fast-md5

A small #![no_std] MD5 implementation with hand-written assembly cores for x86_64 and aarch64, plus a portable Rust fallback for every other target.

The assembly was ported from animetosho/md5-optimisation, whose author has placed the original work in the public domain (discussion #4). This crate is licensed BSD-2-Clause for the Rust glue and tests; the assembly stanzas inherit their public-domain status from upstream.

⚠️ MD5 is broken for cryptographic use

MD5 is trivially vulnerable to collision attacks and must not be used for digital signatures, certificate fingerprints, password hashing, or any other security-sensitive integrity check. This crate exists to support:

Legacy protocols that bake MD5 into their wire format (RADIUS, NTLM, parts of TLS 1.0, etc.) — this is the motivating use-case.
Non-cryptographic content addressing, deduplication, and checksumming where collisions are not adversarial.

For new designs, use BLAKE3 or SHA-256.

Usage

Add it to your Cargo.toml:

[dependencies]
fast-md5 = "1.0.0"

One-shot hashing:

let digest = fast_md5::digest(b"The quick brown fox jumps over the lazy dog");
assert_eq!(hex::encode(digest), "9e107d9d372bb6826bd81d3542a419d6");

Streaming:

let mut h = fast_md5::Md5::new();
h.update(b"The quick brown fox ");
h.update(b"jumps over the lazy dog");
let digest = h.finalize();

Architecture support

Target	Implementation
`x86_64`	Inline assembly (NoLEA + GOpt schedule)
`aarch64`	Inline assembly
everything else	Portable Rust fallback

The compression function is selected at compile time — there is no runtime dispatch and no feature flags to remember.

`no_std`

The crate is #![no_std] and performs no heap allocation. The Md5 state is <128 bytes and can live on the stack.

Testing strategy

The fallback (portable Rust) transform is compiled in cfg(test) on every target, regardless of host architecture, so that module-level unit tests of the compression function can run everywhere. On assembly hosts the test suite cross-checks the active assembly transform against the fallback block-for-block over randomized inputs.

The full RFC 1321 §A.5 test suite plus the "a" * 1_000_000 long-input vector are covered.

cargo test

Benchmarks

Throughput benchmarks compare fast-md5 against:

md-5 — the RustCrypto MD5 crate.
AWS-LC, via the aws-lc-sys FFI bindings. (aws-lc-rs intentionally does not expose MD5 in its public API, so the bench links the C entrypoint directly.)

Run them with:

cargo bench

On Apple Silicon (M-series) you should see fast-md5 come in roughly on par with AWS-LC and ~5–10% faster than RustCrypto across the 64 B – 1 MiB range. Exact numbers depend on your CPU; criterion will write per-input HTML reports under target/criterion/.

Acknowledgements

All algorithmic credit belongs to @animetosho for the optimization work and the public-domain release of the original C header. This crate is just a faithful port to Rust inline assembly with some scaffolding around it.

License

BSD-2-Clause. See LICENSE.

fast-md5 1.0.0