xxhash-wasm

A WebAssembly implementation of xxHash, a fast non-cryptographic hash algorithm. It can be called seamlessly from JavaScript. You can use it like any other JavaScript library but still get the benefits of WebAssembly, no special setup needed.

Table of Contents

• Installation
  • From npm
  • From Unpkg
    • ES Modules
    • UMD build
  • Cloudflare Workers
• Usage
  • Streaming Example
  • Node
• Performance
  • Engine Requirements
• API
  • h32
  • h64
  • Streaming
• Comparison to xxhashjs
  • Benchmarks
  • Bundle size

Installation

From npm

npm install --save xxhash-wasm

Or with Yarn:

yarn add xxhash-wasm

From Unpkg

ES Modules

<script type="module">
  import xxhash from "https://unpkg.com/xxhash-wasm/esm/xxhash-wasm.js";
</script>

UMD build

<script src="https://unpkg.com/xxhash-wasm/umd/xxhash-wasm.js"></script>

The global xxhash will be available.

Cloudflare Workers

If you are using Cloudflare Workers (workerd) you can use the installed npm package as is. The xxhash-wasm package is compatible with Cloudflare Workers.

import xxhash from "xxhash-wasm";

Importing it will pick the correct file base on the conditional import from the package.json.

Usage

The WebAssembly is contained in the JavaScript bundle, so you don't need to manually fetch it and create a new WebAssembly instance.

import xxhash from "xxhash-wasm";

// Creates the WebAssembly instance.
xxhash().then(hasher => {
  const input = "The string that is being hashed";

  // 32-bit version
  hasher.h32(input); // 3998627172 (decimal representation)
  // For convenience, get hash as string of its zero-padded hex representation
  hasher.h32ToString(input); // "ee563564"

  // 64-bit version
  hasher.h64(input); // 5776724552493396044n (BigInt)
  // For convenience, get hash as string of its zero-padded hex representation
  hasher.h64ToString(input); // "502b0c5fc4a5704c"
});

Or with async/await and destructuring:

// Creates the WebAssembly instance.
const { h32, h64 } = await xxhash();

const input = "The string that is being hashed";
// 32-bit version
h32(input); // 3998627172 (decimal representation)
// 64-bit version
h64(input); // 5776724552493396044n (BigInt)

Streaming Example

xxhash-wasm supports a crypto-like streaming api, useful for avoiding memory consumption when hashing large amounts of data:

const { create32, create64 } = await xxhash();

// 32-bit version
create32()
  .update("some data")
  // update accepts either a string or Uint8Array
  .update(Uint8Array.from([1, 2, 3]))
  .digest(); // 955607085

// 64-bit version
create64()
  .update("some data")
  // update accepts either a string or Uint8Array
  .update(Uint8Array.from([1, 2, 3]))
  .digest(); // 883044157688673477n

Node

It doesn't matter whether you are using CommonJS or ES Modules in Node (e.g. with "type": "module" in package.json or using the explicit file extensions .cjs or .mjs respectively), importing xxhash-wasm will always load the corresponding module, as both bundles are provided and specified in the exports field of its package.json, therefore the appropriate one will automatically be selected.

Using ES Modules

import xxhash from "xxhash-wasm";

Using CommonJS

const xxhash = require("xxhash-wasm");

Performance

For performance sensitive applications, xxhash-wasm provides the h** and h**Raw APIs, which return raw numeric hash results rather than zero-padded hex strings. The overhead of the string conversion in the h**ToString APIs can be as much as 20% of overall runtime when hashing small byte-size inputs, and the string result is often inconsequential (for example when simply checking if the the resulting hashes are the same). When necessary, getting a zero-padded hex string from the provided number or BigInt results is easily achieved via result.toString(16).padStart(16, "0") and the h**ToString APIs are purely for convenience.

The h**, h**ToString, and streaming APIs make use of TextEncoder.encodeInto to directly encode strings as a stream of UTF-8 bytes into the WebAssembly memory buffer, meaning that for string-hashing purposes, these APIs will be significantly faster than converting the string to bytes externally and using the Raw API. That said, for large strings it may be beneficial to consider the streaming API or another approach to encoding, as encodeInto is forced to allocate 3-times the string length to account for the chance the input string contains high-byte-count code units.

If possible, defer the encoding of the string to the hashing, unless you need to use the encoded string (bytes) for other purposes as well, or you are creating the bytes differently (e.g. different encoding), in which case it's much more efficient to use the h**Raw APIs instead of having to unnecessarily convert them to a string first.

Engine Requirements

In an effort to make this library as performant as possible, it uses several recent additions to browsers, Node and the WebAssembly specification. Notably, these include:

1. BigInt support in WebAssembly
2. Bulk memory operations in WebAssembly
3. TextEncoder.encodeInto

Taking all of these requirements into account, xxhash-wasm should be compatible with:

• Chrome >= 85
• Edge >= 79
• Firefox >= 79
• Safari >= 15.0
• Node >= 15.0

If support for an older engine is required, `xxhash-wasm@0.4.2` is available with much broader engine support, but 3-4x slower hashing performance.

API

const {
  h32,
  h32ToString,
  h32Raw,
  create32,
  h64,
  h64ToString,
  h64Raw,
  create64,
} = await xxhash();

Create a WebAssembly instance.

h32

h32(input: string, [seed: u32]): number

Generate a 32-bit hash of the UTF-8 encoded bytes of input. The optional seed is a u32 and any number greater than the maximum (0xffffffff) is wrapped, which means that 0xffffffff + 1 = 0.

Returns a u32 number containing the hash value.

h32ToString(input: string, [seed: u32]): string

Same as h32, but returning a zero-padded hex string.

h32Raw(input: Uint8Array, [seed: u32]): number

Same as h32 but with a Uint8Array as input instead of a string.

h64

h64(input: string, [seed: bigint]): bigint

Generate a 64-bit hash of the UTF-8 encoded bytes of input. The optional seed is a u64 provided as a BigInt.

Returns a u64 bigint containing the hash value.

h64ToString(input: string, [seed: bigint]): string

Same as h64, but returning a zero-padded hex string.

h64Raw(input: Uint8Array, [seed: bigint]): bigint

Same as h64 but with a Uint8Array as input instead of a string.

Streaming

type XXHash<T> {
  update(input: string | Uint8Array): XXHash<T>;
  digest(): T
}

The streaming API mirrors Node's built-in crypto.createHash, providing update and digest methods to add data to the hash and compute the final hash value, respectively.

create32([seed: number]): XXHash<number>

Create a 32-bit hash for streaming applications.

create64([seed: bigint]): XXHash<bigint>

Create a 64-bit hash for streaming applications.

Comparison to xxhashjs

xxhashjs is implemented in pure JavaScript and because JavaScript is lacking support for 64-bit integers, it uses a workaround with cuint. Not only is that a big performance hit, but it also increases the bundle size by quite a bit when it's used in the browser.

This library (xxhash-wasm) has the big advantage that WebAssembly supports u64 and also some instructions (e.g. rotl), which would otherwise have to be emulated. However, The downside is that you have to initialise a WebAssembly instance, which takes a little over 2ms in Node and about 1ms in the browser. But once the instance is created, it can be used without any further overhead. For the benchmarks below, the instantiation is done before the benchmark and therefore it's excluded from the results, since it wouldn't make sense to always create a new WebAssembly instance.

Benchmarks

Benchmarks are using Benchmark.js with random strings of different lengths. Higher is better

xxhash-wasm outperforms xxhashjs significantly, the 32-bit is up to 90 times faster (generally increases as the size of the input grows), and the 64-bit is up to 350 times faster (generally increases as the size of the input grows).

The 64-bit version is the faster algorithm but there is a small degree of overhead involved in using BigInts, and so it retains a performance advantage over all lengths over xxhashjs and the 32-bit algorithm above ~1000 bytes.

xxhash-wasm also significantly outperforms Node's built-in hash algorithms, making it suitable for use in a wide variety of situations, where non-cryptographic hashes are acceptable. Benchmarks from an x64 MacBook Pro running Node 17.3:

If suitable for your use case, the Raw API offers significant throughput improvements over the string-hashing API, particularly for smaller inputs, assuming that you have access to the Uint8Array already (see also the Performance section):

Bundle size

Both libraries can be used in the browser and they provide a UMD bundle. The bundles are self-contained, that means they can be included and used without having to add any other dependencies. The table shows the bundle size of the minified versions. Lower is better.

| | xxhashjs | xxhash-wasm | | -------------- | ---------- | ------------- | | Bundle size | 41.5kB | 11.4kB | | Gzipped Size | 10.3kB | 2.3kB |

Changelog

This changelog keeps all release notes in one place and mirrors the release notes from the GitHub releases, except for older versions, where no GitHub releases had been created.

v1.1.0

Cloudflare Workers Support

Cloudflare Workers disallow loading WebAssembly modules from an ArrayBuffer for security reasons.

They support conditional import from the workerd field in package.json, therefore xxhash-wasm now includes an additional workerd package that includes the WASM in a separate file. (#51)

You can install it with npm as usual and import it with:

import xxhash from "xxhash-wasm";

Performance

• Set state with .subarray() over .slice() to avoid unnecessary copy that will be discarded right afterwards (#52)

v1.0.2

• Add support for Typescript's nodenext module resolution (#33)

v1.0.1

• Export data types separately + fixed bigint data type (#28)

v1.0.0

This big release includes an up to a 3-4x performance improvement in most cases and a new streaming API similar to Node's built-in crypto API. To fully utilise the performance improvements, there are some breaking changes in the API and newer engine requirements.

3-4x Performance improvements

To achieve these substantial performance improvements, a handful of new features have been used, which are fairly recent additions to the browsers, Node and the WebAssembly specification. These include the following:

1. BigInt support in WebAssembly
2. Bulk memory operations in WebAssembly
3. TextEncoder.encodeInto

Taking all of these requirements into account, v1.0.0 should be compatible with:

• Chrome >= 85
• Edge >= 79
• Firefox >= 79
• Safari >= 15.0
• Node >= 15.0

If support for an older engine is required, `xxhash-wasm@0.4.2` is available with much broader engine support, but 3-4x slower hashing performance.

Besides the features regarding memory optimisations for WebAssembly, the biggest addition is the use of BigInt, which avoids the whole workaround that was previously used in order to represent u64 integers in JavaScript. That makes everything a lot simpler and faster, but that also brings some breaking changes of the 64-bit API.

The TextEncoder.encodeInto allows to encode the string as UTF-8 bytes directly into the WebAssembly memory, meaning that if you have the string and hash it directly, it will be faster than encoding it yourself and then using the Raw API. If possible, defer the encoding of the string to the hashing, unless you need to use the encoded string (bytes) for other purposes as well, or you are creating the bytes differently (e.g. different encoding), in which case it's much more efficient to use the h**Raw APIs instead of having to unnecessarily convert them to a string first.

Streaming API

The streaming API allows to build up the input that is being hashed in an iterative manner, which is particularly helpful for larger inputs which are collected over time instead of having it all at once in memory. It is kept in line with Node's crypto.createHash, hence the streams are initialised with create32/create64 and then .update(string | Uint8Array) is used to add an input, which can either be a string or a Uint8Array, and finally .digest() needs to be called to finalise the hash.

const { create32, create64 } = await xxhash();

// 32-bit version
create32()
  .update("some data")
  // update accepts either a string or Uint8Array
  .update(Uint8Array.from([1, 2, 3]))
  .digest(); // 955607085

// 64-bit version
create64()
  .update("some data")
  // update accepts either a string or Uint8Array
  .update(Uint8Array.from([1, 2, 3]))
  .digest(); // 883044157688673477n

Breaking Changes

64-bit seed as BigInt

64-bit hash APIs now use BigInt, where the seed is now a single BigInt instead of being split into the two halves seedHigh and seedLow. This makes it much simpler to use and avoids any workarounds for previous limitations.

h64(input: string, [seedHigh: u32, seedLow: u32]): string
h64Raw(input: Uint8Array, [seedHigh: u32, seedLow: u32]): Uint8Array

h64(input: string, [seed: BigInt]): BigInt
h64ToString(input: string, [seed: BigInt]): string
h64Raw(input: Uint8Array, [seed: BigInt]): BigInt

h32/h64 return numbers instead of strings

The hashes are numbers but were previously converted to a string of their a zero-padded hex string representations, mainly to keep the 32-bit in line with the 64-bit version, which could not be expressed by a single number without BigInt. This overhead is unnecessary for many applications and therefore the performance suffers. Now h32 returns a number and h64 a BigInt. For convenience, h32ToString and h64ToString have been added to get the hash as a string, which can also be achieved by converting them manually, e.g. hash64.toString(16).padStart(16, "0").

h32(input: string, [seed: u32]): string
h64(input: string, [seedHigh: u32, seedLow: u32]): string

h32(input: string, [seed: u32]): number
h64(input: string, [seed: BigInt]): BigInt

// New *ToString methods for convenience and to get old behaviour
h32ToString(input: string, [seed: u32]): string
h64ToString(input: string, [seed: BigInt]): string

v0.4.2

• Fix 64-bit hex representation when second part has leading zeros (#23)

v0.4.1

• Initialise TextEncoder lazily

v0.4.0

• TypeScript definitions
• h32Raw and h64Raw APIs for use with Uint8Array

v0.3.1

WebAssembly is optimised by binaryen:

• Faster
• Smaller

v0.3.0

New API to avoid reinitialising WASM instances

v0.2.0

Include a CommonJS bundle for Node.js