Struct GpuContext 

pub struct GpuContext {
    pub device: Device,
    pub queue: Queue,
}

Available on crate feature wgpu only.

A headless wgpu device + queue for running the compute kernel.

Callers that already own a wgpu::Device should build StencilEvalPipeline and StencilTableGpu directly; this is a convenience for standalone use and tests. new returns None when no adapter is available.

Fields

device: Device

The wgpu device.

queue: Queue

The wgpu queue.

Implementations

impl GpuContext

pub fn new() -> Option<Self>

Create a headless compute context, or None if no GPU adapter is available (so callers can gracefully fall back to the CPU path).

pub async fn new_async() -> Option<Self>

Async variant of new.

pub fn evaluate( &self, table: &StencilTable, input: &[f32], components: u32, ) -> Result<Vec<f32>, KernelError>

Evaluate table over a tightly-packed input buffer of components-vectors, returning the packed output.

Equivalent to StencilTable::interpolate for [f32; components] data. Builds all transient GPU resources per call; for repeated evaluation hold a StencilEvalPipeline + StencilTableGpu and reuse buffers via evaluate_stencils.

pub fn evaluate_sparse( &self, table: &StencilTable, input: &[f32], components: u32, affected: &[u32], prior_output: &[f32], ) -> Result<Vec<f32>, KernelError>

Sparse re-evaluation: recompute only the affected output rows from input, splicing them into prior_output (the previous dense result). Pair with RefinementResult::affected_outputs for incremental edits.

The result equals a full dense re-evaluation when affected covers every output row that actually changed.

Trait Implementations

impl Debug for GpuContext

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.