Download the PHP package awesome/simdjson_plus without Composer

awesome/simdjson_plus

🚀 Blazing-fast JSON encoding and decoding for PHP, powered by the simdjson project.

This is a fork of JakubOnderka/simdjson_php (which is a fork of crazyxman/simdjson_php)

Since the simdjson PECL extension seems to be unmaintained, or at least slow to accept PRs for improvements, we packaged this up under a new name (simdjson_plus) to avoid naming conflicts and published it on Packagist (instead of PECL) for easier installation.

It's a drop-in replacement for the PECL extension, with additional features from JakubOnderka, such as accelerated JSON encoding (not just decoding) and optimizations.

Performance Comparison: How Fast is simdjson_plus?

Tests were conducted using PHP 8.3 on an Apple M1 Max. For test specification see TwitterDecodeBench.php and TwitterEncoderBench.php.

Additionally, simdjson_plus reduces memory usage compared to json_decode(). For example, when decoding twitter.json, memory consumption drops from 3.01 MB to 2.47 MB due to efficient array key deduplication.

Related SIMD-accelerated PHP extensions

• crc_fast PHP extension for SIMD-accelerated CRC calculations at >100GiB/s.
• simdutf PHP extension for Unicode validation and transcoding at billions of characters per second using the simdutf project.

Requirements

• PHP 8.1+ (PHP 8.2+ recommended for maximum performance)
• g++ (version 7 or better) or clang++ (version 6 or better)
• A 64-bit system with a command-line shell (e.g., Linux, macOS, FreeBSD)

Changes

See the change log.

Installing

Use Composer to install this library using PIE:

Compilation Instructions for Linux

To compile simdjson_plus, run the following commands:

Once installed, add this line to your php.ini file:

Usage Examples

Encoder

Most of available options of default json_encode() method are not supported by simdjson_encode() as they are usually useless.

simdjson_encode($value) method has similar behaviour as json_encode($value, JSON_UNESCAPED_SLASHES | JSON_UNESCAPED_UNICODE | JSON_THROW_ON_ERROR)

Supported options are:

• SIMDJSON_PRETTY_PRINT - use whitespace in returned data to format it
• SIMDJSON_INVALID_UTF8_SUBSTITUTE - convert invalid UTF-8 characters to \0xfffd (Unicode Character 'REPLACEMENT CHARACTER' �)
• SIMDJSON_INVALID_UTF8_IGNORE - ignore invalid UTF-8 characters
• SIMDJSON_APPEND_NEWLINE - append new line character (\n) to end of encoded string. This is useful when encoding data to JSONL format as PHP strings are immutable.

Differences are:

• uses different algorithm to convert floating-point number to string, so string format can be slightly different
• even when JSON_UNESCAPED_UNICODE is enabled, PHP json_encode() escapes some Unicode chars that do not need to be escaped. simdjson_encode() escape just Unicode chars that needs to be escaped by JSON spec.
• simdjson will throw SimdJsonEncoderException exception in case of error

Base64 encoding

JSON format do not support binary data. Common way how to transfer binary data in JSON encoding is using base64 encoding. If you need to include base64 encoded value into JSON, you can use SimdJsonBase64Encode class that offers optimised converting to base64 value into JSON and use less memory. As creating new object in PHP is relatively slow, this approach make sense for string longer than 1 kB.

You can also use base64url encoding (RFC 4648 §5) by setting second argument to true: new SimdJsonBase64Encode($fileContent, true);

Encode to stream

For large data sets, simdjson_plus provides the simdjson_encode_to_stream() function to save data directly to a file or output buffer.

Decoder

There are some differences from json_decode() due to the implementation of the underlying simdjson library. This will throw a SimdJsonDecoderException if simdjson rejects the JSON.

Note that the simdjson PECL is using a fork of the simdjson C library to imitate php's handling of integers and floats in JSON.

1) The maximum string length that can be passed to simdjson_decode() is 4GiB (4294967295 bytes). json_decode() can decode longer strings.

2) The handling of max depth is counted slightly differently for empty vs non-empty objects/arrays. In json_decode, an array with a scalar has the same depth as an array with no elements. In simdjson_decode, an array with a scalar is one level deeper than an array with no elements. For typical use cases, this shouldn't matter. (e.g. simdjson_decode('[[]]', true, 2) will succeed but json_decode('[[]]', true, 2) and simdjson_decode('[[1]]', true, 2) will fail.)

Decode from stream

If you need to decode a big file from JSON format that you want to save to a file or send to a user, you can use the simdjson_decode_from_stream method.

Benchmarks

See the benchmark folder for more benchmarks.