Accel: GPGPU Framework for Rust

crate	crates.io	docs.rs	GitLab Pages
accel				CUDA-based GPGPU framework
accel-core				Helper for writing device code
accel-derive				Procedural macro for generating kernel code

Requirements

• Minimum Supported Rust Version (MSRV) is 1.42.0
• Install CUDA on your system
  • accel depends on CUDA Device APIs through rust-cuda/cuda-sys
  • accel does not depend on CUDA Runtime APIs. It means that a compiled binary requires only libcuda.so at runtime, which is far lighter than entire CUDA development toolkit.
• Setup NVPTX target of Rust
  • Install nightly-2020-05-01 toolchain with nvptx64-nvidia-cuda target, and rust-ptx-linker
  • There is an setup script for them:

curl -sSL https://gitlab.com/termoshtt/accel/raw/master/setup_nvptx_toolchain.sh | bash

Or, you can use docekr container

Limitations

This project is still in early stage. There are several limitations as following:

• For runtime on CPU

  • Windows and macOS are not supported
  • f64 and Complex number supports are missing
  • Texture/Surface object handling is missing
  • Async features based on CUDA Stream and Events are disabled until async/.await support

• For writting GPU kernel code

  • libstd cannot be used in writting kernel
  • Rust slice cannot be used in writing kernel
  • Shared memory cannot be used

Contribution

This project is developed on GitLab and mirrored to GitHub.

Sponsors

• RICOS Co. Ltd
  • GPU instances for CI and development

Links

Projects which accel depends on:

• rust-cuda/cuda-sys: CUDA Runtime and Driver API binding to Rust
• denzp/rust-ptx-linker: Linker for PTX files generated by rustc

Related Projects:

• rust-cuda/wg: Working group for Rust CUDA Team
• denzp/rust-ptx-builder: Another CUDA kernel builder from Rust crate
• bheisler/RustaCUDA: Another CUDA-based Rust flamework