xxHash is an Extremely fast Hash algorithm, running at RAM speed limits. It successfully completes the SMHasher test suite which evaluates collision, dispersion and randomness qualities of hash functions. Code is highly portable, and hashes are identical on all platforms (little / big endian).

Branch	Status
master
dev

The benchmark uses SMHasher speed test, compiled with Visual 2010 on a Windows Seven 32-bit box. The reference system uses a Core 2 Duo @3GHz

Q.Score is a measure of quality of the hash function. It depends on successfully passing SMHasher test set. 10 is a perfect score. Algorithms with a score < 5 are not listed on this table.

A more recent version, XXH64, has been created thanks to Mathias Westerdahl, which offers superior speed and dispersion for 64-bit systems. Note however that 32-bit applications will still run faster using the 32-bit version.

SMHasher speed test, compiled using GCC 4.8.2, on Linux Mint 64-bit. The reference system uses a Core i5-3340M @2.7GHz

This project also includes a command line utility, named xxhsum, offering similar features as md5sum, thanks to Takayuki Matsuoka contributions.

The library files xxhash.c and xxhash.h are BSD licensed. The utility xxhsum is GPL licensed.

The following macros can be set at compilation time, they modify xxhash behavior. They are all disabled by default.

• XXH_INLINE_ALL : Make all functions inline, with bodies directly included within xxhash.h. There is no need for an xxhash.o module in this case. Inlining functions is generally beneficial for speed on small keys. It's especially effective when key length is a compile time constant, with observed performance improvement in the +200% range . See this article for details.
• XXH_ACCEPT_NULL_INPUT_POINTER : if set to 1, when input is a null-pointer, xxhash result is the same as a zero-length key (instead of a dereference segfault).
• XXH_FORCE_MEMORY_ACCESS : default method 0 uses a portable memcpy() notation. Method 1 uses a gcc-specific packed attribute, which can provide better performance for some targets. Method 2 forces unaligned reads, which is not standard compliant, but might sometimes be the only way to extract better performance.
• XXH_CPU_LITTLE_ENDIAN : by default, endianess is determined at compile time. It's possible to skip auto-detection and force format to little-endian, by setting this macro to 1. Setting it to 0 forces big-endian.
• XXH_PRIVATE_API : same impact as XXH_INLINE_ALL. Name underlines that symbols will not be published on library public interface.
• XXH_NAMESPACE : prefix all symbols with the value of XXH_NAMESPACE. Useful to evade symbol naming collisions, in case of multiple inclusions of xxHash source code. Client applications can still use regular function name, symbols are automatically translated through xxhash.h.
• XXH_STATIC_LINKING_ONLY : gives access to state declaration for static allocation. Incompatible with dynamic linking, due to risks of ABI changes.
• XXH_NO_LONG_LONG : removes support for XXH64, for targets without 64-bit support.
• XXH_IMPORT : should only be defined for dynamic linking, it prevents linkage errors with MSVC.

Calling xxhash 64-bit variant from a C program :

#include "xxhash.h"

unsigned long long calcul_hash(const void* buffer, size_t length)
{
    unsigned long long const seed = 0;   /* or any other value */
    unsigned long long const hash = XXH64(buffer, length, seed);
    return hash;
}

Using streaming variant is more involved, but makes it possible to provide data in multiple rounds :

#include "stdlib.h"   /* abort() */
#include "xxhash.h"


unsigned long long calcul_hash_streaming(someCustomType handler)
{
    /* create a hash state */
    XXH64_state_t* const state = XXH64_createState();
    if (state==NULL) abort();

    size_t const bufferSize = SOME_SIZE;
    void* const buffer = malloc(bufferSize);
    if (buffer==NULL) abort();

    /* Initialize state with selected seed */
    unsigned long long const seed = 0;   /* or any other value */
    XXH_errorcode const resetResult = XXH64_reset(state, seed);
    if (resetResult == XXH_ERROR) abort();

    /* Feed the state with input data, any size, any number of times */
    (...)
    while ( /* any condition */ ) {
        size_t const length = get_more_data(buffer, bufferSize, handler);   
        XXH_errorcode const updateResult = XXH64_update(state, buffer, length);
        if (updateResult == XXH_ERROR) abort();
        (...)
    }
    (...)

    /* Get the hash */
    XXH64_hash_t const hash = XXH64_digest(state);

    /* State can then be re-used; in this example, it is simply freed  */
    free(buffer);
    XXH64_freeState(state);

    return (unsigned long long)hash;
}

Starting with v0.7.0, the library includes a new algorithm, named XXH3, able to generate 64 and 128-bits hashes.

The new algorithm is much faster than its predecessors, for both long and small inputs, as can be observed in following graphs :

The algorithm is currently labelled experimental, as it may change in a future version. To access it, one need to unlock its declaration using macro XXH_STATIC_LINKING_ONLY. It can be used for ephemeral data, and for tests, but avoid storing long-term hash values yet. XXH3 will be stabilized in a future version. This period will be used to collect users' feedback.

Beyond the C reference version, xxHash is also available on many programming languages, thanks to great contributors. They are listed here.

• The "master" branch is considered stable, at all times.
• The "dev" branch is the one where all contributions must be merged before being promoted to master.
  • If you plan to propose a patch, please commit into the "dev" branch, or its own feature branch. Direct commit to "master" are not permitted.