Struct BinaryPlan 

pub struct BinaryPlan<T: Element, const N: usize> { /* private fields */ }

Binary elementwise plan.

T: Element is the kernel’s element type (today: must be f32). const N: usize is the tensor rank — fixed at compile time to keep the descriptor heap-free and the rank invariants type-checked.

Implementations

impl<T: Element, const N: usize> BinaryPlan<T, N>

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

Pick a kernel for desc. Returns Error::Unsupported if the (kind, T::KIND) pair isn’t wired today.

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

Validate that this plan can launch with args.

Accepts both fully-contiguous operands (which take the contig fast path) and broadcast / strided operands (which take the strided path). For each axis d, each input operand must satisfy shape[d] == y.shape[d] (no broadcast on that axis) or shape[d] == 1 && stride[d] == 0 (broadcast on that axis). The output must be exactly desc.shape and is conventionally contiguous, though the strided kernel accepts arbitrary y strides too.

pub fn workspace_size(&self) -> usize

Workspace size in bytes. Always 0 for the trailblazer SKU.

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: BinaryArgs<'_, T, N>, ) -> Result<()>

Launch.