GitHub - ovasylenko/g6k-rs

g6k-rs is a research-oriented Rust library for lattice reduction, enumeration, and sieving on CPUs, Apple Metal, and NVIDIA CUDA.

The crate includes floating-point and arbitrary-precision reduction, exact invariant checks, deterministic examples, and optional GPU backends. It is an experimental implementation: APIs may change, and it has not been audited for use in production cryptographic systems.

Try it

The quickest path uses only the portable floating-point feature set:

git clone https://github.com/ovasylenko/g6k-rs.git
cd g6k-rs
./scripts/quickstart.sh

The script runs a deterministic LLL/BKZ example and verifies its determinant invariant. The first build compiles dependencies and therefore takes longer than subsequent runs.

Add the crate to a Rust project with:

[dependencies]
g6k-rs = "0.1"

use g6k_rs::{LLLParams, LatticeBasis, lll_reduce};

let mut basis = LatticeBasis::from_rows(vec![
    vec![1.0, 1.0],
    vec![-1.0, 2.0],
]);
lll_reduce(&mut basis, &LLLParams::default()).unwrap();

The default build enables arbitrary-precision support through the mpz feature. Set default-features = false for the floating-point API only.

What is implemented

• Floating-point LLL, BKZ, deep insertion, slide reduction, and self-dual reduction.
• Babai-style CVP, pruned enumeration, rerandomized enumeration, and pruning profile optimization.
• BDGL, BGJ1, and HK3-style lattice sieves with deterministic CPU and parallel paths.
• Arbitrary-precision lattice arithmetic and reduction through rug/GMP/MPFR.
• Optional Metal and CUDA sieve, enumeration, and Seysen-conditioning paths.
• Exact or independently recomputed checks for determinant preservation, coordinate reconstruction, integral output, and returned candidate norms.

See the examples guide for task-oriented snippets and project status for maturity and limitations.

Relationship to G6K and fplll

The name describes the project's goal: exploring a Rust implementation of the lattice-reduction and sieving problem space associated with the General Sieve Kernel. It is not an official port, release, or drop-in replacement for fplll/G6K.

No claim is made that g6k-rs is faster, more complete, or a compatible replacement. Read PROVENANCE.md before redistributing or relicensing the project.

Verification

# Portable floating-point build and tests
cargo test --no-default-features

# Default build, including arbitrary precision
cargo test

# Formatting and linting
cargo fmt --check
cargo clippy --all-targets -- -D warnings

The last locally observed default-platform run completed 1,813 tests with no failures; ignored tests include slow probes and hardware-dependent paths. That number is a development reference, not a substitute for the current CI result. GPU tests require the corresponding feature, toolchain, and device.

GPU features

• metal-gpu (macOS): Metal sieve kernels, GPU Seysen conditioning, and Metal-backed reduction entry points.
• metal-fp16: experimental half-precision database storage on top of metal-gpu.
• cuda-gpu (Linux/NVIDIA): CUDA sieve, enumeration, and Seysen kernels. The build requires nvcc; set G6K_CUDA_ARCH to the target architecture, such as sm_89.

Use the GPU testing guide for build and validation commands. Hardware measurements and the rules for making performance claims live in BENCHMARKING.md.

Documentation

• Examples — how to use each major workflow.
• Status and limitations — what is stable, experimental, or hardware-dependent.
• Benchmarking — reproducible commands and historical measurements.
• Provenance — independent-authorship statement, research sources, and contribution policy.
• Algorithm and paper map — fidelity notes for research modules.
• Contributing — required checks and evidence standards.

License

Licensed under either the Apache License 2.0 or the MIT License, at your option.

Maintainer: Oleksii Vasylenko (hello@ovasylenko.com).