xxHash

xxHash - Extremely fast hash algorithm
======================================

[Website](https://xxhash.com/) · [API documentation](lib/README.md#documentation) · [`xxhsum` manual](cli/xxhsum.1.md) · [Format specification](doc/xxhash_spec.md)

xxHash is an extremely fast non-cryptographic hash algorithm, working at RAM
speed limits. It is highly portable and produces identical hashes on all
platforms, including little- and big-endian systems. Once finalized, algorithm
outputs remain stable across xxHash releases.

The library includes the following algorithms:

- XXH32 generates 32-bit hashes, using 32-bit arithmetic.
- XXH64 generates 64-bit hashes, using 64-bit arithmetic.
- XXH3, available since v0.8.0, generates 64-bit or 128-bit hashes using
  vectorized arithmetic. The 128-bit variant is called XXH128.

For new applications, `XXH3_64bits()` is the recommended default. Use
`XXH3_128bits()` when a 128-bit hash is required.

Benchmarks
----------

The benchmarked reference system uses an Intel i7-9700K CPU running Ubuntu
x64 20.04. The [open source benchmark program] is compiled with Clang v10.0
using `-O3`.

| Hash Name | Width | Bandwidth | Small Data Velocity | Comment |
| --- | ---: | ---: | ---: | --- |
| **XXH3** (AVX2) | 64 | 59.4 GB/s | 133.1 | |
| **XXH128** (AVX2) | 128 | 57.9 GB/s | 118.1 | |
| **XXH3** (SSE2) | 64 | 31.5 GB/s | 133.1 | |
| **XXH128** (SSE2) | 128 | 29.6 GB/s | 118.1 | |
| *memcpy*, from RAM | N/A | 28.0 GB/s | N/A | *for reference* |
| City64 | 64 | 22.0 GB/s | 76.6 | |
| T1ha2 | 64 | 22.0 GB/s | 99.0 | Slightly worse [collisions] |
| City128 | 128 | 21.7 GB/s | 57.7 | |
| **XXH64** | 64 | 19.4 GB/s | 71.0 | |
| SpookyHash | 64 | 19.3 GB/s | 53.2 | |
| Mum | 64 | 18.0 GB/s | 67.0 | Slightly worse [collisions] |
| **XXH32** | 32 | 9.7 GB/s | 71.9 | |
| City32 | 32 | 9.1 GB/s | 66.0 | |
| Murmur3 | 32 | 3.9 GB/s | 56.1 | |
| SipHash | 64 | 3.0 GB/s | 43.2 | |
| FNV64 | 64 | 1.2 GB/s | 62.7 | Poor avalanche properties |
| Blake2 | 256 | 1.1 GB/s | 5.1 | Cryptographic |
| SHA1 | 160 | 0.8 GB/s | 5.6 | Cryptographic but broken |
| MD5 | 128 | 0.6 GB/s | 7.8 | Cryptographic but broken |

[open source benchmark program]: tests/bench
[collisions]: https://github.com/Cyan4973/xxHash/wiki/Collision-ratio-comparison#collision-study

For current measurements on additional platforms, see the
[benchmarks on xxhash.com](https://xxhash.com/#benchmarks).

Note 1: Small data velocity is a _rough_ evaluation of an algorithm's
efficiency on small data. For more detailed analysis, see the next section.

Note 2: Some algorithms feature _faster than RAM_ speed. They can only reach
their full speed potential when input is already in CPU cache (L3 or better).
Otherwise, they are limited by RAM speed.

### Small data

Performance on large data is only one part of the picture.
Hashing is also very useful in constructions like hash tables and bloom filters.
In these use cases, it is common to hash many small inputs, sometimes only a
few bytes