Enum DeviceBackend 

pub enum DeviceBackend {
    Cpu,
}

Active backend selected at runtime.

Obtain via DeviceBackend::auto_detect for automatic selection, or DeviceBackend::from_preference to request a specific backend.

Feature gating

• Cuda variant requires --features cuda.
• Avx2 variant requires --features avx2 on an x86_64 host.
• Cpu is always present and is the fallback of last resort.

Variants

Cpu

Scalar CPU path, delegates to zer-compare (Rayon parallel).

Implementations

impl DeviceBackend

pub fn auto_detect() -> Self

Auto-detect the best available backend: CUDA → AVX2 → CPU scalar.

Never panics; always returns a usable backend. Tracing output explains which path was selected and why alternatives were skipped.

pub fn cpu() -> Self

Force the scalar CPU backend regardless of available hardware.

Useful in tests where deterministic, non-SIMD behaviour is required.

pub fn from_preference(pref: BackendPreference) -> Result<Self, GpuError>

Request a specific backend.

Returns Err(BackendUnavailable) when:

• The requested feature flag was not compiled in.
• The hardware initialisation fails (e.g. no CUDA GPU present).
• The requested ISA extension is absent at runtime (AVX2).

pub fn run<K: Kernel>(&self, input: K::Input<'_>) -> Result<K::Output, GpuError>

where Self: KernelDispatch<K>,

Dispatch kernel K on this backend.

pub fn name(&self) -> &'static str

Human-readable name of the active backend.

pub fn is_gpu(&self) -> bool

true when a GPU backend (CUDA or Vulkan) is active.

pub fn is_accelerated(&self) -> bool

true when this backend provides any hardware acceleration (GPU or SIMD), i.e. it is not the scalar CPU fallback.

pub fn available_vram_bytes(&self) -> Option<u64>

Available VRAM in bytes, or None for CPU/AVX2 paths.

pub fn total_vram_bytes(&self) -> Option<u64>

Total (installed) VRAM in bytes, or None for CPU/AVX2 paths.

Trait Implementations

impl KernelDispatch<EmReduce> for DeviceBackend

fn dispatch(&self, input: EmReduceInput<'_>) -> Result<EmReduceOutput, GpuError>

impl KernelDispatch<HelloBackend> for DeviceBackend

fn dispatch( &self, input: HelloBackendInput, ) -> Result<HelloBackendOutput, GpuError>