diff --git a/src/borg/algorithms/xxh64/xxhash.c b/src/borg/algorithms/xxh64/xxhash.c index 0d0b3a529..66e29833a 100644 --- a/src/borg/algorithms/xxh64/xxhash.c +++ b/src/borg/algorithms/xxh64/xxhash.c @@ -33,6 +33,12 @@ */ +/* since xxhash.c can be included (via XXH_INLINE_ALL), + * it's good practice to protect it with guard + * in case of multiples inclusions */ +#ifndef XXHASH_C_01393879 +#define XXHASH_C_01393879 + /* ************************************* * Tuning parameters ***************************************/ @@ -50,38 +56,29 @@ * Prefer these methods in priority order (0 > 1 > 2) */ #ifndef XXH_FORCE_MEMORY_ACCESS /* can be defined externally, on command line for example */ -# 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__) ) +# if !defined(__clang__) && defined(__GNUC__) && defined(__ARM_FEATURE_UNALIGNED) && defined(__ARM_ARCH) && (__ARM_ARCH == 6) # define XXH_FORCE_MEMORY_ACCESS 2 -# elif defined(__INTEL_COMPILER) || \ - (defined(__GNUC__) && ( defined(__ARM_ARCH_7__) || defined(__ARM_ARCH_7A__) || defined(__ARM_ARCH_7R__) || defined(__ARM_ARCH_7M__) || defined(__ARM_ARCH_7S__) )) +# elif !defined(__clang__) && ((defined(__INTEL_COMPILER) && !defined(_WIN32)) || \ + (defined(__GNUC__) && (defined(__ARM_ARCH) && __ARM_ARCH >= 7))) # define XXH_FORCE_MEMORY_ACCESS 1 # endif #endif /*!XXH_ACCEPT_NULL_INPUT_POINTER : - * If the input pointer is a null pointer, xxHash default behavior is to trigger a memory access error, since it is a bad pointer. - * When this option is enabled, xxHash output for null input pointers will be the same as a null-length input. - * By default, this option is disabled. To enable it, uncomment below define : + * If input pointer is NULL, xxHash default behavior is to dereference it, triggering a segfault. + * When this macro is enabled, xxHash actively checks input for null pointer. + * It it is, result for null input pointers is the same as a null-length input. */ -/* #define XXH_ACCEPT_NULL_INPUT_POINTER 1 */ - -/*!XXH_FORCE_NATIVE_FORMAT : - * By default, xxHash library provides endian-independant Hash values, based on little-endian convention. - * Results are therefore identical for little-endian and big-endian CPU. - * This comes at a performance cost for big-endian CPU, since some swapping is required to emulate little-endian format. - * Should endian-independance be of no importance for your application, you may set the #define below to 1, - * to improve speed for Big-endian CPU. - * This option has no impact on Little_Endian CPU. - */ -#ifndef XXH_FORCE_NATIVE_FORMAT /* can be defined externally */ -# define XXH_FORCE_NATIVE_FORMAT 0 +#ifndef XXH_ACCEPT_NULL_INPUT_POINTER /* can be defined externally */ +# define XXH_ACCEPT_NULL_INPUT_POINTER 0 #endif /*!XXH_FORCE_ALIGN_CHECK : * This is a minor performance trick, only useful with lots of very small keys. * It means : check for aligned/unaligned input. - * The check costs one initial branch per hash; set to 0 when the input data - * is guaranteed to be aligned. + * The check costs one initial branch per hash; + * set it to 0 when the input is guaranteed to be aligned, + * or when alignment doesn't matter for performance. */ #ifndef XXH_FORCE_ALIGN_CHECK /* can be defined externally */ # if defined(__i386) || defined(_M_IX86) || defined(__x86_64__) || defined(_M_X64) @@ -91,19 +88,33 @@ # endif #endif +/*!XXH_REROLL: + * Whether to reroll XXH32_finalize, and XXH64_finalize, + * instead of using an unrolled jump table/if statement loop. + * + * This is automatically defined on -Os/-Oz on GCC and Clang. */ +#ifndef XXH_REROLL +# if defined(__OPTIMIZE_SIZE__) +# define XXH_REROLL 1 +# else +# define XXH_REROLL 0 +# endif +#endif /* ************************************* * Includes & Memory related functions ***************************************/ -/* Modify the local functions below should you wish to use some other memory routines */ -/* for malloc(), free() */ +/*! Modify the local functions below should you wish to use some other memory routines +* for malloc(), free() */ #include static void* XXH_malloc(size_t s) { return malloc(s); } static void XXH_free (void* p) { free(p); } -/* for memcpy() */ +/*! and for memcpy() */ #include static void* XXH_memcpy(void* dest, const void* src, size_t size) { return memcpy(dest,src,size); } +#include /* ULLONG_MAX */ + #define XXH_STATIC_LINKING_ONLY #include "xxhash.h" @@ -113,58 +124,82 @@ static void* XXH_memcpy(void* dest, const void* src, size_t size) { return memcp ***************************************/ #ifdef _MSC_VER /* Visual Studio */ # pragma warning(disable : 4127) /* disable: C4127: conditional expression is constant */ -# define FORCE_INLINE static __forceinline +# define XXH_FORCE_INLINE static __forceinline +# define XXH_NO_INLINE static __declspec(noinline) #else -# if defined (__STDC_VERSION__) && __STDC_VERSION__ >= 199901L /* C99 */ +# if defined (__cplusplus) || defined (__STDC_VERSION__) && __STDC_VERSION__ >= 199901L /* C99 */ # ifdef __GNUC__ -# define FORCE_INLINE static inline __attribute__((always_inline)) +# define XXH_FORCE_INLINE static inline __attribute__((always_inline)) +# define XXH_NO_INLINE static __attribute__((noinline)) # else -# define FORCE_INLINE static inline +# define XXH_FORCE_INLINE static inline +# define XXH_NO_INLINE static # endif # else -# define FORCE_INLINE static +# define XXH_FORCE_INLINE static +# define XXH_NO_INLINE static # endif /* __STDC_VERSION__ */ #endif + +/* ************************************* +* Debug +***************************************/ +/* DEBUGLEVEL is expected to be defined externally, + * typically through compiler command line. + * Value must be a number. */ +#ifndef DEBUGLEVEL +# define DEBUGLEVEL 0 +#endif + +#if (DEBUGLEVEL>=1) +# include /* note : can still be disabled with NDEBUG */ +# define XXH_ASSERT(c) assert(c) +#else +# define XXH_ASSERT(c) ((void)0) +#endif + +/* note : use after variable declarations */ +#define XXH_STATIC_ASSERT(c) { enum { XXH_sa = 1/(int)(!!(c)) }; } + + /* ************************************* * Basic Types ***************************************/ -#ifndef MEM_MODULE -# if !defined (__VMS) && (defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) ) -# include - typedef uint8_t BYTE; - typedef uint16_t U16; - typedef uint32_t U32; - typedef int32_t S32; -# else - typedef unsigned char BYTE; - typedef unsigned short U16; - typedef unsigned int U32; - typedef signed int S32; -# endif +#if !defined (__VMS) \ + && (defined (__cplusplus) \ + || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) ) +# include + typedef uint8_t xxh_u8; +#else + typedef unsigned char xxh_u8; #endif +typedef XXH32_hash_t xxh_u32; + + +/* === Memory access === */ #if (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==2)) /* Force direct memory access. Only works on CPU which support unaligned memory access in hardware */ -static U32 XXH_read32(const void* memPtr) { return *(const U32*) memPtr; } +static xxh_u32 XXH_read32(const void* memPtr) { return *(const xxh_u32*) memPtr; } #elif (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==1)) /* __pack instructions are safer, but compiler specific, hence potentially problematic for some compilers */ /* currently only defined for gcc and icc */ -typedef union { U32 u32; } __attribute__((packed)) unalign; -static U32 XXH_read32(const void* ptr) { return ((const unalign*)ptr)->u32; } +typedef union { xxh_u32 u32; } __attribute__((packed)) unalign; +static xxh_u32 XXH_read32(const void* ptr) { return ((const unalign*)ptr)->u32; } #else /* portable and safe solution. Generally efficient. * see : http://stackoverflow.com/a/32095106/646947 */ -static U32 XXH_read32(const void* memPtr) +static xxh_u32 XXH_read32(const void* memPtr) { - U32 val; + xxh_u32 val; memcpy(&val, memPtr, sizeof(val)); return val; } @@ -172,26 +207,58 @@ static U32 XXH_read32(const void* memPtr) #endif /* XXH_FORCE_DIRECT_MEMORY_ACCESS */ +/* === Endianess === */ +typedef enum { XXH_bigEndian=0, XXH_littleEndian=1 } XXH_endianess; + +/* XXH_CPU_LITTLE_ENDIAN can be defined externally, for example on the compiler command line */ +#ifndef XXH_CPU_LITTLE_ENDIAN +# if defined(_WIN32) /* Windows is always little endian */ \ + || defined(__LITTLE_ENDIAN__) \ + || (defined(__BYTE_ORDER__) && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__) +# define XXH_CPU_LITTLE_ENDIAN 1 +# elif defined(__BIG_ENDIAN__) \ + || (defined(__BYTE_ORDER__) && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__) +# define XXH_CPU_LITTLE_ENDIAN 0 +# else +static int XXH_isLittleEndian(void) +{ + const union { xxh_u32 u; xxh_u8 c[4]; } one = { 1 }; /* don't use static : performance detrimental */ + return one.c[0]; +} +# define XXH_CPU_LITTLE_ENDIAN XXH_isLittleEndian() +# endif +#endif + + + + /* **************************************** * Compiler-specific Functions and Macros ******************************************/ -#define GCC_VERSION (__GNUC__ * 100 + __GNUC_MINOR__) +#define XXH_GCC_VERSION (__GNUC__ * 100 + __GNUC_MINOR__) +#ifndef __has_builtin +# define __has_builtin(x) 0 +#endif + +#if !defined(NO_CLANG_BUILTIN) && __has_builtin(__builtin_rotateleft32) && __has_builtin(__builtin_rotateleft64) +# define XXH_rotl32 __builtin_rotateleft32 +# define XXH_rotl64 __builtin_rotateleft64 /* Note : although _rotl exists for minGW (GCC under windows), performance seems poor */ -#if defined(_MSC_VER) +#elif defined(_MSC_VER) # define XXH_rotl32(x,r) _rotl(x,r) # define XXH_rotl64(x,r) _rotl64(x,r) #else -# define XXH_rotl32(x,r) ((x << r) | (x >> (32 - r))) -# define XXH_rotl64(x,r) ((x << r) | (x >> (64 - r))) +# define XXH_rotl32(x,r) (((x) << (r)) | ((x) >> (32 - (r)))) +# define XXH_rotl64(x,r) (((x) << (r)) | ((x) >> (64 - (r)))) #endif #if defined(_MSC_VER) /* Visual Studio */ # define XXH_swap32 _byteswap_ulong -#elif GCC_VERSION >= 403 +#elif XXH_GCC_VERSION >= 403 # define XXH_swap32 __builtin_bswap32 #else -static U32 XXH_swap32 (U32 x) +static xxh_u32 XXH_swap32 (xxh_u32 x) { return ((x << 24) & 0xff000000 ) | ((x << 8) & 0x00ff0000 ) | @@ -201,75 +268,425 @@ static U32 XXH_swap32 (U32 x) #endif -/* ************************************* -* Architecture Macros -***************************************/ -typedef enum { XXH_bigEndian=0, XXH_littleEndian=1 } XXH_endianess; - -/* XXH_CPU_LITTLE_ENDIAN can be defined externally, for example on the compiler command line */ -#ifndef XXH_CPU_LITTLE_ENDIAN - static const int g_one = 1; -# define XXH_CPU_LITTLE_ENDIAN (*(const char*)(&g_one)) -#endif - - /* *************************** * Memory reads *****************************/ typedef enum { XXH_aligned, XXH_unaligned } XXH_alignment; -FORCE_INLINE U32 XXH_readLE32_align(const void* ptr, XXH_endianess endian, XXH_alignment align) +XXH_FORCE_INLINE xxh_u32 XXH_readLE32(const void* ptr) { - if (align==XXH_unaligned) - return endian==XXH_littleEndian ? XXH_read32(ptr) : XXH_swap32(XXH_read32(ptr)); - else - return endian==XXH_littleEndian ? *(const U32*)ptr : XXH_swap32(*(const U32*)ptr); + return XXH_CPU_LITTLE_ENDIAN ? XXH_read32(ptr) : XXH_swap32(XXH_read32(ptr)); } -FORCE_INLINE U32 XXH_readLE32(const void* ptr, XXH_endianess endian) +static xxh_u32 XXH_readBE32(const void* ptr) { - return XXH_readLE32_align(ptr, endian, XXH_unaligned); + return XXH_CPU_LITTLE_ENDIAN ? XXH_swap32(XXH_read32(ptr)) : XXH_read32(ptr); } +XXH_FORCE_INLINE xxh_u32 +XXH_readLE32_align(const void* ptr, XXH_alignment align) +{ + if (align==XXH_unaligned) { + return XXH_readLE32(ptr); + } else { + return XXH_CPU_LITTLE_ENDIAN ? *(const xxh_u32*)ptr : XXH_swap32(*(const xxh_u32*)ptr); + } +} + + /* ************************************* -* Macros +* Misc ***************************************/ -#define XXH_STATIC_ASSERT(c) { enum { XXH_static_assert = 1/(int)(!!(c)) }; } /* use only *after* variable declarations */ XXH_PUBLIC_API unsigned XXH_versionNumber (void) { return XXH_VERSION_NUMBER; } + +/* ******************************************************************* +* 32-bit hash functions +*********************************************************************/ +static const xxh_u32 PRIME32_1 = 0x9E3779B1U; /* 0b10011110001101110111100110110001 */ +static const xxh_u32 PRIME32_2 = 0x85EBCA77U; /* 0b10000101111010111100101001110111 */ +static const xxh_u32 PRIME32_3 = 0xC2B2AE3DU; /* 0b11000010101100101010111000111101 */ +static const xxh_u32 PRIME32_4 = 0x27D4EB2FU; /* 0b00100111110101001110101100101111 */ +static const xxh_u32 PRIME32_5 = 0x165667B1U; /* 0b00010110010101100110011110110001 */ + +static xxh_u32 XXH32_round(xxh_u32 acc, xxh_u32 input) +{ + acc += input * PRIME32_2; + acc = XXH_rotl32(acc, 13); + acc *= PRIME32_1; +#if defined(__GNUC__) && defined(__SSE4_1__) && !defined(XXH_ENABLE_AUTOVECTORIZE) + /* UGLY HACK: + * This inline assembly hack forces acc into a normal register. This is the + * only thing that prevents GCC and Clang from autovectorizing the XXH32 loop + * (pragmas and attributes don't work for some resason) without globally + * disabling SSE4.1. + * + * The reason we want to avoid vectorization is because despite working on + * 4 integers at a time, there are multiple factors slowing XXH32 down on + * SSE4: + * - There's a ridiculous amount of lag from pmulld (10 cycles of latency on newer chips!) + * making it slightly slower to multiply four integers at once compared to four + * integers independently. Even when pmulld was fastest, Sandy/Ivy Bridge, it is + * still not worth it to go into SSE just to multiply unless doing a long operation. + * + * - Four instructions are required to rotate, + * movqda tmp, v // not required with VEX encoding + * pslld tmp, 13 // tmp <<= 13 + * psrld v, 19 // x >>= 19 + * por v, tmp // x |= tmp + * compared to one for scalar: + * roll v, 13 // reliably fast across the board + * shldl v, v, 13 // Sandy Bridge and later prefer this for some reason + * + * - Instruction level parallelism is actually more beneficial here because the + * SIMD actually serializes this operation: While v1 is rotating, v2 can load data, + * while v3 can multiply. SSE forces them to operate together. + * + * How this hack works: + * __asm__("" // Declare an assembly block but don't declare any instructions + * : // However, as an Input/Output Operand, + * "+r" // constrain a read/write operand (+) as a general purpose register (r). + * (acc) // and set acc as the operand + * ); + * + * Because of the 'r', the compiler has promised that seed will be in a + * general purpose register and the '+' says that it will be 'read/write', + * so it has to assume it has changed. It is like volatile without all the + * loads and stores. + * + * Since the argument has to be in a normal register (not an SSE register), + * each time XXH32_round is called, it is impossible to vectorize. */ + __asm__("" : "+r" (acc)); +#endif + return acc; +} + +/* mix all bits */ +static xxh_u32 XXH32_avalanche(xxh_u32 h32) +{ + h32 ^= h32 >> 15; + h32 *= PRIME32_2; + h32 ^= h32 >> 13; + h32 *= PRIME32_3; + h32 ^= h32 >> 16; + return(h32); +} + +#define XXH_get32bits(p) XXH_readLE32_align(p, align) + +static xxh_u32 +XXH32_finalize(xxh_u32 h32, const xxh_u8* ptr, size_t len, XXH_alignment align) +{ +#define PROCESS1 \ + h32 += (*ptr++) * PRIME32_5; \ + h32 = XXH_rotl32(h32, 11) * PRIME32_1 ; + +#define PROCESS4 \ + h32 += XXH_get32bits(ptr) * PRIME32_3; \ + ptr+=4; \ + h32 = XXH_rotl32(h32, 17) * PRIME32_4 ; + + /* Compact rerolled version */ + if (XXH_REROLL) { + len &= 15; + while (len >= 4) { + PROCESS4; + len -= 4; + } + while (len > 0) { + PROCESS1; + --len; + } + return XXH32_avalanche(h32); + } else { + switch(len&15) /* or switch(bEnd - p) */ { + case 12: PROCESS4; + /* fallthrough */ + case 8: PROCESS4; + /* fallthrough */ + case 4: PROCESS4; + return XXH32_avalanche(h32); + + case 13: PROCESS4; + /* fallthrough */ + case 9: PROCESS4; + /* fallthrough */ + case 5: PROCESS4; + PROCESS1; + return XXH32_avalanche(h32); + + case 14: PROCESS4; + /* fallthrough */ + case 10: PROCESS4; + /* fallthrough */ + case 6: PROCESS4; + PROCESS1; + PROCESS1; + return XXH32_avalanche(h32); + + case 15: PROCESS4; + /* fallthrough */ + case 11: PROCESS4; + /* fallthrough */ + case 7: PROCESS4; + /* fallthrough */ + case 3: PROCESS1; + /* fallthrough */ + case 2: PROCESS1; + /* fallthrough */ + case 1: PROCESS1; + /* fallthrough */ + case 0: return XXH32_avalanche(h32); + } + XXH_ASSERT(0); + return h32; /* reaching this point is deemed impossible */ + } +} + +XXH_FORCE_INLINE xxh_u32 +XXH32_endian_align(const xxh_u8* input, size_t len, xxh_u32 seed, XXH_alignment align) +{ + const xxh_u8* bEnd = input + len; + xxh_u32 h32; + +#if defined(XXH_ACCEPT_NULL_INPUT_POINTER) && (XXH_ACCEPT_NULL_INPUT_POINTER>=1) + if (input==NULL) { + len=0; + bEnd=input=(const xxh_u8*)(size_t)16; + } +#endif + + if (len>=16) { + const xxh_u8* const limit = bEnd - 15; + xxh_u32 v1 = seed + PRIME32_1 + PRIME32_2; + xxh_u32 v2 = seed + PRIME32_2; + xxh_u32 v3 = seed + 0; + xxh_u32 v4 = seed - PRIME32_1; + + do { + v1 = XXH32_round(v1, XXH_get32bits(input)); input += 4; + v2 = XXH32_round(v2, XXH_get32bits(input)); input += 4; + v3 = XXH32_round(v3, XXH_get32bits(input)); input += 4; + v4 = XXH32_round(v4, XXH_get32bits(input)); input += 4; + } while (input < limit); + + h32 = XXH_rotl32(v1, 1) + XXH_rotl32(v2, 7) + + XXH_rotl32(v3, 12) + XXH_rotl32(v4, 18); + } else { + h32 = seed + PRIME32_5; + } + + h32 += (xxh_u32)len; + + return XXH32_finalize(h32, input, len&15, align); +} + + +XXH_PUBLIC_API XXH32_hash_t XXH32 (const void* input, size_t len, XXH32_hash_t seed) +{ +#if 0 + /* Simple version, good for code maintenance, but unfortunately slow for small inputs */ + XXH32_state_t state; + XXH32_reset(&state, seed); + XXH32_update(&state, (const xxh_u8*)input, len); + return XXH32_digest(&state); + +#else + + if (XXH_FORCE_ALIGN_CHECK) { + if ((((size_t)input) & 3) == 0) { /* Input is 4-bytes aligned, leverage the speed benefit */ + return XXH32_endian_align((const xxh_u8*)input, len, seed, XXH_aligned); + } } + + return XXH32_endian_align((const xxh_u8*)input, len, seed, XXH_unaligned); +#endif +} + + + +/*====== Hash streaming ======*/ + +XXH_PUBLIC_API XXH32_state_t* XXH32_createState(void) +{ + return (XXH32_state_t*)XXH_malloc(sizeof(XXH32_state_t)); +} +XXH_PUBLIC_API XXH_errorcode XXH32_freeState(XXH32_state_t* statePtr) +{ + XXH_free(statePtr); + return XXH_OK; +} + +XXH_PUBLIC_API void XXH32_copyState(XXH32_state_t* dstState, const XXH32_state_t* srcState) +{ + memcpy(dstState, srcState, sizeof(*dstState)); +} + +XXH_PUBLIC_API XXH_errorcode XXH32_reset(XXH32_state_t* statePtr, XXH32_hash_t seed) +{ + XXH32_state_t state; /* using a local state to memcpy() in order to avoid strict-aliasing warnings */ + memset(&state, 0, sizeof(state)); + state.v1 = seed + PRIME32_1 + PRIME32_2; + state.v2 = seed + PRIME32_2; + state.v3 = seed + 0; + state.v4 = seed - PRIME32_1; + /* do not write into reserved, planned to be removed in a future version */ + memcpy(statePtr, &state, sizeof(state) - sizeof(state.reserved)); + return XXH_OK; +} + + +XXH_PUBLIC_API XXH_errorcode +XXH32_update(XXH32_state_t* state, const void* input, size_t len) +{ + if (input==NULL) +#if defined(XXH_ACCEPT_NULL_INPUT_POINTER) && (XXH_ACCEPT_NULL_INPUT_POINTER>=1) + return XXH_OK; +#else + return XXH_ERROR; +#endif + + { const xxh_u8* p = (const xxh_u8*)input; + const xxh_u8* const bEnd = p + len; + + state->total_len_32 += (XXH32_hash_t)len; + state->large_len |= (XXH32_hash_t)((len>=16) | (state->total_len_32>=16)); + + if (state->memsize + len < 16) { /* fill in tmp buffer */ + XXH_memcpy((xxh_u8*)(state->mem32) + state->memsize, input, len); + state->memsize += (XXH32_hash_t)len; + return XXH_OK; + } + + if (state->memsize) { /* some data left from previous update */ + XXH_memcpy((xxh_u8*)(state->mem32) + state->memsize, input, 16-state->memsize); + { const xxh_u32* p32 = state->mem32; + state->v1 = XXH32_round(state->v1, XXH_readLE32(p32)); p32++; + state->v2 = XXH32_round(state->v2, XXH_readLE32(p32)); p32++; + state->v3 = XXH32_round(state->v3, XXH_readLE32(p32)); p32++; + state->v4 = XXH32_round(state->v4, XXH_readLE32(p32)); + } + p += 16-state->memsize; + state->memsize = 0; + } + + if (p <= bEnd-16) { + const xxh_u8* const limit = bEnd - 16; + xxh_u32 v1 = state->v1; + xxh_u32 v2 = state->v2; + xxh_u32 v3 = state->v3; + xxh_u32 v4 = state->v4; + + do { + v1 = XXH32_round(v1, XXH_readLE32(p)); p+=4; + v2 = XXH32_round(v2, XXH_readLE32(p)); p+=4; + v3 = XXH32_round(v3, XXH_readLE32(p)); p+=4; + v4 = XXH32_round(v4, XXH_readLE32(p)); p+=4; + } while (p<=limit); + + state->v1 = v1; + state->v2 = v2; + state->v3 = v3; + state->v4 = v4; + } + + if (p < bEnd) { + XXH_memcpy(state->mem32, p, (size_t)(bEnd-p)); + state->memsize = (unsigned)(bEnd-p); + } + } + + return XXH_OK; +} + + +XXH_PUBLIC_API XXH32_hash_t XXH32_digest (const XXH32_state_t* state) +{ + xxh_u32 h32; + + if (state->large_len) { + h32 = XXH_rotl32(state->v1, 1) + + XXH_rotl32(state->v2, 7) + + XXH_rotl32(state->v3, 12) + + XXH_rotl32(state->v4, 18); + } else { + h32 = state->v3 /* == seed */ + PRIME32_5; + } + + h32 += state->total_len_32; + + return XXH32_finalize(h32, (const xxh_u8*)state->mem32, state->memsize, XXH_aligned); +} + + +/*====== Canonical representation ======*/ + +/*! Default XXH result types are basic unsigned 32 and 64 bits. +* The canonical representation follows human-readable write convention, aka big-endian (large digits first). +* These functions allow transformation of hash result into and from its canonical format. +* This way, hash values can be written into a file or buffer, remaining comparable across different systems. +*/ + +XXH_PUBLIC_API void XXH32_canonicalFromHash(XXH32_canonical_t* dst, XXH32_hash_t hash) +{ + XXH_STATIC_ASSERT(sizeof(XXH32_canonical_t) == sizeof(XXH32_hash_t)); + if (XXH_CPU_LITTLE_ENDIAN) hash = XXH_swap32(hash); + memcpy(dst, &hash, sizeof(*dst)); +} + +XXH_PUBLIC_API XXH32_hash_t XXH32_hashFromCanonical(const XXH32_canonical_t* src) +{ + return XXH_readBE32(src); +} + + #ifndef XXH_NO_LONG_LONG /* ******************************************************************* -* 64-bits hash functions +* 64-bit hash functions *********************************************************************/ -#define XXH_get32bits(p) XXH_readLE32_align(p, endian, align) - /*====== Memory access ======*/ -#ifndef MEM_MODULE -# define MEM_MODULE -# if !defined (__VMS) && (defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) ) -# include - typedef uint64_t U64; -# else - 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. */ -# endif -#endif +typedef XXH64_hash_t xxh_u64; +/*! XXH_REROLL_XXH64: + * Whether to reroll the XXH64_finalize() loop. + * + * Just like XXH32, we can unroll the XXH64_finalize() loop. This can be a performance gain + * on 64-bit hosts, as only one jump is required. + * + * However, on 32-bit hosts, because arithmetic needs to be done with two 32-bit registers, + * and 64-bit arithmetic needs to be simulated, it isn't beneficial to unroll. The code becomes + * ridiculously large (the largest function in the binary on i386!), and rerolling it saves + * anywhere from 3kB to 20kB. It is also slightly faster because it fits into cache better + * and is more likely to be inlined by the compiler. + * + * If XXH_REROLL is defined, this is ignored and the loop is always rerolled. */ +#ifndef XXH_REROLL_XXH64 +# if (defined(__ILP32__) || defined(_ILP32)) /* ILP32 is often defined on 32-bit GCC family */ \ + || !(defined(__x86_64__) || defined(_M_X64) || defined(_M_AMD64) /* x86-64 */ \ + || defined(_M_ARM64) || defined(__aarch64__) || defined(__arm64__) /* aarch64 */ \ + || defined(__PPC64__) || defined(__PPC64LE__) || defined(__ppc64__) || defined(__powerpc64__) /* ppc64 */ \ + || defined(__mips64__) || defined(__mips64)) /* mips64 */ \ + || (!defined(SIZE_MAX) || SIZE_MAX < ULLONG_MAX) /* check limits */ +# define XXH_REROLL_XXH64 1 +# else +# define XXH_REROLL_XXH64 0 +# endif +#endif /* !defined(XXH_REROLL_XXH64) */ + #if (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==2)) /* Force direct memory access. Only works on CPU which support unaligned memory access in hardware */ -static U64 XXH_read64(const void* memPtr) { return *(const U64*) memPtr; } +static xxh_u64 XXH_read64(const void* memPtr) { return *(const xxh_u64*) memPtr; } #elif (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==1)) /* __pack instructions are safer, but compiler specific, hence potentially problematic for some compilers */ /* currently only defined for gcc and icc */ -typedef union { U32 u32; U64 u64; } __attribute__((packed)) unalign64; - -static U64 XXH_read64(const void* ptr) { return ((const unalign64*)ptr)->u64; } +typedef union { xxh_u32 u32; xxh_u64 u64; } __attribute__((packed)) unalign64; +static xxh_u64 XXH_read64(const void* ptr) { return ((const unalign64*)ptr)->u64; } #else @@ -277,9 +694,9 @@ static U64 XXH_read64(const void* ptr) { return ((const unalign64*)ptr)->u64; } * see : http://stackoverflow.com/a/32095106/646947 */ -static U64 XXH_read64(const void* memPtr) +static xxh_u64 XXH_read64(const void* memPtr) { - U64 val; + xxh_u64 val; memcpy(&val, memPtr, sizeof(val)); return val; } @@ -288,10 +705,10 @@ static U64 XXH_read64(const void* memPtr) #if defined(_MSC_VER) /* Visual Studio */ # define XXH_swap64 _byteswap_uint64 -#elif GCC_VERSION >= 403 +#elif XXH_GCC_VERSION >= 403 # define XXH_swap64 __builtin_bswap64 #else -static U64 XXH_swap64 (U64 x) +static xxh_u64 XXH_swap64 (xxh_u64 x) { return ((x << 56) & 0xff00000000000000ULL) | ((x << 40) & 0x00ff000000000000ULL) | @@ -304,34 +721,35 @@ static U64 XXH_swap64 (U64 x) } #endif -FORCE_INLINE U64 XXH_readLE64_align(const void* ptr, XXH_endianess endian, XXH_alignment align) +XXH_FORCE_INLINE xxh_u64 XXH_readLE64(const void* ptr) { - if (align==XXH_unaligned) - return endian==XXH_littleEndian ? XXH_read64(ptr) : XXH_swap64(XXH_read64(ptr)); - else - return endian==XXH_littleEndian ? *(const U64*)ptr : XXH_swap64(*(const U64*)ptr); + return XXH_CPU_LITTLE_ENDIAN ? XXH_read64(ptr) : XXH_swap64(XXH_read64(ptr)); } -FORCE_INLINE U64 XXH_readLE64(const void* ptr, XXH_endianess endian) -{ - return XXH_readLE64_align(ptr, endian, XXH_unaligned); -} - -static U64 XXH_readBE64(const void* ptr) +static xxh_u64 XXH_readBE64(const void* ptr) { return XXH_CPU_LITTLE_ENDIAN ? XXH_swap64(XXH_read64(ptr)) : XXH_read64(ptr); } +XXH_FORCE_INLINE xxh_u64 +XXH_readLE64_align(const void* ptr, XXH_alignment align) +{ + if (align==XXH_unaligned) + return XXH_readLE64(ptr); + else + return XXH_CPU_LITTLE_ENDIAN ? *(const xxh_u64*)ptr : XXH_swap64(*(const xxh_u64*)ptr); +} + /*====== xxh64 ======*/ -static const U64 PRIME64_1 = 11400714785074694791ULL; -static const U64 PRIME64_2 = 14029467366897019727ULL; -static const U64 PRIME64_3 = 1609587929392839161ULL; -static const U64 PRIME64_4 = 9650029242287828579ULL; -static const U64 PRIME64_5 = 2870177450012600261ULL; +static const xxh_u64 PRIME64_1 = 0x9E3779B185EBCA87ULL; /* 0b1001111000110111011110011011000110000101111010111100101010000111 */ +static const xxh_u64 PRIME64_2 = 0xC2B2AE3D27D4EB4FULL; /* 0b1100001010110010101011100011110100100111110101001110101101001111 */ +static const xxh_u64 PRIME64_3 = 0x165667B19E3779F9ULL; /* 0b0001011001010110011001111011000110011110001101110111100111111001 */ +static const xxh_u64 PRIME64_4 = 0x85EBCA77C2B2AE63ULL; /* 0b1000010111101011110010100111011111000010101100101010111001100011 */ +static const xxh_u64 PRIME64_5 = 0x27D4EB2F165667C5ULL; /* 0b0010011111010100111010110010111100010110010101100110011111000101 */ -static U64 XXH64_round(U64 acc, U64 input) +static xxh_u64 XXH64_round(xxh_u64 acc, xxh_u64 input) { acc += input * PRIME64_2; acc = XXH_rotl64(acc, 31); @@ -339,7 +757,7 @@ static U64 XXH64_round(U64 acc, U64 input) return acc; } -static U64 XXH64_mergeRound(U64 acc, U64 val) +static xxh_u64 XXH64_mergeRound(xxh_u64 acc, xxh_u64 val) { val = XXH64_round(0, val); acc ^= val; @@ -347,33 +765,168 @@ static U64 XXH64_mergeRound(U64 acc, U64 val) return acc; } -FORCE_INLINE U64 XXH64_endian_align(const void* input, size_t len, U64 seed, XXH_endianess endian, XXH_alignment align) +static xxh_u64 XXH64_avalanche(xxh_u64 h64) { - const BYTE* p = (const BYTE*)input; - const BYTE* const bEnd = p + len; - U64 h64; -#define XXH_get64bits(p) XXH_readLE64_align(p, endian, align) + h64 ^= h64 >> 33; + h64 *= PRIME64_2; + h64 ^= h64 >> 29; + h64 *= PRIME64_3; + h64 ^= h64 >> 32; + return h64; +} -#ifdef XXH_ACCEPT_NULL_INPUT_POINTER - if (p==NULL) { + +#define XXH_get64bits(p) XXH_readLE64_align(p, align) + +static xxh_u64 +XXH64_finalize(xxh_u64 h64, const xxh_u8* ptr, size_t len, XXH_alignment align) +{ +#define PROCESS1_64 \ + h64 ^= (*ptr++) * PRIME64_5; \ + h64 = XXH_rotl64(h64, 11) * PRIME64_1; + +#define PROCESS4_64 \ + h64 ^= (xxh_u64)(XXH_get32bits(ptr)) * PRIME64_1; \ + ptr+=4; \ + h64 = XXH_rotl64(h64, 23) * PRIME64_2 + PRIME64_3; + +#define PROCESS8_64 { \ + xxh_u64 const k1 = XXH64_round(0, XXH_get64bits(ptr)); \ + ptr+=8; \ + h64 ^= k1; \ + h64 = XXH_rotl64(h64,27) * PRIME64_1 + PRIME64_4; \ +} + + /* Rerolled version for 32-bit targets is faster and much smaller. */ + if (XXH_REROLL || XXH_REROLL_XXH64) { + len &= 31; + while (len >= 8) { + PROCESS8_64; + len -= 8; + } + if (len >= 4) { + PROCESS4_64; + len -= 4; + } + while (len > 0) { + PROCESS1_64; + --len; + } + return XXH64_avalanche(h64); + } else { + switch(len & 31) { + case 24: PROCESS8_64; + /* fallthrough */ + case 16: PROCESS8_64; + /* fallthrough */ + case 8: PROCESS8_64; + return XXH64_avalanche(h64); + + case 28: PROCESS8_64; + /* fallthrough */ + case 20: PROCESS8_64; + /* fallthrough */ + case 12: PROCESS8_64; + /* fallthrough */ + case 4: PROCESS4_64; + return XXH64_avalanche(h64); + + case 25: PROCESS8_64; + /* fallthrough */ + case 17: PROCESS8_64; + /* fallthrough */ + case 9: PROCESS8_64; + PROCESS1_64; + return XXH64_avalanche(h64); + + case 29: PROCESS8_64; + /* fallthrough */ + case 21: PROCESS8_64; + /* fallthrough */ + case 13: PROCESS8_64; + /* fallthrough */ + case 5: PROCESS4_64; + PROCESS1_64; + return XXH64_avalanche(h64); + + case 26: PROCESS8_64; + /* fallthrough */ + case 18: PROCESS8_64; + /* fallthrough */ + case 10: PROCESS8_64; + PROCESS1_64; + PROCESS1_64; + return XXH64_avalanche(h64); + + case 30: PROCESS8_64; + /* fallthrough */ + case 22: PROCESS8_64; + /* fallthrough */ + case 14: PROCESS8_64; + /* fallthrough */ + case 6: PROCESS4_64; + PROCESS1_64; + PROCESS1_64; + return XXH64_avalanche(h64); + + case 27: PROCESS8_64; + /* fallthrough */ + case 19: PROCESS8_64; + /* fallthrough */ + case 11: PROCESS8_64; + PROCESS1_64; + PROCESS1_64; + PROCESS1_64; + return XXH64_avalanche(h64); + + case 31: PROCESS8_64; + /* fallthrough */ + case 23: PROCESS8_64; + /* fallthrough */ + case 15: PROCESS8_64; + /* fallthrough */ + case 7: PROCESS4_64; + /* fallthrough */ + case 3: PROCESS1_64; + /* fallthrough */ + case 2: PROCESS1_64; + /* fallthrough */ + case 1: PROCESS1_64; + /* fallthrough */ + case 0: return XXH64_avalanche(h64); + } + } + /* impossible to reach */ + XXH_ASSERT(0); + return 0; /* unreachable, but some compilers complain without it */ +} + +XXH_FORCE_INLINE xxh_u64 +XXH64_endian_align(const xxh_u8* input, size_t len, xxh_u64 seed, XXH_alignment align) +{ + const xxh_u8* bEnd = input + len; + xxh_u64 h64; + +#if defined(XXH_ACCEPT_NULL_INPUT_POINTER) && (XXH_ACCEPT_NULL_INPUT_POINTER>=1) + if (input==NULL) { len=0; - bEnd=p=(const BYTE*)(size_t)32; + bEnd=input=(const xxh_u8*)(size_t)32; } #endif if (len>=32) { - const BYTE* const limit = bEnd - 32; - U64 v1 = seed + PRIME64_1 + PRIME64_2; - U64 v2 = seed + PRIME64_2; - U64 v3 = seed + 0; - U64 v4 = seed - PRIME64_1; + const xxh_u8* const limit = bEnd - 32; + xxh_u64 v1 = seed + PRIME64_1 + PRIME64_2; + xxh_u64 v2 = seed + PRIME64_2; + xxh_u64 v3 = seed + 0; + xxh_u64 v4 = seed - PRIME64_1; do { - v1 = XXH64_round(v1, XXH_get64bits(p)); p+=8; - v2 = XXH64_round(v2, XXH_get64bits(p)); p+=8; - v3 = XXH64_round(v3, XXH_get64bits(p)); p+=8; - v4 = XXH64_round(v4, XXH_get64bits(p)); p+=8; - } while (p<=limit); + v1 = XXH64_round(v1, XXH_get64bits(input)); input+=8; + v2 = XXH64_round(v2, XXH_get64bits(input)); input+=8; + v3 = XXH64_round(v3, XXH_get64bits(input)); input+=8; + v4 = XXH64_round(v4, XXH_get64bits(input)); input+=8; + } while (input<=limit); h64 = XXH_rotl64(v1, 1) + XXH_rotl64(v2, 7) + XXH_rotl64(v3, 12) + XXH_rotl64(v4, 18); h64 = XXH64_mergeRound(h64, v1); @@ -385,60 +938,30 @@ FORCE_INLINE U64 XXH64_endian_align(const void* input, size_t len, U64 seed, XXH h64 = seed + PRIME64_5; } - h64 += (U64) len; + h64 += (xxh_u64) len; - while (p+8<=bEnd) { - U64 const k1 = XXH64_round(0, XXH_get64bits(p)); - h64 ^= k1; - h64 = XXH_rotl64(h64,27) * PRIME64_1 + PRIME64_4; - p+=8; - } - - if (p+4<=bEnd) { - h64 ^= (U64)(XXH_get32bits(p)) * PRIME64_1; - h64 = XXH_rotl64(h64, 23) * PRIME64_2 + PRIME64_3; - p+=4; - } - - while (p> 33; - h64 *= PRIME64_2; - h64 ^= h64 >> 29; - h64 *= PRIME64_3; - h64 ^= h64 >> 32; - - return h64; + return XXH64_finalize(h64, input, len, align); } -XXH_PUBLIC_API unsigned long long XXH64 (const void* input, size_t len, unsigned long long seed) +XXH_PUBLIC_API XXH64_hash_t XXH64 (const void* input, size_t len, XXH64_hash_t seed) { #if 0 /* Simple version, good for code maintenance, but unfortunately slow for small inputs */ - XXH64_CREATESTATE_STATIC(state); - XXH64_reset(state, seed); - XXH64_update(state, input, len); - return XXH64_digest(state); + XXH64_state_t state; + XXH64_reset(&state, seed); + XXH64_update(&state, (const xxh_u8*)input, len); + return XXH64_digest(&state); + #else - XXH_endianess endian_detected = (XXH_endianess)XXH_CPU_LITTLE_ENDIAN; if (XXH_FORCE_ALIGN_CHECK) { if ((((size_t)input) & 7)==0) { /* Input is aligned, let's leverage the speed advantage */ - if ((endian_detected==XXH_littleEndian) || XXH_FORCE_NATIVE_FORMAT) - return XXH64_endian_align(input, len, seed, XXH_littleEndian, XXH_aligned); - else - return XXH64_endian_align(input, len, seed, XXH_bigEndian, XXH_aligned); + return XXH64_endian_align((const xxh_u8*)input, len, seed, XXH_aligned); } } - if ((endian_detected==XXH_littleEndian) || XXH_FORCE_NATIVE_FORMAT) - return XXH64_endian_align(input, len, seed, XXH_littleEndian, XXH_unaligned); - else - return XXH64_endian_align(input, len, seed, XXH_bigEndian, XXH_unaligned); + return XXH64_endian_align((const xxh_u8*)input, len, seed, XXH_unaligned); + #endif } @@ -454,99 +977,94 @@ XXH_PUBLIC_API XXH_errorcode XXH64_freeState(XXH64_state_t* statePtr) return XXH_OK; } -XXH_PUBLIC_API void XXH64_copyState(XXH64_state_t* restrict dstState, const XXH64_state_t* restrict srcState) +XXH_PUBLIC_API void XXH64_copyState(XXH64_state_t* dstState, const XXH64_state_t* srcState) { memcpy(dstState, srcState, sizeof(*dstState)); } -XXH_PUBLIC_API XXH_errorcode XXH64_reset(XXH64_state_t* statePtr, unsigned long long seed) +XXH_PUBLIC_API XXH_errorcode XXH64_reset(XXH64_state_t* statePtr, XXH64_hash_t seed) { XXH64_state_t state; /* using a local state to memcpy() in order to avoid strict-aliasing warnings */ - memset(&state, 0, sizeof(state)-8); /* do not write into reserved, for future removal */ + memset(&state, 0, sizeof(state)); state.v1 = seed + PRIME64_1 + PRIME64_2; state.v2 = seed + PRIME64_2; state.v3 = seed + 0; state.v4 = seed - PRIME64_1; - memcpy(statePtr, &state, sizeof(state)); + /* do not write into reserved64, might be removed in a future version */ + memcpy(statePtr, &state, sizeof(state) - sizeof(state.reserved64)); return XXH_OK; } -FORCE_INLINE XXH_errorcode XXH64_update_endian (XXH64_state_t* state, const void* input, size_t len, XXH_endianess endian) +XXH_PUBLIC_API XXH_errorcode +XXH64_update (XXH64_state_t* state, const void* input, size_t len) { - const BYTE* p = (const BYTE*)input; - const BYTE* const bEnd = p + len; - -#ifdef XXH_ACCEPT_NULL_INPUT_POINTER - if (input==NULL) return XXH_ERROR; + if (input==NULL) +#if defined(XXH_ACCEPT_NULL_INPUT_POINTER) && (XXH_ACCEPT_NULL_INPUT_POINTER>=1) + return XXH_OK; +#else + return XXH_ERROR; #endif - state->total_len += len; + { const xxh_u8* p = (const xxh_u8*)input; + const xxh_u8* const bEnd = p + len; - if (state->memsize + len < 32) { /* fill in tmp buffer */ - XXH_memcpy(((BYTE*)state->mem64) + state->memsize, input, len); - state->memsize += (U32)len; - return XXH_OK; - } + state->total_len += len; - if (state->memsize) { /* tmp buffer is full */ - XXH_memcpy(((BYTE*)state->mem64) + state->memsize, input, 32-state->memsize); - state->v1 = XXH64_round(state->v1, XXH_readLE64(state->mem64+0, endian)); - state->v2 = XXH64_round(state->v2, XXH_readLE64(state->mem64+1, endian)); - state->v3 = XXH64_round(state->v3, XXH_readLE64(state->mem64+2, endian)); - state->v4 = XXH64_round(state->v4, XXH_readLE64(state->mem64+3, endian)); - p += 32-state->memsize; - state->memsize = 0; - } + if (state->memsize + len < 32) { /* fill in tmp buffer */ + XXH_memcpy(((xxh_u8*)state->mem64) + state->memsize, input, len); + state->memsize += (xxh_u32)len; + return XXH_OK; + } - if (p+32 <= bEnd) { - const BYTE* const limit = bEnd - 32; - U64 v1 = state->v1; - U64 v2 = state->v2; - U64 v3 = state->v3; - U64 v4 = state->v4; + if (state->memsize) { /* tmp buffer is full */ + XXH_memcpy(((xxh_u8*)state->mem64) + state->memsize, input, 32-state->memsize); + state->v1 = XXH64_round(state->v1, XXH_readLE64(state->mem64+0)); + state->v2 = XXH64_round(state->v2, XXH_readLE64(state->mem64+1)); + state->v3 = XXH64_round(state->v3, XXH_readLE64(state->mem64+2)); + state->v4 = XXH64_round(state->v4, XXH_readLE64(state->mem64+3)); + p += 32-state->memsize; + state->memsize = 0; + } - do { - v1 = XXH64_round(v1, XXH_readLE64(p, endian)); p+=8; - v2 = XXH64_round(v2, XXH_readLE64(p, endian)); p+=8; - v3 = XXH64_round(v3, XXH_readLE64(p, endian)); p+=8; - v4 = XXH64_round(v4, XXH_readLE64(p, endian)); p+=8; - } while (p<=limit); + if (p+32 <= bEnd) { + const xxh_u8* const limit = bEnd - 32; + xxh_u64 v1 = state->v1; + xxh_u64 v2 = state->v2; + xxh_u64 v3 = state->v3; + xxh_u64 v4 = state->v4; - state->v1 = v1; - state->v2 = v2; - state->v3 = v3; - state->v4 = v4; - } + do { + v1 = XXH64_round(v1, XXH_readLE64(p)); p+=8; + v2 = XXH64_round(v2, XXH_readLE64(p)); p+=8; + v3 = XXH64_round(v3, XXH_readLE64(p)); p+=8; + v4 = XXH64_round(v4, XXH_readLE64(p)); p+=8; + } while (p<=limit); - if (p < bEnd) { - XXH_memcpy(state->mem64, p, (size_t)(bEnd-p)); - state->memsize = (unsigned)(bEnd-p); + state->v1 = v1; + state->v2 = v2; + state->v3 = v3; + state->v4 = v4; + } + + if (p < bEnd) { + XXH_memcpy(state->mem64, p, (size_t)(bEnd-p)); + state->memsize = (unsigned)(bEnd-p); + } } return XXH_OK; } -XXH_PUBLIC_API XXH_errorcode XXH64_update (XXH64_state_t* state_in, const void* input, size_t len) -{ - XXH_endianess endian_detected = (XXH_endianess)XXH_CPU_LITTLE_ENDIAN; - if ((endian_detected==XXH_littleEndian) || XXH_FORCE_NATIVE_FORMAT) - return XXH64_update_endian(state_in, input, len, XXH_littleEndian); - else - return XXH64_update_endian(state_in, input, len, XXH_bigEndian); -} - -FORCE_INLINE U64 XXH64_digest_endian (const XXH64_state_t* state, XXH_endianess endian) +XXH_PUBLIC_API XXH64_hash_t XXH64_digest (const XXH64_state_t* state) { - const BYTE * p = (const BYTE*)state->mem64; - const BYTE* const bEnd = (const BYTE*)state->mem64 + state->memsize; - U64 h64; + xxh_u64 h64; if (state->total_len >= 32) { - U64 const v1 = state->v1; - U64 const v2 = state->v2; - U64 const v3 = state->v3; - U64 const v4 = state->v4; + xxh_u64 const v1 = state->v1; + xxh_u64 const v2 = state->v2; + xxh_u64 const v3 = state->v3; + xxh_u64 const v4 = state->v4; h64 = XXH_rotl64(v1, 1) + XXH_rotl64(v2, 7) + XXH_rotl64(v3, 12) + XXH_rotl64(v4, 18); h64 = XXH64_mergeRound(h64, v1); @@ -554,47 +1072,12 @@ FORCE_INLINE U64 XXH64_digest_endian (const XXH64_state_t* state, XXH_endianess h64 = XXH64_mergeRound(h64, v3); h64 = XXH64_mergeRound(h64, v4); } else { - h64 = state->v3 + PRIME64_5; + h64 = state->v3 /*seed*/ + PRIME64_5; } - h64 += (U64) state->total_len; + h64 += (xxh_u64) state->total_len; - while (p+8<=bEnd) { - U64 const k1 = XXH64_round(0, XXH_readLE64(p, endian)); - h64 ^= k1; - h64 = XXH_rotl64(h64,27) * PRIME64_1 + PRIME64_4; - p+=8; - } - - if (p+4<=bEnd) { - h64 ^= (U64)(XXH_readLE32(p, endian)) * PRIME64_1; - h64 = XXH_rotl64(h64, 23) * PRIME64_2 + PRIME64_3; - p+=4; - } - - while (p> 33; - h64 *= PRIME64_2; - h64 ^= h64 >> 29; - h64 *= PRIME64_3; - h64 ^= h64 >> 32; - - return h64; -} - -XXH_PUBLIC_API unsigned long long XXH64_digest (const XXH64_state_t* state_in) -{ - XXH_endianess endian_detected = (XXH_endianess)XXH_CPU_LITTLE_ENDIAN; - - if ((endian_detected==XXH_littleEndian) || XXH_FORCE_NATIVE_FORMAT) - return XXH64_digest_endian(state_in, XXH_littleEndian); - else - return XXH64_digest_endian(state_in, XXH_bigEndian); + return XXH64_finalize(h64, (const xxh_u8*)state->mem64, (size_t)state->total_len, XXH_aligned); } @@ -612,4 +1095,7 @@ XXH_PUBLIC_API XXH64_hash_t XXH64_hashFromCanonical(const XXH64_canonical_t* src return XXH_readBE64(src); } + #endif /* XXH_NO_LONG_LONG */ + +#endif /* XXHASH_C_01393879 */ diff --git a/src/borg/algorithms/xxh64/xxhash.h b/src/borg/algorithms/xxh64/xxhash.h index 5e5d4cf47..60435bed1 100644 --- a/src/borg/algorithms/xxh64/xxhash.h +++ b/src/borg/algorithms/xxh64/xxhash.h @@ -49,16 +49,19 @@ Lookup3 1.2 GB/s 9 Bob Jenkins SuperFastHash 1.2 GB/s 1 Paul Hsieh CityHash64 1.05 GB/s 10 Pike & Alakuijala FNV 0.55 GB/s 5 Fowler, Noll, Vo -CRC32 0.43 GB/s 9 +CRC32 0.43 GB/s † 9 MD5-32 0.33 GB/s 10 Ronald L. Rivest SHA1-32 0.28 GB/s 10 +Note †: other CRC32 implementations can be over 40x faster than SMHasher's: +http://fastcompression.blogspot.com/2019/03/presenting-xxh3.html?showComment=1552696407071#c3490092340461170735 + Q.Score is a measure of quality of the hash function. It depends on successfully passing SMHasher test set. 10 is a perfect score. -A 64-bits version, named XXH64, is available since r35. -It offers much better speed, but for 64-bits applications only. +A 64-bit version, named XXH64, is available since r35. +It offers much better speed, but for 64-bit applications only. Name Speed on 64 bits Speed on 32 bits XXH64 13.8 GB/s 1.9 GB/s XXH32 6.8 GB/s 6.0 GB/s @@ -67,21 +70,11 @@ XXH32 6.8 GB/s 6.0 GB/s #ifndef XXHASH_H_5627135585666179 #define XXHASH_H_5627135585666179 1 -#define XXH_STATIC_LINKING_ONLY - #if defined (__cplusplus) extern "C" { #endif -/* **************************** -* Compiler specifics -******************************/ -#if !(defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L)) /* ! C99 */ -# define restrict /* disable restrict */ -#endif - - /* **************************** * Definitions ******************************/ @@ -90,18 +83,21 @@ typedef enum { XXH_OK=0, XXH_ERROR } XXH_errorcode; /* **************************** -* API modifier -******************************/ -/** XXH_PRIVATE_API -* This is useful to include xxhash functions in `static` mode -* in order to inline them, and remove their symbol from the public list. -* Methodology : -* #define XXH_PRIVATE_API -* #include "xxhash.h" -* `xxhash.c` is automatically included. -* It's not useful to compile and link it as a separate module. -*/ -#ifdef XXH_PRIVATE_API + * API modifier + ******************************/ +/** XXH_INLINE_ALL (and XXH_PRIVATE_API) + * This build macro includes xxhash functions in `static` mode + * in order to inline them, and remove their symbol from the public list. + * Inlining offers great performance improvement on small keys, + * and dramatic ones when length is expressed as a compile-time constant. + * See https://fastcompression.blogspot.com/2018/03/xxhash-for-small-keys-impressive-power.html . + * Methodology : + * #define XXH_INLINE_ALL + * #include "xxhash.h" + * `xxhash.c` is automatically included. + * It's not useful to compile and link it as a separate object. + */ +#if defined(XXH_INLINE_ALL) || defined(XXH_PRIVATE_API) # ifndef XXH_STATIC_LINKING_ONLY # define XXH_STATIC_LINKING_ONLY # endif @@ -112,23 +108,32 @@ typedef enum { XXH_OK=0, XXH_ERROR } XXH_errorcode; # elif defined(_MSC_VER) # define XXH_PUBLIC_API static __inline # else -# define XXH_PUBLIC_API static /* this version may generate warnings for unused static functions; disable the relevant warning */ + /* this version may generate warnings for unused static functions */ +# define XXH_PUBLIC_API static # endif #else -# define XXH_PUBLIC_API /* do nothing */ -#endif /* XXH_PRIVATE_API */ +# if defined(WIN32) && defined(_MSC_VER) && (defined(XXH_IMPORT) || defined(XXH_EXPORT)) +# ifdef XXH_EXPORT +# define XXH_PUBLIC_API __declspec(dllexport) +# elif XXH_IMPORT +# define XXH_PUBLIC_API __declspec(dllimport) +# endif +# else +# define XXH_PUBLIC_API /* do nothing */ +# endif +#endif /* XXH_INLINE_ALL || XXH_PRIVATE_API */ -/*!XXH_NAMESPACE, aka Namespace Emulation : - -If you want to include _and expose_ xxHash functions from within your own library, -but also want to avoid symbol collisions with other libraries which may also include xxHash, - -you can use XXH_NAMESPACE, to automatically prefix any public symbol from xxhash library -with the value of XXH_NAMESPACE (therefore, avoid NULL and numeric values). - -Note that no change is required within the calling program as long as it includes `xxhash.h` : -regular symbol name will be automatically translated by this header. -*/ +/*! XXH_NAMESPACE, aka Namespace Emulation : + * + * If you want to include _and expose_ xxHash functions from within your own library, + * but also want to avoid symbol collisions with other libraries which may also include xxHash, + * + * you can use XXH_NAMESPACE, to automatically prefix any public symbol from xxhash library + * with the value of XXH_NAMESPACE (therefore, avoid NULL and numeric values). + * + * Note that no change is required within the calling program as long as it includes `xxhash.h` : + * regular symbol name will be automatically translated by this header. + */ #ifdef XXH_NAMESPACE # define XXH_CAT(A,B) A##B # define XXH_NAME2(A,B) XXH_CAT(A,B) @@ -158,31 +163,122 @@ regular symbol name will be automatically translated by this header. * Version ***************************************/ #define XXH_VERSION_MAJOR 0 -#define XXH_VERSION_MINOR 6 +#define XXH_VERSION_MINOR 7 #define XXH_VERSION_RELEASE 2 #define XXH_VERSION_NUMBER (XXH_VERSION_MAJOR *100*100 + XXH_VERSION_MINOR *100 + XXH_VERSION_RELEASE) XXH_PUBLIC_API unsigned XXH_versionNumber (void); + +/*-********************************************************************** +* 32-bit hash +************************************************************************/ +#if !defined (__VMS) \ + && (defined (__cplusplus) \ + || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) ) +# include + typedef uint32_t XXH32_hash_t; +#else +# include +# if UINT_MAX == 0xFFFFFFFFUL + typedef unsigned int XXH32_hash_t; +# else +# if ULONG_MAX == 0xFFFFFFFFUL + typedef unsigned long XXH32_hash_t; +# else +# error "unsupported platform : need a 32-bit type" +# endif +# endif +#endif + +/*! XXH32() : + Calculate the 32-bit hash of sequence "length" bytes stored at memory address "input". + The memory between input & input+length must be valid (allocated and read-accessible). + "seed" can be used to alter the result predictably. + Speed on Core 2 Duo @ 3 GHz (single thread, SMHasher benchmark) : 5.4 GB/s */ +XXH_PUBLIC_API XXH32_hash_t XXH32 (const void* input, size_t length, XXH32_hash_t seed); + +/*====== Streaming ======*/ + +/* + * Streaming functions generate the xxHash value from an incrememtal input. + * This method is slower than single-call functions, due to state management. + * For small inputs, prefer `XXH32()` and `XXH64()`, which are better optimized. + * + * XXH state must first be allocated, using XXH*_createState() . + * + * Start a new hash by initializing state with a seed, using XXH*_reset(). + * + * Then, feed the hash state by calling XXH*_update() as many times as necessary. + * The function returns an error code, with 0 meaning OK, and any other value meaning there is an error. + * + * Finally, a hash value can be produced anytime, by using XXH*_digest(). + * This function returns the nn-bits hash as an int or long long. + * + * It's still possible to continue inserting input into the hash state after a digest, + * and generate some new hash values later on, by invoking again XXH*_digest(). + * + * When done, release the state, using XXH*_freeState(). + */ + +typedef struct XXH32_state_s XXH32_state_t; /* incomplete type */ +XXH_PUBLIC_API XXH32_state_t* XXH32_createState(void); +XXH_PUBLIC_API XXH_errorcode XXH32_freeState(XXH32_state_t* statePtr); +XXH_PUBLIC_API void XXH32_copyState(XXH32_state_t* dst_state, const XXH32_state_t* src_state); + +XXH_PUBLIC_API XXH_errorcode XXH32_reset (XXH32_state_t* statePtr, XXH32_hash_t seed); +XXH_PUBLIC_API XXH_errorcode XXH32_update (XXH32_state_t* statePtr, const void* input, size_t length); +XXH_PUBLIC_API XXH32_hash_t XXH32_digest (const XXH32_state_t* statePtr); + +/*====== Canonical representation ======*/ + +/* Default return values from XXH functions are basic unsigned 32 and 64 bits. + * This the simplest and fastest format for further post-processing. + * However, this leaves open the question of what is the order of bytes, + * since little and big endian conventions will write the same number differently. + * + * The canonical representation settles this issue, + * by mandating big-endian convention, + * aka, the same convention as human-readable numbers (large digits first). + * When writing hash values to storage, sending them over a network, or printing them, + * it's highly recommended to use the canonical representation, + * to ensure portability across a wider range of systems, present and future. + * + * The following functions allow transformation of hash values into and from canonical format. + */ + +typedef struct { unsigned char digest[4]; } XXH32_canonical_t; +XXH_PUBLIC_API void XXH32_canonicalFromHash(XXH32_canonical_t* dst, XXH32_hash_t hash); +XXH_PUBLIC_API XXH32_hash_t XXH32_hashFromCanonical(const XXH32_canonical_t* src); + + #ifndef XXH_NO_LONG_LONG /*-********************************************************************** -* 64-bits hash +* 64-bit hash ************************************************************************/ -typedef unsigned long long XXH64_hash_t; +#if !defined (__VMS) \ + && (defined (__cplusplus) \ + || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) ) +# include + typedef uint64_t XXH64_hash_t; +#else + /* the following type must have a width of 64-bit */ + typedef unsigned long long XXH64_hash_t; +#endif /*! XXH64() : - Calculate the 64-bits hash of sequence of length "len" stored at memory address "input". + Calculate the 64-bit hash of sequence of length "len" stored at memory address "input". "seed" can be used to alter the result predictably. - This function runs faster on 64-bits systems, but slower on 32-bits systems (see benchmark). + This function runs faster on 64-bit systems, but slower on 32-bit systems (see benchmark). */ -XXH_PUBLIC_API XXH64_hash_t XXH64 (const void* input, size_t length, unsigned long long seed); +XXH_PUBLIC_API XXH64_hash_t XXH64 (const void* input, size_t length, XXH64_hash_t seed); /*====== Streaming ======*/ typedef struct XXH64_state_s XXH64_state_t; /* incomplete type */ XXH_PUBLIC_API XXH64_state_t* XXH64_createState(void); XXH_PUBLIC_API XXH_errorcode XXH64_freeState(XXH64_state_t* statePtr); -XXH_PUBLIC_API void XXH64_copyState(XXH64_state_t* restrict dst_state, const XXH64_state_t* restrict src_state); +XXH_PUBLIC_API void XXH64_copyState(XXH64_state_t* dst_state, const XXH64_state_t* src_state); -XXH_PUBLIC_API XXH_errorcode XXH64_reset (XXH64_state_t* statePtr, unsigned long long seed); +XXH_PUBLIC_API XXH_errorcode XXH64_reset (XXH64_state_t* statePtr, XXH64_hash_t seed); XXH_PUBLIC_API XXH_errorcode XXH64_update (XXH64_state_t* statePtr, const void* input, size_t length); XXH_PUBLIC_API XXH64_hash_t XXH64_digest (const XXH64_state_t* statePtr); @@ -190,50 +286,296 @@ XXH_PUBLIC_API XXH64_hash_t XXH64_digest (const XXH64_state_t* statePtr); typedef struct { unsigned char digest[8]; } XXH64_canonical_t; XXH_PUBLIC_API void XXH64_canonicalFromHash(XXH64_canonical_t* dst, XXH64_hash_t hash); XXH_PUBLIC_API XXH64_hash_t XXH64_hashFromCanonical(const XXH64_canonical_t* src); + + #endif /* XXH_NO_LONG_LONG */ + #ifdef XXH_STATIC_LINKING_ONLY /* ================================================================================================ - This section contains definitions which are not guaranteed to remain stable. + This section contains declarations which are not guaranteed to remain stable. They may change in future versions, becoming incompatible with a different version of the library. - They shall only be used with static linking. - Never use these definitions in association with dynamic linking ! + These declarations should only be used with static linking. + Never use them in association with dynamic linking ! =================================================================================================== */ -/* These definitions are only meant to allow allocation of XXH state - statically, on stack, or in a struct for example. - Do not use members directly. */ +/* These definitions are only present to allow + * static allocation of XXH state, on stack or in a struct for example. + * Never **ever** use members directly. */ - struct XXH32_state_s { - unsigned total_len_32; - unsigned large_len; - unsigned v1; - unsigned v2; - unsigned v3; - unsigned v4; - unsigned mem32[4]; /* buffer defined as U32 for alignment */ - unsigned memsize; - unsigned reserved; /* never read nor write, will be removed in a future version */ - }; /* typedef'd to XXH32_state_t */ +struct XXH32_state_s { + XXH32_hash_t total_len_32; + XXH32_hash_t large_len; + XXH32_hash_t v1; + XXH32_hash_t v2; + XXH32_hash_t v3; + XXH32_hash_t v4; + XXH32_hash_t mem32[4]; + XXH32_hash_t memsize; + XXH32_hash_t reserved; /* never read nor write, might be removed in a future version */ +}; /* typedef'd to XXH32_state_t */ +#ifndef XXH_NO_LONG_LONG /* remove 64-bit support */ +struct XXH64_state_s { + XXH64_hash_t total_len; + XXH64_hash_t v1; + XXH64_hash_t v2; + XXH64_hash_t v3; + XXH64_hash_t v4; + XXH64_hash_t mem64[4]; + XXH32_hash_t memsize; + XXH32_hash_t reserved32; /* required for padding anyway */ + XXH64_hash_t reserved64; /* never read nor write, might be removed in a future version */ +}; /* typedef'd to XXH64_state_t */ +#endif /* XXH_NO_LONG_LONG */ + + +/*-********************************************************************** +* XXH3 +* New experimental hash +************************************************************************/ #ifndef XXH_NO_LONG_LONG - struct XXH64_state_s { - unsigned long long total_len; - unsigned long long v1; - unsigned long long v2; - unsigned long long v3; - unsigned long long v4; - unsigned long long mem64[4]; /* buffer defined as U64 for alignment */ - unsigned memsize; - unsigned reserved[2]; /* never read nor write, will be removed in a future version */ - }; /* typedef'd to XXH64_state_t */ + + +/* ============================================ + * XXH3 is a new hash algorithm, + * featuring improved speed performance for both small and large inputs. + * See full speed analysis at : http://fastcompression.blogspot.com/2019/03/presenting-xxh3.html + * In general, expect XXH3 to run about ~2x faster on large inputs, + * and >3x faster on small ones, though exact differences depend on platform. + * + * The algorithm is portable, will generate the same hash on all platforms. + * It benefits greatly from vectorization units, but does not require it. + * + * XXH3 offers 2 variants, _64bits and _128bits. + * When only 64 bits are needed, prefer calling the _64bits variant : + * it reduces the amount of mixing, resulting in faster speed on small inputs. + * It's also generally simpler to manipulate a scalar return type than a struct. + * + * The XXH3 algorithm is still considered experimental. + * Produced results can still change between versions. + * Results produced by v0.7.x are not comparable with results from v0.7.y . + * It's nonetheless possible to use XXH3 for ephemeral data (local sessions), + * but avoid storing values in long-term storage for later reads. + * + * The API supports one-shot hashing, streaming mode, and custom secrets. + * + * There are still a number of opened questions that community can influence during the experimental period. + * I'm trying to list a few of them below, though don't consider this list as complete. + * + * - 128-bits output type : currently defined as a structure of two 64-bits fields. + * That's because 128-bit values do not exist in C standard. + * Note that it means that, at byte level, result is not identical depending on endianess. + * However, at field level, they are identical on all platforms. + * The canonical representation solves the issue of identical byte-level representation across platforms, + * which is necessary for serialization. + * Q1 : Would there be a better representation for a 128-bit hash result ? + * Q2 : Are the names of the inner 64-bit fields important ? Should they be changed ? + * + * - Prototype XXH128() : XXH128() uses the same arguments as XXH64(), for consistency. + * It means it maps to XXH3_128bits_withSeed(). + * This variant is slightly slower than XXH3_128bits(), + * because the seed is now part of the algorithm, and can't be simplified. + * Is that a good idea ? + * + * - Seed type for XXH128() : currently, it's a single 64-bit value, like the 64-bit variant. + * It could be argued that it's more logical to offer a 128-bit seed input parameter for a 128-bit hash. + * But 128-bit seed is more difficult to use, since it requires to pass a structure instead of a scalar value. + * Such a variant could either replace current one, or become an additional one. + * Farmhash, for example, offers both variants (the 128-bits seed variant is called `doubleSeed`). + * Follow up question : if both 64-bit and 128-bit seeds are allowed, which variant should be called XXH128 ? + * + * - Result for len==0 : Currently, the result of hashing a zero-length input is always `0`. + * It seems okay as a return value when using "default" secret and seed. + * But is it still fine to return `0` when secret or seed are non-default ? + * Are there use cases which could depend on generating a different hash result for zero-length input when the secret is different ? + * + * - Consistency (1) : Streaming XXH128 uses an XXH3 state, which is the same state as XXH3_64bits(). + * It means a 128bit streaming loop must invoke the following symbols : + * XXH3_createState(), XXH3_128bits_reset(), XXH3_128bits_update() (loop), XXH3_128bits_digest(), XXH3_freeState(). + * Is that consistent enough ? + * + * - Consistency (2) : The canonical representation of `XXH3_64bits` is provided by existing functions + * XXH64_canonicalFromHash(), and reverse operation XXH64_hashFromCanonical(). + * As a mirror, canonical functions for XXH128_hash_t results generated by `XXH3_128bits` + * are XXH128_canonicalFromHash() and XXH128_hashFromCanonical(). + * Which means, `XXH3` doesn't appear in the names, because canonical functions operate on a type, + * independently of which algorithm was used to generate that type. + * Is that consistent enough ? + */ + +#ifdef XXH_NAMESPACE +# define XXH3_64bits XXH_NAME2(XXH_NAMESPACE, XXH3_64bits) +# define XXH3_64bits_withSecret XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_withSecret) +# define XXH3_64bits_withSeed XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_withSeed) + +# define XXH3_createState XXH_NAME2(XXH_NAMESPACE, XXH3_createState) +# define XXH3_freeState XXH_NAME2(XXH_NAMESPACE, XXH3_freeState) +# define XXH3_copyState XXH_NAME2(XXH_NAMESPACE, XXH3_copyState) + +# define XXH3_64bits_reset XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_reset) +# define XXH3_64bits_reset_withSeed XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_reset_withSeed) +# define XXH3_64bits_reset_withSecret XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_reset_withSecret) +# define XXH3_64bits_update XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_update) +# define XXH3_64bits_digest XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_digest) #endif -# ifdef XXH_PRIVATE_API -# include "xxhash.c" /* include xxhash function bodies as `static`, for inlining */ -# endif +/* XXH3_64bits() : + * default 64-bit variant, using default secret and default seed of 0. + * It's the fastest variant. */ +XXH_PUBLIC_API XXH64_hash_t XXH3_64bits(const void* data, size_t len); + +/* XXH3_64bits_withSecret() : + * It's possible to provide any blob of bytes as a "secret" to generate the hash. + * This makes it more difficult for an external actor to prepare an intentional collision. + * The secret *must* be large enough (>= XXH3_SECRET_SIZE_MIN). + * It should consist of random bytes. + * Avoid repeating same character, or sequences of bytes, + * and especially avoid swathes of \0. + * Failure to respect these conditions will result in a poor quality hash. + */ +#define XXH3_SECRET_SIZE_MIN 136 +XXH_PUBLIC_API XXH64_hash_t XXH3_64bits_withSecret(const void* data, size_t len, const void* secret, size_t secretSize); + +/* XXH3_64bits_withSeed() : + * This variant generates on the fly a custom secret, + * based on the default secret, altered using the `seed` value. + * While this operation is decently fast, note that it's not completely free. + * note : seed==0 produces same results as XXH3_64bits() */ +XXH_PUBLIC_API XXH64_hash_t XXH3_64bits_withSeed(const void* data, size_t len, XXH64_hash_t seed); + + +/* streaming 64-bit */ + +#if defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 201112L) /* C11+ */ +# include +# define XXH_ALIGN(n) alignas(n) +#elif defined(__GNUC__) +# define XXH_ALIGN(n) __attribute__ ((aligned(n))) +#elif defined(_MSC_VER) +# define XXH_ALIGN(n) __declspec(align(n)) +#else +# define XXH_ALIGN(n) /* disabled */ +#endif + +typedef struct XXH3_state_s XXH3_state_t; + +#define XXH3_SECRET_DEFAULT_SIZE 192 /* minimum XXH3_SECRET_SIZE_MIN */ +#define XXH3_INTERNALBUFFER_SIZE 256 +struct XXH3_state_s { + XXH_ALIGN(64) XXH64_hash_t acc[8]; + XXH_ALIGN(64) unsigned char customSecret[XXH3_SECRET_DEFAULT_SIZE]; /* used to store a custom secret generated from the seed. Makes state larger. Design might change */ + XXH_ALIGN(64) unsigned char buffer[XXH3_INTERNALBUFFER_SIZE]; + XXH32_hash_t bufferedSize; + XXH32_hash_t nbStripesPerBlock; + XXH32_hash_t nbStripesSoFar; + XXH32_hash_t secretLimit; + XXH32_hash_t reserved32; + XXH32_hash_t reserved32_2; + XXH64_hash_t totalLen; + XXH64_hash_t seed; + XXH64_hash_t reserved64; + const unsigned char* secret; /* note : there is some padding after, due to alignment on 64 bytes */ +}; /* typedef'd to XXH3_state_t */ + +/* Streaming requires state maintenance. + * This operation costs memory and cpu. + * As a consequence, streaming is slower than one-shot hashing. + * For better performance, prefer using one-shot functions whenever possible. */ + +XXH_PUBLIC_API XXH3_state_t* XXH3_createState(void); +XXH_PUBLIC_API XXH_errorcode XXH3_freeState(XXH3_state_t* statePtr); +XXH_PUBLIC_API void XXH3_copyState(XXH3_state_t* dst_state, const XXH3_state_t* src_state); + + +/* XXH3_64bits_reset() : + * initialize with default parameters. + * result will be equivalent to `XXH3_64bits()`. */ +XXH_PUBLIC_API XXH_errorcode XXH3_64bits_reset(XXH3_state_t* statePtr); +/* XXH3_64bits_reset_withSeed() : + * generate a custom secret from `seed`, and store it into state. + * digest will be equivalent to `XXH3_64bits_withSeed()`. */ +XXH_PUBLIC_API XXH_errorcode XXH3_64bits_reset_withSeed(XXH3_state_t* statePtr, XXH64_hash_t seed); +/* XXH3_64bits_reset_withSecret() : + * `secret` is referenced, and must outlive the hash streaming session. + * secretSize must be >= XXH3_SECRET_SIZE_MIN. + */ +XXH_PUBLIC_API XXH_errorcode XXH3_64bits_reset_withSecret(XXH3_state_t* statePtr, const void* secret, size_t secretSize); + +XXH_PUBLIC_API XXH_errorcode XXH3_64bits_update (XXH3_state_t* statePtr, const void* input, size_t length); +XXH_PUBLIC_API XXH64_hash_t XXH3_64bits_digest (const XXH3_state_t* statePtr); + + +/* 128-bit */ + +#ifdef XXH_NAMESPACE +# define XXH128 XXH_NAME2(XXH_NAMESPACE, XXH128) +# define XXH3_128bits XXH_NAME2(XXH_NAMESPACE, XXH3_128bits) +# define XXH3_128bits_withSeed XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_withSeed) +# define XXH3_128bits_withSecret XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_withSecret) + +# define XXH3_128bits_reset XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_reset) +# define XXH3_128bits_reset_withSeed XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_reset_withSeed) +# define XXH3_128bits_reset_withSecret XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_reset_withSecret) +# define XXH3_128bits_update XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_update) +# define XXH3_128bits_digest XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_digest) + +# define XXH128_isEqual XXH_NAME2(XXH_NAMESPACE, XXH128_isEqual) +# define XXH128_cmp XXH_NAME2(XXH_NAMESPACE, XXH128_cmp) +# define XXH128_canonicalFromHash XXH_NAME2(XXH_NAMESPACE, XXH128_canonicalFromHash) +# define XXH128_hashFromCanonical XXH_NAME2(XXH_NAMESPACE, XXH128_hashFromCanonical) +#endif + +typedef struct { + XXH64_hash_t low64; + XXH64_hash_t high64; +} XXH128_hash_t; + +XXH_PUBLIC_API XXH128_hash_t XXH128(const void* data, size_t len, XXH64_hash_t seed); +XXH_PUBLIC_API XXH128_hash_t XXH3_128bits(const void* data, size_t len); +XXH_PUBLIC_API XXH128_hash_t XXH3_128bits_withSeed(const void* data, size_t len, XXH64_hash_t seed); /* == XXH128() */ +XXH_PUBLIC_API XXH128_hash_t XXH3_128bits_withSecret(const void* data, size_t len, const void* secret, size_t secretSize); + +XXH_PUBLIC_API XXH_errorcode XXH3_128bits_reset(XXH3_state_t* statePtr); +XXH_PUBLIC_API XXH_errorcode XXH3_128bits_reset_withSeed(XXH3_state_t* statePtr, XXH64_hash_t seed); +XXH_PUBLIC_API XXH_errorcode XXH3_128bits_reset_withSecret(XXH3_state_t* statePtr, const void* secret, size_t secretSize); + +XXH_PUBLIC_API XXH_errorcode XXH3_128bits_update (XXH3_state_t* statePtr, const void* input, size_t length); +XXH_PUBLIC_API XXH128_hash_t XXH3_128bits_digest (const XXH3_state_t* statePtr); + + +/* Note : for better performance, following functions can be inlined, + * using XXH_INLINE_ALL */ + +/* return : 1 is equal, 0 if different */ +XXH_PUBLIC_API int XXH128_isEqual(XXH128_hash_t h1, XXH128_hash_t h2); + +/* This comparator is compatible with stdlib's qsort(). + * return : >0 if *h128_1 > *h128_2 + * <0 if *h128_1 < *h128_2 + * =0 if *h128_1 == *h128_2 */ +XXH_PUBLIC_API int XXH128_cmp(const void* h128_1, const void* h128_2); + + +/*====== Canonical representation ======*/ +typedef struct { unsigned char digest[16]; } XXH128_canonical_t; +XXH_PUBLIC_API void XXH128_canonicalFromHash(XXH128_canonical_t* dst, XXH128_hash_t hash); +XXH_PUBLIC_API XXH128_hash_t XXH128_hashFromCanonical(const XXH128_canonical_t* src); + + +#endif /* XXH_NO_LONG_LONG */ + + +/*-********************************************************************** +* XXH_INLINE_ALL +************************************************************************/ +#if defined(XXH_INLINE_ALL) || defined(XXH_PRIVATE_API) +# include "xxhash.c" /* include xxhash function bodies as `static`, for inlining */ +#endif + + #endif /* XXH_STATIC_LINKING_ONLY */