From 844ea60273f4e7862150d9c6430cdb15768e2f34 Mon Sep 17 00:00:00 2001 From: Jens Axboe Date: Thu, 20 Feb 2014 13:21:45 -0800 Subject: [PATCH] Add support for the Google xxhash checksumming function It's a lot faster than the other software defined ones, so will be a useful alternative for the cases where crc32c-intel isn't available. Signed-off-by: Jens Axboe --- HOWTO | 4 + crc/test.c | 30 ++++ crc/xxhash.c | 425 +++++++++++++++++++++++++++++++++++++++++++++++++++ crc/xxhash.h | 177 +++++++++++++++++++++ fio.1 | 2 +- options.c | 4 + verify.c | 46 ++++++ verify.h | 4 + 8 files changed, 691 insertions(+), 1 deletion(-) create mode 100644 crc/xxhash.c create mode 100644 crc/xxhash.h diff --git a/HOWTO b/HOWTO index 9830fa1e..bafad93f 100644 --- a/HOWTO +++ b/HOWTO @@ -1105,6 +1105,10 @@ verify=str If writing to a file, fio can verify the file contents crc7 Use a crc7 sum of the data area and store it in the header of each block. + xxhash Use xxhash as the checksum function. Generally + the fastest software checksum that fio + supports. + sha512 Use sha512 as the checksum function. sha256 Use sha256 as the checksum function. diff --git a/crc/test.c b/crc/test.c index 2f6d9ee7..174bea37 100644 --- a/crc/test.c +++ b/crc/test.c @@ -16,6 +16,7 @@ #include "../crc/sha1.h" #include "../crc/sha256.h" #include "../crc/sha512.h" +#include "../crc/xxhash.h" #define CHUNK 131072U #define NR_CHUNKS 2048U @@ -36,6 +37,7 @@ enum { T_SHA1 = 1U << 6, T_SHA256 = 1U << 7, T_SHA512 = 1U << 8, + T_XXHASH = 1U << 9, }; static void randomize_buf(void *buf, unsigned int size, int seed) @@ -233,6 +235,29 @@ static uint64_t t_sha512(void) return ret; } +static uint64_t t_xxhash(void) +{ + void *state; + struct timeval s; + uint64_t ret; + void *buf; + int i; + + state = XXH32_init(0x8989); + + buf = malloc(CHUNK); + randomize_buf(buf, CHUNK, 0x8989); + + fio_gettime(&s, NULL); + for (i = 0; i < NR_CHUNKS; i++) + XXH32_update(state, buf, CHUNK); + + XXH32_digest(state); + ret = utime_since_now(&s); + free(buf); + return ret; +} + static struct test_type t[] = { { .name = "md5", @@ -279,6 +304,11 @@ static struct test_type t[] = { .mask = T_SHA512, .fn = t_sha512, }, + { + .name = "xxhash", + .mask = T_XXHASH, + .fn = t_xxhash, + }, { .name = NULL, }, diff --git a/crc/xxhash.c b/crc/xxhash.c new file mode 100644 index 00000000..6897de36 --- /dev/null +++ b/crc/xxhash.c @@ -0,0 +1,425 @@ +/* +xxHash - Fast Hash algorithm +Copyright (C) 2012-2014, Yann Collet. +BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php) + +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions are +met: + +* Redistributions of source code must retain the above copyright +notice, this list of conditions and the following disclaimer. +* Redistributions in binary form must reproduce the above +copyright notice, this list of conditions and the following disclaimer +in the documentation and/or other materials provided with the +distribution. + +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + +You can contact the author at : +- xxHash source repository : http://code.google.com/p/xxhash/ +*/ + + +//************************************** +// Tuning parameters +//************************************** +// Unaligned memory access is automatically enabled for "common" CPU, such as x86. +// For others CPU, the compiler will be more cautious, and insert extra code to ensure aligned access is respected. +// If you know your target CPU supports unaligned memory access, you want to force this option manually to improve performance. +// You can also enable this parameter if you know your input data will always be aligned (boundaries of 4, for uint32_t). +#if defined(__ARM_FEATURE_UNALIGNED) || defined(__i386) || defined(_M_IX86) || defined(__x86_64__) || defined(_M_X64) +# define XXH_USE_UNALIGNED_ACCESS 1 +#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. +// This option has a very small performance cost (only measurable on small inputs). +// By default, this option is disabled. To enable it, uncomment below define : +//#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. +// It will improve speed for Big-endian CPU. +// This option has no impact on Little_Endian CPU. +#define XXH_FORCE_NATIVE_FORMAT 0 + + +//************************************** +// Includes & Memory related functions +//************************************** +#include "xxhash.h" +// Modify the local functions below should you wish to use some other memory related routines +// for malloc(), free() +#include +void* XXH_malloc(size_t s) { return malloc(s); } +void XXH_free (void* p) { free(p); } +// for memcpy() +#include +void* XXH_memcpy(void* dest, const void* src, size_t size) { return memcpy(dest,src,size); } + + +#if defined(__GNUC__) && !defined(XXH_USE_UNALIGNED_ACCESS) +# define _PACKED __attribute__ ((packed)) +#else +# define _PACKED +#endif + +#if !defined(XXH_USE_UNALIGNED_ACCESS) && !defined(__GNUC__) +# ifdef __IBMC__ +# pragma pack(1) +# else +# pragma pack(push, 1) +# endif +#endif + +typedef struct _uint32_t_S { uint32_t v; } _PACKED uint32_t_S; + +#if !defined(XXH_USE_UNALIGNED_ACCESS) && !defined(__GNUC__) +# pragma pack(pop) +#endif + +#define A32(x) (((uint32_t_S *)(x))->v) + + +//*************************************** +// Compiler-specific Functions and Macros +//*************************************** +#define GCC_VERSION (__GNUC__ * 100 + __GNUC_MINOR__) + +// Note : although _rotl exists for minGW (GCC under windows), performance seems poor +#if defined(_MSC_VER) +# define XXH_rotl32(x,r) _rotl(x,r) +#else +# define XXH_rotl32(x,r) ((x << r) | (x >> (32 - r))) +#endif + +#if defined(_MSC_VER) // Visual Studio +# define XXH_swap32 _byteswap_ulong +#elif GCC_VERSION >= 403 +# define XXH_swap32 __builtin_bswap32 +#else +static inline uint32_t XXH_swap32 (uint32_t x) { + return ((x << 24) & 0xff000000 ) | + ((x << 8) & 0x00ff0000 ) | + ((x >> 8) & 0x0000ff00 ) | + ((x >> 24) & 0x000000ff );} +#endif + + +//************************************** +// Constants +//************************************** +#define PRIME32_1 2654435761U +#define PRIME32_2 2246822519U +#define PRIME32_3 3266489917U +#define PRIME32_4 668265263U +#define PRIME32_5 374761393U + + +//************************************** +// Architecture Macros +//************************************** +typedef enum { XXH_bigEndian=0, XXH_littleEndian=1 } XXH_endianess; +#ifndef XXH_CPU_LITTLE_ENDIAN // It is possible to define XXH_CPU_LITTLE_ENDIAN externally, for example using a compiler switch + static const int one = 1; +# define XXH_CPU_LITTLE_ENDIAN (*(char*)(&one)) +#endif + + +//************************************** +// Macros +//************************************** +#define XXH_STATIC_ASSERT(c) { enum { XXH_static_assert = 1/(!!(c)) }; } // use only *after* variable declarations + + +//**************************** +// Memory reads +//**************************** +typedef enum { XXH_aligned, XXH_unaligned } XXH_alignment; + +uint32_t XXH_readLE32_align(const uint32_t* ptr, XXH_endianess endian, XXH_alignment align) +{ + if (align==XXH_unaligned) + return endian==XXH_littleEndian ? A32(ptr) : XXH_swap32(A32(ptr)); + else + return endian==XXH_littleEndian ? *ptr : XXH_swap32(*ptr); +} + +uint32_t XXH_readLE32(const uint32_t* ptr, XXH_endianess endian) { return XXH_readLE32_align(ptr, endian, XXH_unaligned); } + + +//**************************** +// Simple Hash Functions +//**************************** +uint32_t XXH32_endian_align(const void* input, int len, uint32_t seed, XXH_endianess endian, XXH_alignment align) +{ + const uint8_t *p = (const uint8_t *)input; + const uint8_t * const bEnd = p + len; + uint32_t h32; + +#ifdef XXH_ACCEPT_NULL_INPUT_POINTER + if (p==NULL) { len=0; p=(const uint8_t *)(size_t)16; } +#endif + + if (len>=16) + { + const uint8_t * const limit = bEnd - 16; + uint32_t v1 = seed + PRIME32_1 + PRIME32_2; + uint32_t v2 = seed + PRIME32_2; + uint32_t v3 = seed + 0; + uint32_t v4 = seed - PRIME32_1; + + do + { + v1 += XXH_readLE32_align((const uint32_t*)p, endian, align) * PRIME32_2; v1 = XXH_rotl32(v1, 13); v1 *= PRIME32_1; p+=4; + v2 += XXH_readLE32_align((const uint32_t*)p, endian, align) * PRIME32_2; v2 = XXH_rotl32(v2, 13); v2 *= PRIME32_1; p+=4; + v3 += XXH_readLE32_align((const uint32_t*)p, endian, align) * PRIME32_2; v3 = XXH_rotl32(v3, 13); v3 *= PRIME32_1; p+=4; + v4 += XXH_readLE32_align((const uint32_t*)p, endian, align) * PRIME32_2; v4 = XXH_rotl32(v4, 13); v4 *= PRIME32_1; p+=4; + } while (p<=limit); + + h32 = XXH_rotl32(v1, 1) + XXH_rotl32(v2, 7) + XXH_rotl32(v3, 12) + XXH_rotl32(v4, 18); + } + else + { + h32 = seed + PRIME32_5; + } + + h32 += (uint32_t) len; + + while (p<=bEnd-4) + { + h32 += XXH_readLE32_align((const uint32_t*)p, endian, align) * PRIME32_3; + h32 = XXH_rotl32(h32, 17) * PRIME32_4 ; + p+=4; + } + + while (p> 15; + h32 *= PRIME32_2; + h32 ^= h32 >> 13; + h32 *= PRIME32_3; + h32 ^= h32 >> 16; + + return h32; +} + + +uint32_t XXH32(const void* input, int len, uint32_t seed) +{ +#if 0 + // Simple version, good for code maintenance, but unfortunately slow for small inputs + void* state = XXH32_init(seed); + XXH32_update(state, input, len); + return XXH32_digest(state); +#else + XXH_endianess endian_detected = (XXH_endianess)XXH_CPU_LITTLE_ENDIAN; + +# if !defined(XXH_USE_UNALIGNED_ACCESS) + if ((((size_t)input) & 3)) // Input is aligned, let's leverage the speed advantage + { + if ((endian_detected==XXH_littleEndian) || XXH_FORCE_NATIVE_FORMAT) + return XXH32_endian_align(input, len, seed, XXH_littleEndian, XXH_aligned); + else + return XXH32_endian_align(input, len, seed, XXH_bigEndian, XXH_aligned); + } +# endif + + if ((endian_detected==XXH_littleEndian) || XXH_FORCE_NATIVE_FORMAT) + return XXH32_endian_align(input, len, seed, XXH_littleEndian, XXH_unaligned); + else + return XXH32_endian_align(input, len, seed, XXH_bigEndian, XXH_unaligned); +#endif +} + + +//**************************** +// Advanced Hash Functions +//**************************** + +int XXH32_sizeofState() +{ + XXH_STATIC_ASSERT(XXH32_SIZEOFSTATE >= sizeof(struct XXH_state32_t)); // A compilation error here means XXH32_SIZEOFSTATE is not large enough + return sizeof(struct XXH_state32_t); +} + + +XXH_errorcode XXH32_resetState(void* state_in, uint32_t seed) +{ + struct XXH_state32_t * state = (struct XXH_state32_t *) state_in; + state->seed = seed; + state->v1 = seed + PRIME32_1 + PRIME32_2; + state->v2 = seed + PRIME32_2; + state->v3 = seed + 0; + state->v4 = seed - PRIME32_1; + state->total_len = 0; + state->memsize = 0; + return XXH_OK; +} + + +void* XXH32_init (uint32_t seed) +{ + void* state = XXH_malloc (sizeof(struct XXH_state32_t)); + XXH32_resetState(state, seed); + return state; +} + + +XXH_errorcode XXH32_update_endian (void* state_in, const void* input, int len, XXH_endianess endian) +{ + struct XXH_state32_t * state = (struct XXH_state32_t *) state_in; + const uint8_t *p = (const uint8_t *)input; + const uint8_t * const bEnd = p + len; + +#ifdef XXH_ACCEPT_NULL_INPUT_POINTER + if (input==NULL) return XXH_ERROR; +#endif + + state->total_len += len; + + if (state->memsize + len < 16) // fill in tmp buffer + { + XXH_memcpy(state->memory + state->memsize, input, len); + state->memsize += len; + return XXH_OK; + } + + if (state->memsize) // some data left from previous update + { + XXH_memcpy(state->memory + state->memsize, input, 16-state->memsize); + { + const uint32_t* p32 = (const uint32_t*)state->memory; + state->v1 += XXH_readLE32(p32, endian) * PRIME32_2; state->v1 = XXH_rotl32(state->v1, 13); state->v1 *= PRIME32_1; p32++; + state->v2 += XXH_readLE32(p32, endian) * PRIME32_2; state->v2 = XXH_rotl32(state->v2, 13); state->v2 *= PRIME32_1; p32++; + state->v3 += XXH_readLE32(p32, endian) * PRIME32_2; state->v3 = XXH_rotl32(state->v3, 13); state->v3 *= PRIME32_1; p32++; + state->v4 += XXH_readLE32(p32, endian) * PRIME32_2; state->v4 = XXH_rotl32(state->v4, 13); state->v4 *= PRIME32_1; p32++; + } + p += 16-state->memsize; + state->memsize = 0; + } + + if (p <= bEnd-16) + { + const uint8_t * const limit = bEnd - 16; + uint32_t v1 = state->v1; + uint32_t v2 = state->v2; + uint32_t v3 = state->v3; + uint32_t v4 = state->v4; + + do + { + v1 += XXH_readLE32((const uint32_t*)p, endian) * PRIME32_2; v1 = XXH_rotl32(v1, 13); v1 *= PRIME32_1; p+=4; + v2 += XXH_readLE32((const uint32_t*)p, endian) * PRIME32_2; v2 = XXH_rotl32(v2, 13); v2 *= PRIME32_1; p+=4; + v3 += XXH_readLE32((const uint32_t*)p, endian) * PRIME32_2; v3 = XXH_rotl32(v3, 13); v3 *= PRIME32_1; p+=4; + v4 += XXH_readLE32((const uint32_t*)p, endian) * PRIME32_2; v4 = XXH_rotl32(v4, 13); v4 *= PRIME32_1; p+=4; + } while (p<=limit); + + state->v1 = v1; + state->v2 = v2; + state->v3 = v3; + state->v4 = v4; + } + + if (p < bEnd) + { + XXH_memcpy(state->memory, p, bEnd-p); + state->memsize = (int)(bEnd-p); + } + + return XXH_OK; +} + +XXH_errorcode XXH32_update (void* state_in, const void* input, int len) +{ + XXH_endianess endian_detected = (XXH_endianess)XXH_CPU_LITTLE_ENDIAN; + + if ((endian_detected==XXH_littleEndian) || XXH_FORCE_NATIVE_FORMAT) + return XXH32_update_endian(state_in, input, len, XXH_littleEndian); + else + return XXH32_update_endian(state_in, input, len, XXH_bigEndian); +} + + + +uint32_t XXH32_intermediateDigest_endian (void* state_in, XXH_endianess endian) +{ + struct XXH_state32_t * state = (struct XXH_state32_t *) state_in; + const uint8_t *p = (const uint8_t *)state->memory; + uint8_t * bEnd = (uint8_t *)state->memory + state->memsize; + uint32_t h32; + + if (state->total_len >= 16) + { + h32 = XXH_rotl32(state->v1, 1) + XXH_rotl32(state->v2, 7) + XXH_rotl32(state->v3, 12) + XXH_rotl32(state->v4, 18); + } + else + { + h32 = state->seed + PRIME32_5; + } + + h32 += (uint32_t) state->total_len; + + while (p<=bEnd-4) + { + h32 += XXH_readLE32((const uint32_t*)p, endian) * PRIME32_3; + h32 = XXH_rotl32(h32, 17) * PRIME32_4; + p+=4; + } + + while (p> 15; + h32 *= PRIME32_2; + h32 ^= h32 >> 13; + h32 *= PRIME32_3; + h32 ^= h32 >> 16; + + return h32; +} + + +uint32_t XXH32_intermediateDigest (void* state_in) +{ + XXH_endianess endian_detected = (XXH_endianess)XXH_CPU_LITTLE_ENDIAN; + + if ((endian_detected==XXH_littleEndian) || XXH_FORCE_NATIVE_FORMAT) + return XXH32_intermediateDigest_endian(state_in, XXH_littleEndian); + else + return XXH32_intermediateDigest_endian(state_in, XXH_bigEndian); +} + + +uint32_t XXH32_digest (void* state_in) +{ + uint32_t h32 = XXH32_intermediateDigest(state_in); + + XXH_free(state_in); + + return h32; +} diff --git a/crc/xxhash.h b/crc/xxhash.h new file mode 100644 index 00000000..4c7aafeb --- /dev/null +++ b/crc/xxhash.h @@ -0,0 +1,177 @@ +/* + xxHash - Fast Hash algorithm + Header File + Copyright (C) 2012-2014, Yann Collet. + BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php) + + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions are + met: + + * Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + * Redistributions in binary form must reproduce the above + copyright notice, this list of conditions and the following disclaimer + in the documentation and/or other materials provided with the + distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT + OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, + SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT + LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, + DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY + THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE + OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + + You can contact the author at : + - xxHash source repository : http://code.google.com/p/xxhash/ +*/ + +/* Notice extracted from xxHash homepage : + +xxHash is an extremely fast Hash algorithm, running at RAM speed limits. +It also successfully passes all tests from the SMHasher suite. + +Comparison (single thread, Windows Seven 32 bits, using SMHasher on a Core 2 Duo @3GHz) + +Name Speed Q.Score Author +xxHash 5.4 GB/s 10 +CrapWow 3.2 GB/s 2 Andrew +MumurHash 3a 2.7 GB/s 10 Austin Appleby +SpookyHash 2.0 GB/s 10 Bob Jenkins +SBox 1.4 GB/s 9 Bret Mulvey +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 +MD5-32 0.33 GB/s 10 Ronald L. Rivest +SHA1-32 0.28 GB/s 10 + +Q.Score is a measure of quality of the hash function. +It depends on successfully passing SMHasher test set. +10 is a perfect score. +*/ + +#pragma once + +#if defined (__cplusplus) +extern "C" { +#endif + +#include + +struct XXH_state32_t +{ + uint64_t total_len; + uint32_t seed; + uint32_t v1; + uint32_t v2; + uint32_t v3; + uint32_t v4; + int memsize; + char memory[16]; +}; + +//**************************** +// Type +//**************************** +typedef enum { XXH_OK=0, XXH_ERROR } XXH_errorcode; + + + +//**************************** +// Simple Hash Functions +//**************************** + +unsigned int XXH32 (const void* input, int len, unsigned int seed); + +/* +XXH32() : + Calculate the 32-bits hash of sequence of length "len" stored at memory address "input". + The memory between input & input+len must be valid (allocated and read-accessible). + "seed" can be used to alter the result predictably. + This function successfully passes all SMHasher tests. + Speed on Core 2 Duo @ 3 GHz (single thread, SMHasher benchmark) : 5.4 GB/s + Note that "len" is type "int", which means it is limited to 2^31-1. + If your data is larger, use the advanced functions below. +*/ + + + +//**************************** +// Advanced Hash Functions +//**************************** + +void* XXH32_init (unsigned int seed); +XXH_errorcode XXH32_update (void* state, const void* input, int len); +unsigned int XXH32_digest (void* state); + +/* +These functions calculate the xxhash of an input provided in several small packets, +as opposed to an input provided as a single block. + +It must be started with : +void* XXH32_init() +The function returns a pointer which holds the state of calculation. + +This pointer must be provided as "void* state" parameter for XXH32_update(). +XXH32_update() can be called as many times as necessary. +The user must provide a valid (allocated) input. +The function returns an error code, with 0 meaning OK, and any other value meaning there is an error. +Note that "len" is type "int", which means it is limited to 2^31-1. +If your data is larger, it is recommended to chunk your data into blocks +of size for example 2^30 (1GB) to avoid any "int" overflow issue. + +Finally, you can end the calculation anytime, by using XXH32_digest(). +This function returns the final 32-bits hash. +You must provide the same "void* state" parameter created by XXH32_init(). +Memory will be freed by XXH32_digest(). +*/ + + +int XXH32_sizeofState(); +XXH_errorcode XXH32_resetState(void* state, unsigned int seed); + +#define XXH32_SIZEOFSTATE 48 +typedef struct { long long ll[(XXH32_SIZEOFSTATE+(sizeof(long long)-1))/sizeof(long long)]; } XXH32_stateSpace_t; +/* +These functions allow user application to make its own allocation for state. + +XXH32_sizeofState() is used to know how much space must be allocated for the xxHash 32-bits state. +Note that the state must be aligned to access 'long long' fields. Memory must be allocated and referenced by a pointer. +This pointer must then be provided as 'state' into XXH32_resetState(), which initializes the state. + +For static allocation purposes (such as allocation on stack, or freestanding systems without malloc()), +use the structure XXH32_stateSpace_t, which will ensure that memory space is large enough and correctly aligned to access 'long long' fields. +*/ + + +unsigned int XXH32_intermediateDigest (void* state); +/* +This function does the same as XXH32_digest(), generating a 32-bit hash, +but preserve memory context. +This way, it becomes possible to generate intermediate hashes, and then continue feeding data with XXH32_update(). +To free memory context, use XXH32_digest(), or free(). +*/ + + + +//**************************** +// Deprecated function names +//**************************** +// The following translations are provided to ease code transition +// You are encouraged to no longer this function names +#define XXH32_feed XXH32_update +#define XXH32_result XXH32_digest +#define XXH32_getIntermediateResult XXH32_intermediateDigest + + + +#if defined (__cplusplus) +} +#endif diff --git a/fio.1 b/fio.1 index 4377052f..81b03f70 100644 --- a/fio.1 +++ b/fio.1 @@ -985,7 +985,7 @@ values are: .RS .RS .TP -.B md5 crc16 crc32 crc32c crc32c-intel crc64 crc7 sha256 sha512 sha1 +.B md5 crc16 crc32 crc32c crc32c-intel crc64 crc7 sha256 sha512 sha1 xxhash Store appropriate checksum in the header of each block. crc32c-intel is hardware accelerated SSE4.2 driven, falls back to regular crc32c if not supported by the system. diff --git a/options.c b/options.c index ac101a3f..dc55c985 100644 --- a/options.c +++ b/options.c @@ -2243,6 +2243,10 @@ struct fio_option fio_options[FIO_MAX_OPTS] = { .oval = VERIFY_SHA512, .help = "Use sha512 checksums for verification", }, + { .ival = "xxhash", + .oval = VERIFY_XXHASH, + .help = "Use xxhash checksums for verification", + }, { .ival = "meta", .oval = VERIFY_META, .help = "Use io information", diff --git a/verify.c b/verify.c index 93731228..9eb532a2 100644 --- a/verify.c +++ b/verify.c @@ -23,6 +23,7 @@ #include "crc/sha256.h" #include "crc/sha512.h" #include "crc/sha1.h" +#include "crc/xxhash.h" static void populate_hdr(struct thread_data *td, struct io_u *io_u, struct verify_header *hdr, unsigned int header_num, @@ -172,6 +173,9 @@ static inline unsigned int __hdr_size(int verify_type) case VERIFY_SHA512: len = sizeof(struct vhdr_sha512); break; + case VERIFY_XXHASH: + len = sizeof(struct vhdr_xxhash); + break; case VERIFY_META: len = sizeof(struct vhdr_meta); break; @@ -404,6 +408,30 @@ static int verify_io_u_meta(struct verify_header *hdr, struct vcont *vc) return ret; } +static int verify_io_u_xxhash(struct verify_header *hdr, struct vcont *vc) +{ + void *p = io_u_verify_off(hdr, vc); + struct vhdr_xxhash *vh = hdr_priv(hdr); + uint32_t hash; + void *state; + + dprint(FD_VERIFY, "xxhash verify io_u %p, len %u\n", vc->io_u, hdr->len); + + state = XXH32_init(1); + XXH32_update(state, p, hdr->len - hdr_size(hdr)); + hash = XXH32_digest(state); + + if (vh->hash == hash) + return 0; + + vc->name = "xxhash"; + vc->good_crc = &vh->hash; + vc->bad_crc = &hash; + vc->crc_len = sizeof(hash); + log_verify_failure(hdr, vc); + return EILSEQ; +} + static int verify_io_u_sha512(struct verify_header *hdr, struct vcont *vc) { void *p = io_u_verify_off(hdr, vc); @@ -793,6 +821,9 @@ int verify_io_u(struct thread_data *td, struct io_u *io_u) case VERIFY_SHA512: ret = verify_io_u_sha512(hdr, &vc); break; + case VERIFY_XXHASH: + ret = verify_io_u_xxhash(hdr, &vc); + break; case VERIFY_META: ret = verify_io_u_meta(hdr, &vc); break; @@ -834,6 +865,16 @@ static void fill_meta(struct verify_header *hdr, struct thread_data *td, vh->offset = io_u->offset + header_num * td->o.verify_interval; } +static void fill_xxhash(struct verify_header *hdr, void *p, unsigned int len) +{ + struct vhdr_xxhash *vh = hdr_priv(hdr); + void *state; + + state = XXH32_init(1); + XXH32_update(state, p, len); + vh->hash = XXH32_digest(state); +} + static void fill_sha512(struct verify_header *hdr, void *p, unsigned int len) { struct vhdr_sha512 *vh = hdr_priv(hdr); @@ -973,6 +1014,11 @@ static void populate_hdr(struct thread_data *td, struct io_u *io_u, io_u, hdr->len); fill_sha512(hdr, data, data_len); break; + case VERIFY_XXHASH: + dprint(FD_VERIFY, "fill xxhash io_u %p, len %u\n", + io_u, hdr->len); + fill_xxhash(hdr, data, data_len); + break; case VERIFY_META: dprint(FD_VERIFY, "fill meta io_u %p, len %u\n", io_u, hdr->len); diff --git a/verify.h b/verify.h index 05d7b81b..dba7743a 100644 --- a/verify.h +++ b/verify.h @@ -16,6 +16,7 @@ enum { VERIFY_CRC7, /* crc7 sum data blocks */ VERIFY_SHA256, /* sha256 sum data blocks */ VERIFY_SHA512, /* sha512 sum data blocks */ + VERIFY_XXHASH, /* xxhash sum data blocks */ VERIFY_META, /* block_num, timestamp etc. */ VERIFY_SHA1, /* sha1 sum data blocks */ VERIFY_PATTERN, /* verify specific patterns */ @@ -66,6 +67,9 @@ struct vhdr_meta { unsigned long time_sec; unsigned long time_usec; }; +struct vhdr_xxhash { + uint32_t hash; +}; /* * Verify helpers -- 2.25.1