Install RustSys

RustSys 0.1.0 is unreleased and is not published on crates.io. Install from this repository with Rust 1.97.1 and Cargo. The CPU workflow needs no external synthesis program, Python runtime or GPU toolkit.

Build from source

Use RustEDA's pinned recursive checkout, or clone the integration branch directly (the standalone main branch can lag this documentation):

git clone --branch codex/unified-brand https://github.com/lab-emi/RustSys.git
cd RustSys

From the repository root:

cargo install --locked --path crates/apps/rustsys-cli
rustsys --version
rustsys synth --help

The default build includes the CUDA runtime, but ordinary --backend cpu synthesis does not require a GPU. For a CPU-only build:

cargo install --locked --path crates/apps/rustsys-cli --no-default-features

Install the companion workspace command if you want immutable revisions, library caching and JSON envelopes:

cargo install --locked --path crates/apps/rusteda-cli
rusteda --version

Cargo installs the binaries into its configured bin directory. Ensure that directory is on PATH. For development without installation, build with cargo build --release -p rustsys-cli -p rusteda-cli and use target/release/rustsys and target/release/rusteda.

What you need for a real design

Supply your synthesizable RTL, its include paths and parameters, a top module, and at least one Liberty library. RustSys emits a flat structural Verilog netlist that instantiates cells from that library. It does not supply a production PDK with the executable.

The quickstart uses a small synthetic library included in the repository. That library is useful for learning and regression tests; use the intended technology library for implementation work.

Platforms and resources

Linux is exercised in CI. Builds on other platforms have not received the same validation. Native CUDA mapping is available; a Metal runtime is planned. See GPU and timing workflows before selecting a non-default backend.

Start with --threads 1 when measuring or integrating a new design. The default is one worker per available CPU. Large designs also need explicit process memory and wall-clock limits; see resource guidance.

Continue with your first netlist, or use the agent workflow to build a revision-aware integration.