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 <axboe@fb.com>

diff --git a/HOWTO b/HOWTO
index 9830fa1ebd9510f693c3978c70c5c9121d7eabdd..bafad93fbdacc050d7c3b694556859af7b3cd9eb 100644 (file)
--- 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 2f6d9ee7f430c1a05a5d00c328b8a35e33e6fc2d..174bea375136f01cefe1aa60e560bce04de875ae 100644 (file)
--- 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 (file)
index 0000000..6897de3
--- /dev/null
+++ b/crc/xxhash.c
@@ -0,0 +1,425 @@
+/*\r
+xxHash - Fast Hash algorithm\r
+Copyright (C) 2012-2014, Yann Collet.\r
+BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)\r
+\r
+Redistribution and use in source and binary forms, with or without\r
+modification, are permitted provided that the following conditions are\r
+met:\r
+\r
+* Redistributions of source code must retain the above copyright\r
+notice, this list of conditions and the following disclaimer.\r
+* Redistributions in binary form must reproduce the above\r
+copyright notice, this list of conditions and the following disclaimer\r
+in the documentation and/or other materials provided with the\r
+distribution.\r
+\r
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS\r
+"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT\r
+LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR\r
+A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT\r
+OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,\r
+SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT\r
+LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,\r
+DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY\r
+THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT\r
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE\r
+OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.\r
+\r
+You can contact the author at :\r
+- xxHash source repository : http://code.google.com/p/xxhash/\r
+*/\r
+\r
+\r
+//**************************************\r
+// Tuning parameters\r
+//**************************************\r
+// Unaligned memory access is automatically enabled for "common" CPU, such as x86.\r
+// For others CPU, the compiler will be more cautious, and insert extra code to ensure aligned access is respected.\r
+// If you know your target CPU supports unaligned memory access, you want to force this option manually to improve performance.\r
+// You can also enable this parameter if you know your input data will always be aligned (boundaries of 4, for uint32_t).\r
+#if defined(__ARM_FEATURE_UNALIGNED) || defined(__i386) || defined(_M_IX86) || defined(__x86_64__) || defined(_M_X64)\r
+#  define XXH_USE_UNALIGNED_ACCESS 1\r
+#endif\r
+\r
+// XXH_ACCEPT_NULL_INPUT_POINTER :\r
+// If the input pointer is a null pointer, xxHash default behavior is to trigger a memory access error, since it is a bad pointer.\r
+// When this option is enabled, xxHash output for null input pointers will be the same as a null-length input.\r
+// This option has a very small performance cost (only measurable on small inputs).\r
+// By default, this option is disabled. To enable it, uncomment below define :\r
+//#define XXH_ACCEPT_NULL_INPUT_POINTER 1\r
+\r
+// XXH_FORCE_NATIVE_FORMAT :\r
+// By default, xxHash library provides endian-independant Hash values, based on little-endian convention.\r
+// Results are therefore identical for little-endian and big-endian CPU.\r
+// This comes at a performance cost for big-endian CPU, since some swapping is required to emulate little-endian format.\r
+// Should endian-independance be of no importance for your application, you may set the #define below to 1.\r
+// It will improve speed for Big-endian CPU.\r
+// This option has no impact on Little_Endian CPU.\r
+#define XXH_FORCE_NATIVE_FORMAT 0\r
+\r
+\r
+//**************************************\r
+// Includes & Memory related functions\r
+//**************************************\r
+#include "xxhash.h"\r
+// Modify the local functions below should you wish to use some other memory related routines\r
+// for malloc(), free()\r
+#include <stdlib.h>\r
+void* XXH_malloc(size_t s) { return malloc(s); }\r
+void  XXH_free  (void* p)  { free(p); }\r
+// for memcpy()\r
+#include <string.h>\r
+void* XXH_memcpy(void* dest, const void* src, size_t size) { return memcpy(dest,src,size); }\r
+\r
+\r
+#if defined(__GNUC__)  && !defined(XXH_USE_UNALIGNED_ACCESS)\r
+#  define _PACKED __attribute__ ((packed))\r
+#else\r
+#  define _PACKED\r
+#endif\r
+\r
+#if !defined(XXH_USE_UNALIGNED_ACCESS) && !defined(__GNUC__)\r
+#  ifdef __IBMC__\r
+#    pragma pack(1)\r
+#  else\r
+#    pragma pack(push, 1)\r
+#  endif\r
+#endif\r
+\r
+typedef struct _uint32_t_S { uint32_t v; } _PACKED uint32_t_S;\r
+\r
+#if !defined(XXH_USE_UNALIGNED_ACCESS) && !defined(__GNUC__)\r
+#  pragma pack(pop)\r
+#endif\r
+\r
+#define A32(x) (((uint32_t_S *)(x))->v)\r
+\r
+\r
+//***************************************\r
+// Compiler-specific Functions and Macros\r
+//***************************************\r
+#define GCC_VERSION (__GNUC__ * 100 + __GNUC_MINOR__)\r
+\r
+// Note : although _rotl exists for minGW (GCC under windows), performance seems poor\r
+#if defined(_MSC_VER)\r
+#  define XXH_rotl32(x,r) _rotl(x,r)\r
+#else\r
+#  define XXH_rotl32(x,r) ((x << r) | (x >> (32 - r)))\r
+#endif\r
+\r
+#if defined(_MSC_VER)     // Visual Studio\r
+#  define XXH_swap32 _byteswap_ulong\r
+#elif GCC_VERSION >= 403\r
+#  define XXH_swap32 __builtin_bswap32\r
+#else\r
+static inline uint32_t XXH_swap32 (uint32_t x) {\r
+    return  ((x << 24) & 0xff000000 ) |\r
+        ((x <<  8) & 0x00ff0000 ) |\r
+        ((x >>  8) & 0x0000ff00 ) |\r
+        ((x >> 24) & 0x000000ff );}\r
+#endif\r
+\r
+\r
+//**************************************\r
+// Constants\r
+//**************************************\r
+#define PRIME32_1   2654435761U\r
+#define PRIME32_2   2246822519U\r
+#define PRIME32_3   3266489917U\r
+#define PRIME32_4    668265263U\r
+#define PRIME32_5    374761393U\r
+\r
+\r
+//**************************************\r
+// Architecture Macros\r
+//**************************************\r
+typedef enum { XXH_bigEndian=0, XXH_littleEndian=1 } XXH_endianess;\r
+#ifndef XXH_CPU_LITTLE_ENDIAN   // It is possible to define XXH_CPU_LITTLE_ENDIAN externally, for example using a compiler switch\r
+    static const int one = 1;\r
+#   define XXH_CPU_LITTLE_ENDIAN   (*(char*)(&one))\r
+#endif\r
+\r
+\r
+//**************************************\r
+// Macros\r
+//**************************************\r
+#define XXH_STATIC_ASSERT(c)   { enum { XXH_static_assert = 1/(!!(c)) }; }    // use only *after* variable declarations\r
+\r
+\r
+//****************************\r
+// Memory reads\r
+//****************************\r
+typedef enum { XXH_aligned, XXH_unaligned } XXH_alignment;\r
+\r
+uint32_t XXH_readLE32_align(const uint32_t* ptr, XXH_endianess endian, XXH_alignment align)\r
+{ \r
+    if (align==XXH_unaligned)\r
+        return endian==XXH_littleEndian ? A32(ptr) : XXH_swap32(A32(ptr)); \r
+    else\r
+        return endian==XXH_littleEndian ? *ptr : XXH_swap32(*ptr); \r
+}\r
+\r
+uint32_t XXH_readLE32(const uint32_t* ptr, XXH_endianess endian) { return XXH_readLE32_align(ptr, endian, XXH_unaligned); }\r
+\r
+\r
+//****************************\r
+// Simple Hash Functions\r
+//****************************\r
+uint32_t XXH32_endian_align(const void* input, int len, uint32_t seed, XXH_endianess endian, XXH_alignment align)\r
+{\r
+    const uint8_t *p = (const uint8_t *)input;\r
+    const uint8_t * const bEnd = p + len;\r
+    uint32_t h32;\r
+\r
+#ifdef XXH_ACCEPT_NULL_INPUT_POINTER\r
+    if (p==NULL) { len=0; p=(const uint8_t *)(size_t)16; }\r
+#endif\r
+\r
+    if (len>=16)\r
+    {\r
+        const uint8_t * const limit = bEnd - 16;\r
+        uint32_t v1 = seed + PRIME32_1 + PRIME32_2;\r
+        uint32_t v2 = seed + PRIME32_2;\r
+        uint32_t v3 = seed + 0;\r
+        uint32_t v4 = seed - PRIME32_1;\r
+\r
+        do\r
+        {\r
+            v1 += XXH_readLE32_align((const uint32_t*)p, endian, align) * PRIME32_2; v1 = XXH_rotl32(v1, 13); v1 *= PRIME32_1; p+=4;\r
+            v2 += XXH_readLE32_align((const uint32_t*)p, endian, align) * PRIME32_2; v2 = XXH_rotl32(v2, 13); v2 *= PRIME32_1; p+=4;\r
+            v3 += XXH_readLE32_align((const uint32_t*)p, endian, align) * PRIME32_2; v3 = XXH_rotl32(v3, 13); v3 *= PRIME32_1; p+=4;\r
+            v4 += XXH_readLE32_align((const uint32_t*)p, endian, align) * PRIME32_2; v4 = XXH_rotl32(v4, 13); v4 *= PRIME32_1; p+=4;\r
+        } while (p<=limit);\r
+\r
+        h32 = XXH_rotl32(v1, 1) + XXH_rotl32(v2, 7) + XXH_rotl32(v3, 12) + XXH_rotl32(v4, 18);\r
+    }\r
+    else\r
+    {\r
+        h32  = seed + PRIME32_5;\r
+    }\r
+\r
+    h32 += (uint32_t) len;\r
+\r
+    while (p<=bEnd-4)\r
+    {\r
+        h32 += XXH_readLE32_align((const uint32_t*)p, endian, align) * PRIME32_3;\r
+        h32  = XXH_rotl32(h32, 17) * PRIME32_4 ;\r
+        p+=4;\r
+    }\r
+\r
+    while (p<bEnd)\r
+    {\r
+        h32 += (*p) * PRIME32_5;\r
+        h32 = XXH_rotl32(h32, 11) * PRIME32_1 ;\r
+        p++;\r
+    }\r
+\r
+    h32 ^= h32 >> 15;\r
+    h32 *= PRIME32_2;\r
+    h32 ^= h32 >> 13;\r
+    h32 *= PRIME32_3;\r
+    h32 ^= h32 >> 16;\r
+\r
+    return h32;\r
+}\r
+\r
+\r
+uint32_t XXH32(const void* input, int len, uint32_t seed)\r
+{\r
+#if 0\r
+    // Simple version, good for code maintenance, but unfortunately slow for small inputs\r
+    void* state = XXH32_init(seed);\r
+    XXH32_update(state, input, len);\r
+    return XXH32_digest(state);\r
+#else\r
+    XXH_endianess endian_detected = (XXH_endianess)XXH_CPU_LITTLE_ENDIAN;\r
+\r
+#  if !defined(XXH_USE_UNALIGNED_ACCESS)\r
+    if ((((size_t)input) & 3))   // Input is aligned, let's leverage the speed advantage\r
+    {\r
+        if ((endian_detected==XXH_littleEndian) || XXH_FORCE_NATIVE_FORMAT)\r
+            return XXH32_endian_align(input, len, seed, XXH_littleEndian, XXH_aligned);\r
+        else\r
+            return XXH32_endian_align(input, len, seed, XXH_bigEndian, XXH_aligned);\r
+    }\r
+#  endif\r
+\r
+    if ((endian_detected==XXH_littleEndian) || XXH_FORCE_NATIVE_FORMAT)\r
+        return XXH32_endian_align(input, len, seed, XXH_littleEndian, XXH_unaligned);\r
+    else\r
+        return XXH32_endian_align(input, len, seed, XXH_bigEndian, XXH_unaligned);\r
+#endif\r
+}\r
+\r
+\r
+//****************************\r
+// Advanced Hash Functions\r
+//****************************\r
+\r
+int XXH32_sizeofState() \r
+{\r
+    XXH_STATIC_ASSERT(XXH32_SIZEOFSTATE >= sizeof(struct XXH_state32_t));   // A compilation error here means XXH32_SIZEOFSTATE is not large enough\r
+    return sizeof(struct XXH_state32_t); \r
+}\r
+\r
+\r
+XXH_errorcode XXH32_resetState(void* state_in, uint32_t seed)\r
+{ \r
+    struct XXH_state32_t * state = (struct XXH_state32_t *) state_in;\r
+    state->seed = seed;\r
+    state->v1 = seed + PRIME32_1 + PRIME32_2;\r
+    state->v2 = seed + PRIME32_2;\r
+    state->v3 = seed + 0;\r
+    state->v4 = seed - PRIME32_1;\r
+    state->total_len = 0;\r
+    state->memsize = 0;\r
+    return XXH_OK;\r
+}\r
+\r
+\r
+void* XXH32_init (uint32_t seed)\r
+{\r
+    void* state = XXH_malloc (sizeof(struct XXH_state32_t));\r
+    XXH32_resetState(state, seed);\r
+    return state;\r
+}\r
+\r
+\r
+XXH_errorcode XXH32_update_endian (void* state_in, const void* input, int len, XXH_endianess endian)\r
+{\r
+    struct XXH_state32_t * state = (struct XXH_state32_t *) state_in;\r
+    const uint8_t *p = (const uint8_t *)input;\r
+    const uint8_t * const bEnd = p + len;\r
+\r
+#ifdef XXH_ACCEPT_NULL_INPUT_POINTER\r
+    if (input==NULL) return XXH_ERROR;\r
+#endif\r
+\r
+    state->total_len += len;\r
+\r
+    if (state->memsize + len < 16)   // fill in tmp buffer\r
+    {\r
+        XXH_memcpy(state->memory + state->memsize, input, len);\r
+        state->memsize +=  len;\r
+        return XXH_OK;\r
+    }\r
+\r
+    if (state->memsize)   // some data left from previous update\r
+    {\r
+        XXH_memcpy(state->memory + state->memsize, input, 16-state->memsize);\r
+        {\r
+            const uint32_t* p32 = (const uint32_t*)state->memory;\r
+            state->v1 += XXH_readLE32(p32, endian) * PRIME32_2; state->v1 = XXH_rotl32(state->v1, 13); state->v1 *= PRIME32_1; p32++;\r
+            state->v2 += XXH_readLE32(p32, endian) * PRIME32_2; state->v2 = XXH_rotl32(state->v2, 13); state->v2 *= PRIME32_1; p32++; \r
+            state->v3 += XXH_readLE32(p32, endian) * PRIME32_2; state->v3 = XXH_rotl32(state->v3, 13); state->v3 *= PRIME32_1; p32++;\r
+            state->v4 += XXH_readLE32(p32, endian) * PRIME32_2; state->v4 = XXH_rotl32(state->v4, 13); state->v4 *= PRIME32_1; p32++;\r
+        }\r
+        p += 16-state->memsize;\r
+        state->memsize = 0;\r
+    }\r
+\r
+    if (p <= bEnd-16)\r
+    {\r
+        const uint8_t * const limit = bEnd - 16;\r
+        uint32_t v1 = state->v1;\r
+        uint32_t v2 = state->v2;\r
+        uint32_t v3 = state->v3;\r
+        uint32_t v4 = state->v4;\r
+\r
+        do\r
+        {\r
+            v1 += XXH_readLE32((const uint32_t*)p, endian) * PRIME32_2; v1 = XXH_rotl32(v1, 13); v1 *= PRIME32_1; p+=4;\r
+            v2 += XXH_readLE32((const uint32_t*)p, endian) * PRIME32_2; v2 = XXH_rotl32(v2, 13); v2 *= PRIME32_1; p+=4;\r
+            v3 += XXH_readLE32((const uint32_t*)p, endian) * PRIME32_2; v3 = XXH_rotl32(v3, 13); v3 *= PRIME32_1; p+=4;\r
+            v4 += XXH_readLE32((const uint32_t*)p, endian) * PRIME32_2; v4 = XXH_rotl32(v4, 13); v4 *= PRIME32_1; p+=4;\r
+        } while (p<=limit);\r
+\r
+        state->v1 = v1;\r
+        state->v2 = v2;\r
+        state->v3 = v3;\r
+        state->v4 = v4;\r
+    }\r
+\r
+    if (p < bEnd)\r
+    {\r
+        XXH_memcpy(state->memory, p, bEnd-p);\r
+        state->memsize = (int)(bEnd-p);\r
+    }\r
+\r
+    return XXH_OK;\r
+}\r
+\r
+XXH_errorcode XXH32_update (void* state_in, const void* input, int len)\r
+{\r
+    XXH_endianess endian_detected = (XXH_endianess)XXH_CPU_LITTLE_ENDIAN;\r
+\r
+    if ((endian_detected==XXH_littleEndian) || XXH_FORCE_NATIVE_FORMAT)\r
+        return XXH32_update_endian(state_in, input, len, XXH_littleEndian);\r
+    else\r
+        return XXH32_update_endian(state_in, input, len, XXH_bigEndian);\r
+}\r
+\r
+\r
+\r
+uint32_t XXH32_intermediateDigest_endian (void* state_in, XXH_endianess endian)\r
+{\r
+    struct XXH_state32_t * state = (struct XXH_state32_t *) state_in;\r
+    const uint8_t *p = (const uint8_t *)state->memory;\r
+    uint8_t * bEnd = (uint8_t *)state->memory + state->memsize;\r
+    uint32_t h32;\r
+\r
+    if (state->total_len >= 16)\r
+    {\r
+        h32 = XXH_rotl32(state->v1, 1) + XXH_rotl32(state->v2, 7) + XXH_rotl32(state->v3, 12) + XXH_rotl32(state->v4, 18);\r
+    }\r
+    else\r
+    {\r
+        h32  = state->seed + PRIME32_5;\r
+    }\r
+\r
+    h32 += (uint32_t) state->total_len;\r
+\r
+    while (p<=bEnd-4)\r
+    {\r
+        h32 += XXH_readLE32((const uint32_t*)p, endian) * PRIME32_3;\r
+        h32  = XXH_rotl32(h32, 17) * PRIME32_4;\r
+        p+=4;\r
+    }\r
+\r
+    while (p<bEnd)\r
+    {\r
+        h32 += (*p) * PRIME32_5;\r
+        h32 = XXH_rotl32(h32, 11) * PRIME32_1;\r
+        p++;\r
+    }\r
+\r
+    h32 ^= h32 >> 15;\r
+    h32 *= PRIME32_2;\r
+    h32 ^= h32 >> 13;\r
+    h32 *= PRIME32_3;\r
+    h32 ^= h32 >> 16;\r
+\r
+    return h32;\r
+}\r
+\r
+\r
+uint32_t XXH32_intermediateDigest (void* state_in)\r
+{\r
+    XXH_endianess endian_detected = (XXH_endianess)XXH_CPU_LITTLE_ENDIAN;\r
+\r
+    if ((endian_detected==XXH_littleEndian) || XXH_FORCE_NATIVE_FORMAT)\r
+        return XXH32_intermediateDigest_endian(state_in, XXH_littleEndian);\r
+    else\r
+        return XXH32_intermediateDigest_endian(state_in, XXH_bigEndian);\r
+}\r
+\r
+\r
+uint32_t XXH32_digest (void* state_in)\r
+{\r
+    uint32_t h32 = XXH32_intermediateDigest(state_in);\r
+\r
+    XXH_free(state_in);\r
+\r
+    return h32;\r
+}\r

diff --git a/crc/xxhash.h b/crc/xxhash.h
new file mode 100644 (file)
index 0000000..4c7aafe
--- /dev/null
+++ b/crc/xxhash.h
@@ -0,0 +1,177 @@
+/*\r
+   xxHash - Fast Hash algorithm\r
+   Header File\r
+   Copyright (C) 2012-2014, Yann Collet.\r
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)\r
+\r
+   Redistribution and use in source and binary forms, with or without\r
+   modification, are permitted provided that the following conditions are\r
+   met:\r
+  \r
+       * Redistributions of source code must retain the above copyright\r
+   notice, this list of conditions and the following disclaimer.\r
+       * Redistributions in binary form must reproduce the above\r
+   copyright notice, this list of conditions and the following disclaimer\r
+   in the documentation and/or other materials provided with the\r
+   distribution.\r
+  \r
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS\r
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT\r
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR\r
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT\r
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,\r
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT\r
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,\r
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY\r
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT\r
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE\r
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.\r
+\r
+   You can contact the author at :\r
+   - xxHash source repository : http://code.google.com/p/xxhash/\r
+*/\r
+\r
+/* Notice extracted from xxHash homepage :\r
+\r
+xxHash is an extremely fast Hash algorithm, running at RAM speed limits.\r
+It also successfully passes all tests from the SMHasher suite.\r
+\r
+Comparison (single thread, Windows Seven 32 bits, using SMHasher on a Core 2 Duo @3GHz)\r
+\r
+Name            Speed       Q.Score   Author\r
+xxHash          5.4 GB/s     10\r
+CrapWow         3.2 GB/s      2       Andrew\r
+MumurHash 3a    2.7 GB/s     10       Austin Appleby\r
+SpookyHash      2.0 GB/s     10       Bob Jenkins\r
+SBox            1.4 GB/s      9       Bret Mulvey\r
+Lookup3         1.2 GB/s      9       Bob Jenkins\r
+SuperFastHash   1.2 GB/s      1       Paul Hsieh\r
+CityHash64      1.05 GB/s    10       Pike & Alakuijala\r
+FNV             0.55 GB/s     5       Fowler, Noll, Vo\r
+CRC32           0.43 GB/s     9\r
+MD5-32          0.33 GB/s    10       Ronald L. Rivest\r
+SHA1-32         0.28 GB/s    10\r
+\r
+Q.Score is a measure of quality of the hash function. \r
+It depends on successfully passing SMHasher test set. \r
+10 is a perfect score.\r
+*/\r
+\r
+#pragma once\r
+\r
+#if defined (__cplusplus)\r
+extern "C" {\r
+#endif\r
+\r
+#include <inttypes.h>\r
+\r
+struct XXH_state32_t\r
+{\r
+    uint64_t total_len;\r
+    uint32_t seed;\r
+    uint32_t v1;\r
+    uint32_t v2;\r
+    uint32_t v3;\r
+    uint32_t v4;\r
+    int memsize;\r
+    char memory[16];\r
+};\r
+\r
+//****************************\r
+// Type\r
+//****************************\r
+typedef enum { XXH_OK=0, XXH_ERROR } XXH_errorcode;\r
+\r
+\r
+\r
+//****************************\r
+// Simple Hash Functions\r
+//****************************\r
+\r
+unsigned int XXH32 (const void* input, int len, unsigned int seed);\r
+\r
+/*\r
+XXH32() :\r
+    Calculate the 32-bits hash of sequence of length "len" stored at memory address "input".\r
+    The memory between input & input+len must be valid (allocated and read-accessible).\r
+    "seed" can be used to alter the result predictably.\r
+    This function successfully passes all SMHasher tests.\r
+    Speed on Core 2 Duo @ 3 GHz (single thread, SMHasher benchmark) : 5.4 GB/s\r
+    Note that "len" is type "int", which means it is limited to 2^31-1.\r
+    If your data is larger, use the advanced functions below.\r
+*/\r
+\r
+\r
+\r
+//****************************\r
+// Advanced Hash Functions\r
+//****************************\r
+\r
+void*         XXH32_init   (unsigned int seed);\r
+XXH_errorcode XXH32_update (void* state, const void* input, int len);\r
+unsigned int  XXH32_digest (void* state);\r
+\r
+/*\r
+These functions calculate the xxhash of an input provided in several small packets,\r
+as opposed to an input provided as a single block.\r
+\r
+It must be started with :\r
+void* XXH32_init()\r
+The function returns a pointer which holds the state of calculation.\r
+\r
+This pointer must be provided as "void* state" parameter for XXH32_update().\r
+XXH32_update() can be called as many times as necessary.\r
+The user must provide a valid (allocated) input.\r
+The function returns an error code, with 0 meaning OK, and any other value meaning there is an error.\r
+Note that "len" is type "int", which means it is limited to 2^31-1. \r
+If your data is larger, it is recommended to chunk your data into blocks \r
+of size for example 2^30 (1GB) to avoid any "int" overflow issue.\r
+\r
+Finally, you can end the calculation anytime, by using XXH32_digest().\r
+This function returns the final 32-bits hash.\r
+You must provide the same "void* state" parameter created by XXH32_init().\r
+Memory will be freed by XXH32_digest().\r
+*/\r
+\r
+\r
+int           XXH32_sizeofState();\r
+XXH_errorcode XXH32_resetState(void* state, unsigned int seed);\r
+\r
+#define       XXH32_SIZEOFSTATE 48\r
+typedef struct { long long ll[(XXH32_SIZEOFSTATE+(sizeof(long long)-1))/sizeof(long long)]; } XXH32_stateSpace_t;\r
+/*\r
+These functions allow user application to make its own allocation for state.\r
+\r
+XXH32_sizeofState() is used to know how much space must be allocated for the xxHash 32-bits state.\r
+Note that the state must be aligned to access 'long long' fields. Memory must be allocated and referenced by a pointer.\r
+This pointer must then be provided as 'state' into XXH32_resetState(), which initializes the state.\r
+\r
+For static allocation purposes (such as allocation on stack, or freestanding systems without malloc()),\r
+use the structure XXH32_stateSpace_t, which will ensure that memory space is large enough and correctly aligned to access 'long long' fields.\r
+*/\r
+\r
+\r
+unsigned int XXH32_intermediateDigest (void* state);\r
+/*\r
+This function does the same as XXH32_digest(), generating a 32-bit hash,\r
+but preserve memory context.\r
+This way, it becomes possible to generate intermediate hashes, and then continue feeding data with XXH32_update().\r
+To free memory context, use XXH32_digest(), or free().\r
+*/\r
+\r
+\r
+\r
+//****************************\r
+// Deprecated function names\r
+//****************************\r
+// The following translations are provided to ease code transition\r
+// You are encouraged to no longer this function names\r
+#define XXH32_feed   XXH32_update\r
+#define XXH32_result XXH32_digest\r
+#define XXH32_getIntermediateResult XXH32_intermediateDigest\r
+\r
+\r
+\r
+#if defined (__cplusplus)\r
+}\r
+#endif\r

diff --git a/fio.1 b/fio.1
index 4377052ff61ae88224f4668304c7f8a6de4874b6..81b03f70ece7e8e0e7d4768687a3811d532599ab 100644 (file)
--- 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 ac101a3fe27ae7baa68ea21583f4ea41ebb2c4f7..dc55c985cc19b0d2940642cb1cac49dc3cd1aee7 100644 (file)
--- 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 93731228f1b68661e0edd3293bc491ccb650774d..9eb532a28e080a66a2483943f09fed16efd9ecfa 100644 (file)
--- 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 05d7b81b804460e7c49738acb42f83d4dad8bd6e..dba7743a75d671447e082a4e4571ae615949a725 100644 (file)
--- 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