Struct InterpolatePlan 

pub struct InterpolatePlan<T: Element> { /* private fields */ }

interpolate plan.

Spatial resample of an NCHW input. PyTorch F.interpolate. Coordinate mapping: src = (dst + 0.5) * (src_size / dst_size) - 0.5 (align_corners=false); corner samples clamp to the input boundary.

When to use: forward 2-D bilinear resample. Pair with InterpolateBackwardPlan for autograd.

Dtypes: {f32, f64, f16, bf16}.

Shape limits: rank-4 NCHW input [N, C, IH, IW]; output [N, C, OH, OW]; all extents non-negative.

Modes: only Bilinear2d is wired in the trailblazer. Nearest2d / Bicubic2d / Trilinear3d / Linear1d / Area2d are reserved on the enum and return Unsupported.

Workspace: none.

Precision guarantee: deterministic, bit-stable on identical hardware. No atomics on FW.

Implementations

impl<T: Element> InterpolatePlan<T>

pub fn select( _stream: &Stream, desc: &InterpolateDescriptor, _pref: PlanPreference, ) -> Result<Self>

Pick a kernel for desc.

pub fn can_implement(&self, args: &InterpolateArgs<'_, T>) -> Result<()>

Validate args.

pub fn workspace_size(&self) -> usize

Workspace size (zero).

pub fn sku(&self) -> KernelSku

Identity of the kernel this plan picked.

pub fn precision_guarantee(&self) -> PrecisionGuarantee

Numerical guarantees for this plan’s kernel.

pub fn run( &self, stream: &Stream, _workspace: Workspace<'_>, args: InterpolateArgs<'_, T>, ) -> Result<()>

Launch.