Rust Running on Every GPU

9 months ago

A demo showcases a single Rust codebase running on all major GPU platforms including CUDA, SPIR-V, Metal, DirectX 12, WebGPU, and CPU fallback.
Traditional GPU programming uses specialized languages like WGSL, GLSL, MSL, HLSL, Slang, or Triton, increasing complexity.
Rust projects like Rust GPU, Rust CUDA, and Naga enable compiling standard Rust code directly to GPU targets.
The demo implements a bitonic sort with shared compute logic across all targets, written entirely in Rust.
Backend selection is done via Rust feature flags and compilation targets, allowing dynamic GPU technology selection.
Kernels are compiled to the appropriate device format and embedded into the binary at build time.
Rust's no_std support is crucial for GPU programming, as the standard library isn't available on GPUs.
Conditional compilation, newtypes, enums, traits, and generics are leveraged for type-safe and efficient GPU code.
The project uses Rust's module system, workspaces, and standard tools like rustfmt and clippy for organization and quality.
GPU kernel code can be tested on the CPU using standard Rust testing tools, improving development and debugging.
Despite progress, the developer experience has rough edges, including compiler backend integration and debugging challenges.
The Rust community is encouraged to contribute to improving the GPU programming experience in Rust.

Hasty Briefsbeta