1# crc32fast [![Build Status][travis-img]][travis] [![Crates.io][crates-img]][crates] [![Documentation][docs-img]][docs] 2 3[travis-img]: https://travis-ci.com/srijs/rust-crc32fast.svg?branch=master 4[travis]: https://travis-ci.com/srijs/rust-crc32fast 5[crates-img]: https://img.shields.io/crates/v/crc32fast.svg 6[crates]: https://crates.io/crates/crc32fast 7[docs-img]: https://docs.rs/crc32fast/badge.svg 8[docs]: https://docs.rs/crc32fast 9 10_Fast, SIMD-accelerated CRC32 (IEEE) checksum computation_ 11 12## Usage 13 14### Simple usage 15 16For simple use-cases, you can call the `hash` convenience function to 17directly compute the CRC32 checksum for a given byte slice: 18 19```rust 20let checksum = crc32fast::hash(b"foo bar baz"); 21``` 22 23### Advanced usage 24 25For use-cases that require more flexibility or performance, for example when 26processing large amounts of data, you can create and manipulate a `Hasher`: 27 28```rust 29use crc32fast::Hasher; 30 31let mut hasher = Hasher::new(); 32hasher.update(b"foo bar baz"); 33let checksum = hasher.finalize(); 34``` 35 36## Performance 37 38This crate contains multiple CRC32 implementations: 39 40- A fast baseline implementation which processes up to 16 bytes per iteration 41- An optimized implementation for modern `x86` using `sse` and `pclmulqdq` instructions 42- An optimized implementation for `aarch64` using `crc32` instructions 43 44Calling the `Hasher::new` constructor at runtime will perform a feature detection to select the most 45optimal implementation for the current CPU feature set. 46 47| crate | version | variant | ns/iter | MB/s | 48| ----------------------------------- | ------- | --------- | ------- | ---- | 49| [crc](https://crates.io/crates/crc) | 1.8.1 | n/a | 4,926 | 207 | 50| crc32fast (this crate) | 1.0.0 | baseline | 683 | 1499 | 51| crc32fast (this crate) | 1.0.0 | pclmulqdq | 140 | 7314 | 52 53## Memory Safety 54 55Due to the use of SIMD intrinsics for the optimized implementations, this crate contains some amount of `unsafe` code. 56 57In order to ensure memory safety, the relevant code has been fuzz tested using [afl.rs](https://github.com/rust-fuzz/afl.rs) with millions of iterations in both `debug` and `release` build settings. You can inspect the test setup in the `fuzz` sub-directory, which also has instructions on how to run the tests yourself. 58 59On top of that, every commit is tested using an address sanitizer in CI to catch any out of bounds memory accesses. 60 61Even though neither fuzzing nor sanitization has revealed any safety bugs yet, please don't hesitate to file an issue if you run into any crashes or other unexpected behaviour. 62 63## Available feature flags 64 65### `std` (default: enabled) 66 67This library supports being built without the Rust `std` library, which is useful for low-level use-cases such as embedded where no operating system is available. To build the crate in a `no_std` context, disable the default `std` feature. 68 69Note: Because runtime CPU feature detection requires OS support, the specialized SIMD implementations will be unavailable when the `std` feature is disabled. 70 71### `nightly` (default: disabled) 72 73This feature flag enables unstable features that are only available on the `nightly` channel. Keep in mind that when enabling this feature flag, you 74might experience breaking changes when updating compiler versions. 75 76Currently, enabling this feature flag will make the optimized `aarch64` implementation available. 77 78## License 79 80This project is licensed under either of 81 82- Apache License, Version 2.0, ([LICENSE-APACHE](LICENSE-APACHE) or 83 http://www.apache.org/licenses/LICENSE-2.0) 84- MIT license ([LICENSE-MIT](LICENSE-MIT) or 85 http://opensource.org/licenses/MIT) 86 87at your option. 88 89### Contribution 90 91Unless you explicitly state otherwise, any contribution intentionally submitted 92for inclusion in this project by you, as defined in the Apache-2.0 license, 93shall be dual licensed as above, without any additional terms or conditions. 94
