ruda 0.1.52

Ruda GPU kernel API and portable runtime with CUDA, HIP and WGPU backends.
docs.rs failed to build ruda-0.1.52
Please check the build logs for more information.
See Builds for ideas on how to fix a failed build, or Metadata for how to configure docs.rs builds.
If you believe this is docs.rs' fault, open an issue.
Visit the last successful build: ruda-0.1.48

ruda

Unified entry point for Ruda GPU kernels and compute backends. Select CUDA, HIP, or WGPU through Cargo features. The portable host runtime lives in ruda-runtime and remains available through ruda::runtime.

Interfaces

  • runtime::client and runtime::server: submit work and implement compute services.
  • runtime::storage, memory_management, and allocator: manage backend allocations.
  • runtime::backend, compiler, and kernel: implement runtime and compiled-kernel contracts.
  • dsl and prelude: kernel DSL and launch interfaces, enabled by any GPU backend.
  • cuda, hip, and wgpu: the selected driver's public API.

Usage

Cargo package: ruda. Rust import: ruda.

cargo add ruda --features cuda

For AMD HIP or a cross-vendor WGPU backend, choose one of:

cargo add ruda --features hip
cargo add ruda --features wgpu

CUDA requires a compatible NVIDIA driver and CUDA Toolkit; HIP requires a compatible AMD GPU and ROCm/HIP environment; WGPU requires a compatible GPU and graphics driver. Cargo adds the dependency; it does not install these drivers.

Use ruda::cuda::{CudaDevice, CudaRuntime}, ruda::hip::{AmdDevice, HipRuntime}, or ruda::wgpu::{WgpuDevice, WgpuRuntime} with ruda::prelude::*.

From the source workspace, run the same vector-add kernel with the selected backend:

cargo run -p ruda --example vector-add --features cuda
cargo run -p ruda --example vector-add --features hip
cargo run -p ruda --example vector-add --features wgpu

When enabling multiple backends, select one explicitly for the example:

cargo run -p ruda --example vector-add --features cuda,wgpu -- wgpu

Features

Default features: runtime-default.

Feature Purpose
cuda NVIDIA CUDA backend, exposed as ruda::cuda.
hip AMD ROCm/HIP backend, exposed as ruda::hip.
wgpu Cross-vendor WGPU backend, exposed as ruda::wgpu.
direct-ptx Enable CUDA and its direct PTX compiler; select it with RUDA_CUDA_COMPILER=ptx.
kernel Expose the kernel DSL and prelude without selecting a GPU backend.
runtime Expose the runtime modules.
runtime-std Enable host configuration and standard-library integration.
runtime-storage-bytes Enable byte-backed storage support.
runtime-tracing Enable runtime tracing.

Links