diff --git a/LICENSE b/LICENSE new file mode 100644 --- /dev/null +++ b/LICENSE @@ -0,0 +1,32 @@ +A pure haskell implementation of the xxHash algorithm + +Copyright © 2014 Anchor Systems, Pty Ltd +All rights reserved. + +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: + + 1. Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + 2. 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. + + 3. Neither the name of the project nor the names of its contributors + may be used to endorse or promote products derived from this + software without specific prior written permission. + +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. diff --git a/Setup.hs b/Setup.hs new file mode 100644 --- /dev/null +++ b/Setup.hs @@ -0,0 +1,2 @@ +import Distribution.Simple +main = defaultMain diff --git a/bench/Bench.hs b/bench/Bench.hs new file mode 100644 --- /dev/null +++ b/bench/Bench.hs @@ -0,0 +1,44 @@ +{-# OPTIONS_GHC -fno-warn-orphans #-} +module Main + ( main -- :: IO () + ) where +import Prelude hiding (words) + +import qualified Data.ByteString as B +import qualified Data.ByteString.Lazy as L +import qualified Data.Digest.XXHash as XXHash +import qualified Data.Digest.Adler32 as Adler32 +import qualified Data.Digest.CRC32 as CRC32 +import qualified Data.Digest.Murmur32 as M32 +import qualified Data.Hashable as Hashable + +import Criterion.Main +import Control.DeepSeq +instance NFData M32.Hash32 + +main :: IO () +main = do + med <- L.readFile "/usr/share/dict/all" + med' <- B.readFile "/usr/share/dict/all" + + small <- L.readFile "/usr/share/dict/cracklib-small" + small' <- B.readFile "/usr/share/dict/cracklib-small" + + defaultMain + [ bgroup "24MB text" + [ bench "xxhash (c)" $ nf XXHash.c_xxHash' med' + , bench "Data.Digest.XXHash lazy (haskell)" $ nf XXHash.xxHash med + , bench "Data.Digest.XXHash (haskell)" $ nf XXHash.xxHash' med' + , bench "Data.Digest.adler32 (c, zlib)" $ nf Adler32.adler32 med' + , bench "Data.Digest.crc32 (c, zlib)" $ nf CRC32.crc32 med' + , bench "Data.Hashable (c, FNV?)" $ nf Hashable.hash med' + ] + , bgroup "478K text" + [ bench "Data.Digest.XXHash lazy (haskell)" $ nf XXHash.xxHash small + , bench "Data.Digest.XXHash (haskell)" $ nf XXHash.xxHash' small' + , bench "xxhash (c)" $ nf XXHash.c_xxHash' small' + , bench "Data.Digest.adler32 (c, zlib)" $ nf Adler32.adler32 small' + , bench "Data.Digest.crc32 (c, zlib)" $ nf CRC32.crc32 small' + , bench "Data.Hashable (c, FNV?)" $ nf Hashable.hash small' + ] + ] diff --git a/src/Data/Digest/CXXHash.hsc b/src/Data/Digest/CXXHash.hsc new file mode 100644 --- /dev/null +++ b/src/Data/Digest/CXXHash.hsc @@ -0,0 +1,18 @@ +{-# LANGUAGE CPP #-} +module Data.Digest.CXXHash +( + c_XXH32 +) where + +import qualified Data.ByteString as B +import Data.Word (Word32) +import Foreign.C.String +import Foreign.C.Types + +#include +-- | Compresses a string. +foreign import ccall unsafe "xxhash.h XXH32" + c_XXH32 :: CString -- ^ Data + -> CSize -- ^ Size + -> CUInt -- ^ Seed + -> IO Word32 -- ^ Seed diff --git a/src/Data/Digest/XXHash.hs b/src/Data/Digest/XXHash.hs new file mode 100644 --- /dev/null +++ b/src/Data/Digest/XXHash.hs @@ -0,0 +1,210 @@ +{-# LANGUAGE BangPatterns #-} +{-# LANGUAGE CPP #-} +{-# LANGUAGE MultiParamTypeClasses #-} +{-# LANGUAGE RecordWildCards #-} +{-# LANGUAGE TypeSynonymInstances #-} +{-# OPTIONS_GHC -funbox-strict-fields -fno-warn-orphans #-} + +-- | +-- Module : Data.Digest.XXHash +-- Copyright : (c) Christian Marie 2013 +-- License : BSD3 +-- +-- Maintainer : pingu@anchor.net.au +-- Stability : experimental +-- Portability : portable +-- +-- This module provides a pure implementation of the xxHash algorithm. +-- +-- This implementation is almost as fast as the C version, which is avaliable +-- as c_xxHash'. +-- +-- Criterion benchmarks are avaliable via cabal bench. +-- +-- More information on the algorithm may be found here: +-- +-- +module Data.Digest.XXHash +( + -- *Types + XXHash, + -- *Haskell implementation + xxHash, + xxHash', + -- *C bindings + c_xxHash', +) where + +#include "MachDeps.h" + +import Crypto.Classes (Hash (..), hash) +import Data.Bits +import qualified Data.ByteString as B +import Data.ByteString.Internal (ByteString (PS), inlinePerformIO) +import qualified Data.ByteString.Lazy as L +import qualified Data.Digest.CXXHash as C +import Data.Tagged +import Data.Word (Word32, Word64, Word8) +import Foreign.ForeignPtr (withForeignPtr) +import Foreign.Ptr (Ptr, castPtr, plusPtr, nullPtr) +import Foreign.Storable (peek) +import System.IO.Unsafe (unsafePerformIO) + +-- These all get unboxed. Previous versions used unboxed words, tuples and +-- primitive operations directly. This is much more readable for only a slight +-- performance penalty (less than 5%) +data XXHashCtx = XXHashCtx { a1 :: !Word32 -- ^ Accumulators + , a2 :: !Word32 + , a3 :: !Word32 + , a4 :: !Word32 + , iterations :: !Word32 -- ^ To keep track length + } deriving (Eq, Show) + +type XXHash = Word32 + +-- The user can't currently touch the seed, it is set to zero in the Hash +-- instance definition +type Seed = Word32 + +initializeXXHashCtx :: Seed -> XXHashCtx +initializeXXHashCtx seed = + XXHashCtx a1 a2 a3 a4 iterations + where a1 = seed + prime1 + prime2 + a2 = seed + prime2 + a3 = seed + a4 = seed - prime1 + iterations = 0 + +prime1, prime2, prime3, prime4, prime5 :: Word32 +prime1 = 2654435761 +prime2 = 2246822519 +prime3 = 3266489917 +prime4 = 668265263 +prime5 = 374761393 + +-- This is inlined without any explicit help, the inline here is maybe +-- redundant. +stageOne :: Word32 -> Word32 -> Word32 -> Word32 -> XXHashCtx -> XXHashCtx +stageOne i1 i2 i3 i4 XXHashCtx{..} = + XXHashCtx (vx a1 i1) (vx a2 i2) (vx a3 i3) (vx a4 i4) (iterations + 1) + where + vx v i = ((v + i * prime2) `rotateL` 13) * prime1 +{-# INLINE stageOne #-} + +-- This is always called with a multiple of sixteen, convenient. +updateXXHashCtx :: XXHashCtx -> ByteString -> XXHashCtx +updateXXHashCtx ctx (PS fp os len) = + inlinePerformIO . withForeignPtr fp $ \bs_base_ptr -> + let ptr_beg = bs_base_ptr `plusPtr` os + ptr_end = ptr_beg `plusPtr` len + go ptr !ctx' + | ptr < ptr_end = do + i1 <- peekLE32 ptr 0 + i2 <- peekLE32 ptr 4 + i3 <- peekLE32 ptr 8 + i4 <- peekLE32 ptr 12 + go (ptr `plusPtr` 16) (stageOne i1 i2 i3 i4 ctx') + | otherwise = return ctx' + in go ptr_beg ctx + +finalizeXXHashCtx :: Seed -> XXHashCtx -> ByteString -> XXHash +finalizeXXHashCtx seed ctx (PS fp os len) = + inlinePerformIO . withForeignPtr fp $ \bs_base_ptr -> + let ptr_beg = bs_base_ptr `plusPtr` os + ptr_end = ptr_beg `plusPtr` len + total_len :: Word64 + total_len = fromIntegral len + fromIntegral (iterations ctx) * 16 + go :: Ptr Word8 -> XXHashCtx -> IO XXHash + go ptr !ctx' + | ptr /= nullPtr && ptr <= ptr_end `plusPtr` (-16) = do + i1 <- peekLE32 ptr 0 + i2 <- peekLE32 ptr 4 + i3 <- peekLE32 ptr 8 + i4 <- peekLE32 ptr 12 + go (ptr `plusPtr` 16) (stageOne i1 i2 i3 i4 ctx') + | otherwise = + goEnd ptr $ + if total_len >= 16 + then ctxToDigest ctx' total_len + else seed + prime5 + fromIntegral total_len + goEnd :: Ptr Word8 -> XXHash -> IO XXHash + goEnd ptr !xxhash + | ptr /= nullPtr && ptr <= ptr_end `plusPtr` (-4) = do + b <- peekLE32 ptr 0 + goEnd (ptr `plusPtr` 4) (stageTwo b xxhash) + | ptr == ptr_end = + return $ finalizeXXHash xxhash + | otherwise = do + b <- peekByte ptr + goEnd (ptr `plusPtr` 1) (stageThree b xxhash) + in go ptr_beg ctx + where + stageTwo i xxhash = + ((xxhash + i * prime3) `rotateL` 17) * prime4 + + stageThree i xxhash = + ((xxhash + fromIntegral i * prime5) `rotateL` 11) * prime1 + + ctxToDigest XXHashCtx{..} total_len = + (a1 `rotateL` 1) + (a2 `rotateL` 7) + + (a3 `rotateL` 12) + (a4 `rotateL` 18) + fromIntegral total_len + + finalizeXXHash xxhash = step2 `xor` step2 `shiftR` 16 + where + step1 = (xxhash `xor` (xxhash `shiftR` 15)) * prime2 + step2 = (step1 `xor` (step1 `shiftR` 13)) * prime3 + +-- PeekByte and peekLE32 are more or less borrowed and modified code from +-- this: +-- http://www.serpentine.com/blog/2012/10/02/a-fast-new-siphash-implementation-in-haskell/ +peekByte :: Ptr Word8 -> IO Word8 +peekByte = peek +{-# INLINE peekByte #-} + +peekLE32 :: Ptr Word8 -> Int -> IO Word32 +#if defined(x86_64_HOST_ARCH) || defined(i386_HOST_ARCH) +peekLE32 ptr os = peek (castPtr (ptr `plusPtr` os)) +#else +-- Super slow, around 100x slower. +peekLE32 ptr os = do + let peek8 d = fromIntegral `fmap` peekByte (ptr `plusPtr` (d + os)) + b0 <- peek8 0 + b1 <- peek8 1 + b2 <- peek8 2 + b3 <- peek8 3 + let w = (b3 `shiftL` 24) .|. (b2 `shiftL` 16) .|. + (b1 `shiftL` 8) .|. b0 + return w +#endif +{-# INLINE peekLE32 #-} + +-- Crypto.Classes does our boring stuff for us :D +instance Hash XXHashCtx XXHash where + outputLength = Tagged 32 -- Word32 + -- 128 bits, meaning that the context will never be updated with less + -- than 16 bytes + blockLength = Tagged 128 + initialCtx = initializeXXHashCtx 0 + updateCtx = updateXXHashCtx + finalize = finalizeXXHashCtx 0 + + +-- | +-- Hash a lazy ByteString. +xxHash :: L.ByteString -> XXHash +xxHash = hash + +-- | +-- Hash a strict ByteString. +xxHash' :: ByteString -> XXHash +xxHash' = hash' + +-- | +-- Hash a strict ByteString using the C implementation, the length of the +-- ByteString should be limited to 2^31-1 or the results will be invalid. +-- +-- This is mostly used internally for benchmarking and verification. It's use +-- in production is not recommended. +c_xxHash' :: B.ByteString -> XXHash +c_xxHash' bs = unsafePerformIO . B.useAsCStringLen bs $ \(str, len) -> + C.c_XXH32 str (fromIntegral len) 0 diff --git a/src/cbits/xxhash.c b/src/cbits/xxhash.c new file mode 100644 --- /dev/null +++ b/src/cbits/xxhash.c @@ -0,0 +1,475 @@ +/* +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 U32). +#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 + + +//************************************** +// Compiler Specific Options +//************************************** +// Disable some Visual warning messages +#ifdef _MSC_VER // Visual Studio +# pragma warning(disable : 4127) // disable: C4127: conditional expression is constant +#endif + +#ifdef _MSC_VER // Visual Studio +# define FORCE_INLINE static __forceinline +#else +# ifdef __GNUC__ +# define FORCE_INLINE static inline __attribute__((always_inline)) +# else +# define FORCE_INLINE static inline +# endif +#endif + + +//************************************** +// 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 +FORCE_INLINE void* XXH_malloc(size_t s) { return malloc(s); } +FORCE_INLINE void XXH_free (void* p) { free(p); } +// for memcpy() +#include +FORCE_INLINE void* XXH_memcpy(void* dest, const void* src, size_t size) { return memcpy(dest,src,size); } + + +//************************************** +// Basic Types +//************************************** +#if defined (__STDC_VERSION__) && __STDC_VERSION__ >= 199901L // C99 +# include + typedef uint8_t BYTE; + typedef uint16_t U16; + typedef uint32_t U32; + typedef int32_t S32; + typedef uint64_t U64; +#else + typedef unsigned char BYTE; + typedef unsigned short U16; + typedef unsigned int U32; + typedef signed int S32; + typedef unsigned long long U64; +#endif + +#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 _U32_S { U32 v; } _PACKED U32_S; + +#if !defined(XXH_USE_UNALIGNED_ACCESS) && !defined(__GNUC__) +# pragma pack(pop) +#endif + +#define A32(x) (((U32_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 U32 XXH_swap32 (U32 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; + +FORCE_INLINE U32 XXH_readLE32_align(const U32* 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); +} + +FORCE_INLINE U32 XXH_readLE32(const U32* ptr, XXH_endianess endian) { return XXH_readLE32_align(ptr, endian, XXH_unaligned); } + + +//**************************** +// Simple Hash Functions +//**************************** +FORCE_INLINE U32 XXH32_endian_align(const void* input, int len, U32 seed, XXH_endianess endian, XXH_alignment align) +{ + const BYTE* p = (const BYTE*)input; + const BYTE* const bEnd = p + len; + U32 h32; + +#ifdef XXH_ACCEPT_NULL_INPUT_POINTER + if (p==NULL) { len=0; p=(const BYTE*)(size_t)16; } +#endif + + if (len>=16) + { + const BYTE* const limit = bEnd - 16; + U32 v1 = seed + PRIME32_1 + PRIME32_2; + U32 v2 = seed + PRIME32_2; + U32 v3 = seed + 0; + U32 v4 = seed - PRIME32_1; + + do + { + v1 += XXH_readLE32_align((const U32*)p, endian, align) * PRIME32_2; v1 = XXH_rotl32(v1, 13); v1 *= PRIME32_1; p+=4; + v2 += XXH_readLE32_align((const U32*)p, endian, align) * PRIME32_2; v2 = XXH_rotl32(v2, 13); v2 *= PRIME32_1; p+=4; + v3 += XXH_readLE32_align((const U32*)p, endian, align) * PRIME32_2; v3 = XXH_rotl32(v3, 13); v3 *= PRIME32_1; p+=4; + v4 += XXH_readLE32_align((const U32*)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 += (U32) len; + + while (p<=bEnd-4) + { + h32 += XXH_readLE32_align((const U32*)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; +} + + +U32 XXH32(const void* input, int len, U32 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 +//**************************** + +struct XXH_state32_t +{ + U64 total_len; + U32 seed; + U32 v1; + U32 v2; + U32 v3; + U32 v4; + int memsize; + char memory[16]; +}; + + +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, U32 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 (U32 seed) +{ + void* state = XXH_malloc (sizeof(struct XXH_state32_t)); + XXH32_resetState(state, seed); + return state; +} + + +FORCE_INLINE 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 BYTE* p = (const BYTE*)input; + const BYTE* 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 U32* p32 = (const U32*)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 BYTE* const limit = bEnd - 16; + U32 v1 = state->v1; + U32 v2 = state->v2; + U32 v3 = state->v3; + U32 v4 = state->v4; + + do + { + v1 += XXH_readLE32((const U32*)p, endian) * PRIME32_2; v1 = XXH_rotl32(v1, 13); v1 *= PRIME32_1; p+=4; + v2 += XXH_readLE32((const U32*)p, endian) * PRIME32_2; v2 = XXH_rotl32(v2, 13); v2 *= PRIME32_1; p+=4; + v3 += XXH_readLE32((const U32*)p, endian) * PRIME32_2; v3 = XXH_rotl32(v3, 13); v3 *= PRIME32_1; p+=4; + v4 += XXH_readLE32((const U32*)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); +} + + + +FORCE_INLINE U32 XXH32_intermediateDigest_endian (void* state_in, XXH_endianess endian) +{ + struct XXH_state32_t * state = (struct XXH_state32_t *) state_in; + const BYTE * p = (const BYTE*)state->memory; + BYTE* bEnd = (BYTE*)state->memory + state->memsize; + U32 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 += (U32) state->total_len; + + while (p<=bEnd-4) + { + h32 += XXH_readLE32((const U32*)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; +} + + +U32 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); +} + + +U32 XXH32_digest (void* state_in) +{ + U32 h32 = XXH32_intermediateDigest(state_in); + + XXH_free(state_in); + + return h32; +} diff --git a/src/cbits/xxhash.h b/src/cbits/xxhash.h new file mode 100644 --- /dev/null +++ b/src/cbits/xxhash.h @@ -0,0 +1,164 @@ +/* + 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 + + +//**************************** +// 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/tests/TestSuite.hs b/tests/TestSuite.hs new file mode 100644 --- /dev/null +++ b/tests/TestSuite.hs @@ -0,0 +1,37 @@ +{-# LANGUAGE OverloadedStrings #-} +{-# OPTIONS_GHC -fno-warn-orphans #-} + +module TestSuite where + +import Test.Hspec +import Test.Hspec.QuickCheck +import Test.QuickCheck + +import Data.Digest.XXHash +import qualified Data.ByteString.Char8 as B +import qualified Data.ByteString.Lazy as L + +instance Arbitrary L.ByteString where + arbitrary = L.pack `fmap` arbitrary + shrink xs = L.pack `fmap` shrink (L.unpack xs) + +suite :: Spec +suite = do + describe "Hashing a bytestring" $ do + it "does not explode with sizes up to 2048" $ do + let strings = take 2048 $ iterate ('A':) "" in + (length $ map (xxHash' . B.pack) strings) `shouldBe` 2048 + + it "works with known pairs" $ do + xxHash "" `shouldBe` 46947589 + xxHash "the quick brown fox jumps over a pony" `shouldBe` 2069436874 + xxHash' "small" `shouldBe` 2387791210 + xxHash' "" `shouldBe` 46947589 + let as = B.pack . take 500 $ repeat 'A' in + xxHash' as `shouldBe` 1719491600 + + describe "Fuzzing random strings" $ do + prop "is deterministic" $ \string -> + let h = xxHash string in + (xxHash' $ L.toStrict string) == h && + (c_xxHash' $ L.toStrict string) == h diff --git a/tests/check.hs b/tests/check.hs new file mode 100644 --- /dev/null +++ b/tests/check.hs @@ -0,0 +1,23 @@ +-- +-- Haskell bindings to librados +-- +-- Copyright © 2013- Anchor Systems, Pty Ltd +-- +-- The code in this file, and the program it is a part of, is +-- made available to you by its authors as open source software: +-- you can redistribute it and/or modify it under the terms of +-- the BSD licence. +-- + +{-# LANGUAGE OverloadedStrings #-} +{-# OPTIONS -fno-warn-unused-imports #-} + +module Main where + +import Test.Hspec (hspec) + +import TestSuite (suite) + +main :: IO () +main = do + hspec suite diff --git a/xxhash.cabal b/xxhash.cabal new file mode 100644 --- /dev/null +++ b/xxhash.cabal @@ -0,0 +1,58 @@ +name: xxhash +version: 0.0.1 +license: BSD3 +license-file: LICENSE +author: Christian Marie +maintainer: Christian Marie +description: A Haskell implementation of the xxHash algorithm, for fast + non-cryptographic checksums of ByteStrings. +synopsis: A Haskell implementation of the xxHash algorithm +category: Data, Cryptography +build-type: Simple +cabal-version: >= 1.10 +extra-source-files: + src/cbits/xxhash.c, src/cbits/xxhash.h +source-repository head + type: git + location: https://github.com/christian-marie/xxhash/ + +library + build-depends: base >=4.6 && <4.7, + bytestring, + tagged, + crypto-api + ghc-options: + c-sources: src/cbits/xxhash.c + include-dirs: src/cbits + hs-source-dirs: src + default-language: Haskell2010 + exposed-modules: Data.Digest.XXHash + other-modules: Data.Digest.CXXHash + +test-suite check + type: exitcode-stdio-1.0 + build-depends: base >=4.6 && <4.7, + bytestring, + hspec, + QuickCheck, + xxhash + hs-source-dirs: tests + main-is: check.hs + other-modules: TestSuite + default-language: Haskell2010 + +benchmark bench + hs-source-dirs: bench + main-is: Bench.hs + type: exitcode-stdio-1.0 + build-depends: base >=4.6 && <4.7, + bytestring, + criterion, + digest, + murmur-hash, + deepseq, + hashable, + xxhash + + ghc-options: -rtsopts + default-language: Haskell2010