pub enum Device {
Show 14 variants
Cpu,
Metal,
Mlx,
Ane,
Cuda,
Rocm,
OneApi,
Tpu,
Hexagon,
Gpu,
Vulkan,
OpenGl,
DirectX,
WebGpu,
}Expand description
Target device for graph execution.
Each variant maps to a backend crate gated by a Cargo feature.
Use Device::is_available() to check if the feature is enabled.
Variants§
Cpu
CPU with SIMD (NEON/AVX) + optional BLAS.
Metal
GPU via Apple Metal (Metal Performance Shaders).
Mlx
Apple MLX framework (unified memory GPU).
Ane
Apple Neural Engine.
Cuda
NVIDIA GPU via native CUDA (cuBLAS, cuDNN).
Rocm
AMD GPU via ROCm/HIP.
OneApi
Intel GPU (Arc / Data Center Max) via oneAPI Level Zero.
Tpu
Google TPU via libtpu’s PJRT plugin (no Python).
Hexagon
Qualcomm Hexagon NPU via QNN (AI Engine Direct).
Gpu
Portable GPU via wgpu (Metal/Vulkan/DX12/WebGPU).
Vulkan
Vulkan compute shaders.
OpenGl
OpenGL compute shaders (legacy).
DirectX
DirectX 12 compute (Windows).
WebGpu
WebGPU (WASM target).
Implementations§
Source§impl Device
impl Device
Sourcepub fn name(self) -> &'static str
pub fn name(self) -> &'static str
Human-readable name (no engine-layer info).
is_available / available live in rlx-runtime since they
consult the engine’s backend registry — keeping them out of
the driver layer preserves the one-way dep direction.