cvxgenrust

cvxgenrust takes a parameterized CVXPY optimization problem and generates a Rust solver crate tailored to that problem family. The generated crate reconstructs canonical cone-program data and solves it with Clarabel. It also includes a Python wrapper that can be registered as a custom CVXPY solve method for prototyping.

More details can be found in the associated paper.

Installation

Install the released package from PyPI with:

pip install cvxgenrust

Generated solver projects use Rust, Cargo, Clarabel, and, when the Python wrapper is enabled, a PyO3/maturin build. Install a stable Rust toolchain before building or importing generated extension wrappers.

For development from this repository, use uv to manage dependencies. Once uv is installed, run:

make sync

This installs the default development environment defined by the repository Makefile.

Quick Start

Generate a small nonnegative least-squares solver as a Rust crate:

import cvxpy as cp
import cvxgenrust as cgr

m, n = 3, 2
A = cp.Parameter((m, n), name="A")
b = cp.Parameter(m, name="b")
x = cp.Variable(n, name="x")

problem = cp.Problem(
    cp.Minimize(cp.sum_squares(A @ x - b)),
    [x >= 0],
)

project = cgr.generate_code(
    problem,
    code_dir="nonneg_ls_cgr",
    module_name="nonneg_ls",
)
print("generated:", project.output_dir)

You should always set name= on CVXPY parameters and variables. The generated Rust setters, extractors, metadata, and Python wrapper use those names after code generation.

You can build and run the generated Rust project with:

cd nonneg_ls_cgr
cargo run --example solve

By default, generate_code also compiles the generated Python extension wrapper into the generated project's python/ directory. Pass wrapper=False to only write the Rust crate and Python wrapper sources.

An HTML documentation of the generated project is written to nonneg_ls_cgr/README.html, where you can find more details of the generated code and usage examples.

Structured parameters

cvxgenrust supports real dense, diagonal, symmetric, PSD, NSD, and explicitly sparse CVXPY parameters. Declaring invariant structure with sparsity= keeps known zero entries out of the generated canonical matrices; declaring the same parameter as dense can increase generated code size and solver work. The coordinates excluded by a parameter's sparsity= pattern must remain structural zeros for every update; use a dense parameter if any excluded entry may later become nonzero. Complex and Hermitian parameter layouts are not supported.

Each generated README.html reports the logical shape, packed size, offset, layout, and exact Rust setter order for every parameter. Its generated Python example also shows how to assign values for that problem's layouts.

Related projects

• CVXPYgen: C code generation from CVXPY problems.
• CVXGEN: C code generation for convex optimization in MATLAB.