tachyon

tachyon is a high-performance algorithm library written in Rust, providing SIMD-accelerated implementations for both x86 (AVX/SSE) and ARM (NEON) architectures, to outperform even hand-tuned assembly in many cases.

Features

• Optimized for both x86_64 (SSE, AVX) and aarch64 (NEON) platforms. (AVX-512? Still waiting for it to stabilize... maybe someday.)
• Focused on performance — carefully tuned SIMD intrinsics for critical code paths.
• SIMD implementations often outperform well-known on ARM NEON.
• Designed for performance-critical applications like embedded systems, secure networking, and data processing.

Usage

Add to your Cargo.toml:

tachyon = { git = "https://github.com/SongXiaoXi/tachyon-rs" }

or use cargo:

cargo add tachyon --git https://github.com/SongXiaoXi/tachyon-rs

MD5 Example

use tachyon::crypto::hash::md5;

let digest = md5::compute(b"hello world");

More examples can be found in the bench directory.

Supported Algorithms

Cryptographic

Block Ciphers

• AES-128

Stream Ciphers

• ChaCha20

MACs

• GMAC
• Poly1305

AEAD

• AES-128-GCM
• ChaCha20-Poly1305

Hash Functions

• MD5 (do not use except for non-cryptographic checksums)
• SHA-1 (for legacy compatibility only)
• SHA-256
• SHA-512

Performance Highlights

AES-128-GCM encryption of 64 KiB:

Platform	tachyon-rs	ring(v0.17.14)	OpenSSL
Apple M4	12006.94 MiB/s	8751.87 MiB/s	11043.29 MiB/s(3.5.0)
Apple M1	8334.81 MiB/s	7122.30 MiB/s	8142.95 MiB/s(3.5.0)
Apple A16	9224.50 MiB/s	7813.08 MiB/s	7329.87 MiB/s(3.2.1)
Apple A12z	4855.90 MiB/s	4472.18 MiB/s	4477.18 MiB/s(3.2.1)
Snapdragon 888	4658.82 MiB/s	3950.03 MiB/s	3972.17 MiB/s(3.4.1)
Snapdragon 6Gen1	2449.88 MiB/s	2280.68 MiB/s	2176.36 MiB/s(3.4.1)
Ryzen 9 9950X	33178.55 MiB/s	15612.82 MiB/s	33120.97 MiB/s(3.4.1)
Ryzen Z1 Extreme	18197.73 MiB/s	13798.14 MiB/s	16407.54 MiB/s(3.5.0)

Chacha20-Poly1305 encryption of 64 KiB:

Platform	tachyon-rs	tachyon-rs (tune-cpu=generic)	ring(v0.17.14)	OpenSSL
Apple M4	1729.85 MiB/s	2672.39 MiB/s	2168.00 MiB/s	2355.95 MiB/s(3.5.0)
Apple M1	1147.70 MiB/s	2010.69 MiB/s	1620.36 MiB/s	1816.42 MiB/s(3.5.0)
Apple A16	1240.64 MiB/s	1926.21 MiB/s	1738.82 MiB/s	1932.54 MiB/s(3.2.1)
Snapdragon 888	1435.37 MiB/s	N/A	1325.19 MiB/s	1277.12 MiB/s(3.4.1)
Snapdragon 6Gen1	863.24 MiB/s	N/A	759.92 MiB/s	654.70 MiB/s(3.4.1)
i7-3770	1121.77 MiB/s	N/A	1151.09 MiB/s	1105.63 MiB/s(3.4.1)

MD5 hashing of 64 KiB:

Platform	tachyon-rs	md5(v0.7.0)	OpenSSL
Apple M4	965.56 MiB/s	848.33 MiB/s	981.83 MiB/s(3.5.0)
Apple M1	700.03 MiB/s	607.06 MiB/s	692.56 MiB/s(3.5.0)
Apple A16	751.58 MiB/s	652.44 MiB/s	682.52 MiB/s(3.2.1)
Apple A12z	544.82 MiB/s	445.10 MiB/s	517.96 MiB/s(3.2.1)
Snapdragon 888	620.08 MiB/s	477.37 MiB/s	588.63 MiB/s(3.4.1)
Snapdragon 6Gen1	483.62 MiB/s	369.05 MiB/s	456.73 MiB/s(3.4.1)
i7-10700K	1096.69 MiB/s	609.86 MiB/s	1048.98 MiB/s(3.5.0)
i7-3770	809.03 MiB/s	455.70 MiB/s	757.12 MiB/s(3.4.1)
Ryzen 9 9950X	1243.69 MiB/s	813.45 MiB/s	1186.78 MiB/s(3.4.1)
BCM2711 (RPi4)	323.22 MiB/s	186.70 MiB/s	279.40 MiB/s(3.5.0)

Tip:

• On older Apple devices (e.g., A7–A10), compiling with rust nightly and -Ztune-cpu=generic may produce more favorable NEON code generation.
• avx512+vaes is available in the nightly channel of Rust.

Security Notice

This library is designed for performance benchmarking and experimentation.
It has not been formally audited for cryptographic soundness.
Do not use in production unless you fully understand the risks.

Acknowledgements

This project builds upon the work and ideas of several excellent Rust cryptography projects:

• crypto2: for its clean and pragmatic API design around cryptographic primitives, as well as software fallback implementations of several algorithms.
• RustCrypto AES: used as a software fallback for the AES fixslicing implementation.