH_OK;
734 }
735
XXH64_copyState(XXH64_state_t * dstState,const XXH64_state_t * srcState)736 XXH_PUBLIC_API void XXH64_copyState(XXH64_state_t* dstState, const XXH64_state_t* srcState)
737 {
738 memcpy(dstState, srcState, sizeof(*dstState));
739 }
740
XXH64_reset(XXH64_state_t * statePtr,unsigned long long seed)741 XXH_PUBLIC_API XXH_errorcode XXH64_reset(XXH64_state_t* statePtr, unsigned long long seed)
742 {
743 XXH64_state_t state; /* using a local state to memcpy() in order to avoid strict-aliasing warnings */
744 memset(&state, 0, sizeof(state)-8); /* do not write into reserved, for future removal */
745 state.v1 = seed + PRIME64_1 + PRIME64_2;
746 state.v2 = seed + PRIME64_2;
747 state.v3 = seed + 0;
748 state.v4 = seed - PRIME64_1;
749 memcpy(statePtr, &state, sizeof(state));
750 return XXH_OK;
751 }
752
XXH64_update_endian(XXH64_state_t * state,const void * input,size_t len,XXH_endianess endian)753 XXH_FORCE_INLINE XXH_errorcode XXH64_update_endian (XXH64_state_t* state, const void* input, size_t len, XXH_endianess endian)
754 {
755 const BYTE* p = (const BYTE*)input;
756 const BYTE* const bEnd = p + len;
757
758 #ifdef XXH_ACCEPT_NULL_INPUT_POINTER
759 if (input==NULL) return XXH_ERROR;
760 #endif
761
762 state->total_len += len;
763
764 if (state->memsize + len < 32) { /* fill in tmp buffer */
765 XXH_memcpy(((BYTE*)state->mem64) + state->memsize, input, len);
766 state->memsize += (U32)len;
767 return XXH_OK;
768 }
769
770 if (state->memsize) { /* tmp buffer is full */
771 XXH_memcpy(((BYTE*)state->mem64) + state->memsize, input, 32-state->memsize);
772 state->v1 = XXH64_round(state->v1, XXH_readLE64(state->mem64+0, endian));
773 state->v2 = XXH64_round(state->v2, XXH_readLE64(state->mem64+1, endian));
774 state->v3 = XXH64_round(state->v3, XXH_readLE64(state->mem64+2, endian));
775 state->v4 = XXH64_round(state->v4, XXH_readLE64(state->mem64+3, endian));
776 p += 32-state->memsize;
777 state->memsize = 0;
778 }
779
780 if (p+32 <= bEnd) {
781 const BYTE* const limit = bEnd - 32;
782 U64 v1 = state->v1;
783 U64 v2 = state->v2;
784 U64 v3 = state->v3;
785 U64 v4 = state->v4;
786
787 do {
788 v1 = XXH64_round(v1, XXH_readLE64(p, endian)); p+=8;
789 v2 = XXH64_round(v2, XXH_readLE64(p, endian)); p+=8;
790 v3 = XXH64_round(v3, XXH_readLE64(p, endian)); p+=8;
791 v4 = XXH64_round(v4, XXH_readLE64(p, endian)); p+=8;
792 } while (p<=limit);
793
794 state->v1 = v1;
795 state->v2 = v2;
796 state->v3 = v3;
797 state->v4 = v4;
798 }
799
800 if (p < bEnd) {
801 XXH_memcpy(state->mem64, p, (size_t)(bEnd-p));
802 state->memsize = (unsigned)(bEnd-p);
803 }
804
805 return XXH_OK;
806 }
807
XXH64_update(XXH64_state_t * state_in,const void * input,size_t len)808 XXH_PUBLIC_API XXH_errorcode XXH64_update (XXH64_state_t* state_in, const void* input, size_t len)
809 {
810 XXH_endianess endian_detected = (XXH_endianess)XXH_CPU_LITTLE_ENDIAN;
811
812 if ((endian_detected==XXH_littleEndian) || XXH_FORCE_NATIVE_FORMAT)
813 return XXH64_update_endian(state_in, input, len, XXH_littleEndian);
814 else
815 return XXH64_update_endian(state_in, input, len, XXH_bigEndian);
816 }
817
XXH64_digest_endian(const XXH64_state_t * state,XXH_endianess endian)818 XXH_FORCE_INLINE U64 XXH64_digest_endian (const XXH64_state_t* state, XXH_endianess endian)
819 {
820 const BYTE * p = (const BYTE*)state->mem64;
821 const BYTE* const bEnd = (const BYTE*)state->mem64 + state->memsize;
822 U64 h64;
823
824 if (state->total_len >= 32) {
825 U64 const v1 = state->v1;
826 U64 const v2 = state->v2;
827 U64 const v3 = state->v3;
828 U64 const v4 = state->v4;
829
830 h64 = XXH_rotl64(v1, 1) + XXH_rotl64(v2, 7) + XXH_rotl64(v3, 12) + XXH_rotl64(v4, 18);
831 h64 = XXH64_mergeRound(h64, v1);
832 h64 = XXH64_mergeRound(h64, v2);
833 h64 = XXH64_mergeRound(h64, v3);
834 h64 = XXH64_mergeRound(h64, v4);
835 } else {
836 h64 = state->v3 + PRIME64_5;
837 }
838
839 h64 += (U64) state->total_len;
840
841 while (p+8<=bEnd) {
842 U64 const k1 = XXH64_round(0, XXH_readLE64(p, endian));
843 h64 ^= k1;
844 h64 = XXH_rotl64(h64,27) * PRIME64_1 + PRIME64_4;
845 p+=8;
846 }
847
848 if (p+4<=bEnd) {
849 h64 ^= (U64)(XXH_readLE32(p, endian)) * PRIME64_1;
850 h64 = XXH_rotl64(h64, 23) * PRIME64_2 + PRIME64_3;
851 p+=4;
852 }
853
854 while (p<bEnd) {
855 h64 ^= (*p) * PRIME64_5;
856 h64 = XXH_rotl64(h64, 11) * PRIME64_1;
857 p++;
858 }
859
860 h64 ^= h64 >> 33;
861 h64 *= PRIME64_2;
862 h64 ^= h64 >> 29;
863 h64 *= PRIME64_3;
864 h64 ^= h64 >> 32;
865
866 return h64;
867 }
868
XXH64_digest(const XXH64_state_t * state_in)869 XXH_PUBLIC_API unsigned long long XXH64_digest (const XXH64_state_t* state_in)
870 {
871 XXH_endianess endian_detected = (XXH_endianess)XXH_CPU_LITTLE_ENDIAN;
872
873 if ((endian_detected==XXH_littleEndian) || XXH_FORCE_NATIVE_FORMAT)
874 return XXH64_digest_endian(state_in, XXH_littleEndian);
875 else
876 return XXH64_digest_endian(state_in, XXH_bigEndian);
877 }
878
879
880 /*====== Canonical representation ======*/
881
XXH64_canonicalFromHash(XXH64_canonical_t * dst,XXH64_hash_t hash)882 XXH_PUBLIC_API void XXH64_canonicalFromHash(XXH64_canonical_t* dst, XXH64_hash_t hash)
883 {
884 XXH_STATIC_ASSERT(sizeof(XXH64_canonical_t) == sizeof(XXH64_hash_t));
885 if (XXH_CPU_LITTLE_ENDIAN) hash = XXH_swap64(hash);
886 memcpy(dst, &hash, sizeof(*dst));
887 }
888
XXH64_hashFromCanonical(const XXH64_canonical_t * src)889 XXH_PUBLIC_API XXH64_hash_t XXH64_hashFromCanonical(const XXH64_canonical_t* src)
890 {
891 return XXH_readBE64(src);
892 }
893
894 #endif /* XXH_NO_LONG_LONG */
895