The Rust-CUDA project, an open-source initiative aimed at enabling CUDA compute kernels to be written in the Rust programming language, has officially been restarted. After three years of inactivity since its last release (v0.3 in February 2022), the developers have resumed work and are looking for additional contributors to help advance the project.
Bringing CUDA to Rust
The Rust-CUDA project focuses on compiling Rust code into NVIDIA’s NVVM IR, a derivative of LLVM IR optimized for GPU computing. This intermediate representation (IR) is then converted into PTX (Parallel Thread Execution) code, which runs directly on NVIDIA GPUs.
The rebooted project aims to modernize the codebase, update dependencies, and align with other Rust-based GPU efforts, such as:
- Rust GPU – A similar project targeting SPIR-V for Vulkan-based GPUs.
- Rustc PTX backend – An experimental Rust compiler backend that generates CUDA PTX IR.
- Cudarc – A Rust crate providing high-level CUDA abstractions.
Short-Term Roadmap
The immediate priorities for Rust-CUDA include:
- Merging outstanding pull requests to incorporate pending contributions.
- Cleaning up outdated issues in the repository.
- Updating dependencies to support the latest Rust toolchain and CUDA versions.
- Launching a new project website with updated documentation and examples.
Medium-Term Goals
Once the project is stabilized, developers plan to explore:
- Rust-C++ interoperability, leveraging Rust’s C++ interop features for better CUDA integration.
- Closer collaboration with Rust’s PTX backend to evaluate shifting from NVVM IR to raw PTX.
- Aligning APIs with Rust GPU to unify CUDA and Vulkan development.
Long-Term Vision
Looking further ahead, Rust-CUDA developers hope to:
- Integrate with the Rust compiler, making Rust-CUDA an officially supported backend.
- Unify GPU programming in Rust, abstracting away platform-specific differences.
- Evolve the Rust language to enhance support for GPU programming.
Call for Contributors
The Rust-CUDA team is actively seeking maintainers, contributors, and users to help shape the future of CUDA programming in Rust. Developers interested in participating can find more details on the Rust-GPU GitHub blog.
With this reboot, Rust-CUDA is poised to become a powerful tool for GPU programming in Rust, bridging the gap between high-performance computing and safe, modern software development.
