Struct KernelSku 

pub struct KernelSku {
    pub category: OpCategory,
    pub op: u16,
    pub element: ElementKind,
    pub aux_element: Option<ElementKind>,
    pub layout: Option<LayoutSku>,
    pub epilogue: Option<EpilogueKind>,
    pub arch: ArchSku,
    pub backend: BackendKind,
    pub precision_guarantee: PrecisionGuarantee,
}

Generalized kernel SKU — covers every op category.

Replaces / generalizes the GEMM-specific GemmSku used by Phase 1/2 plan types. Phase 3+ plan types (UnaryPlan, BinaryPlan, TernaryPlan, …) populate and return this struct from their sku() accessor.

The op field is a category-local discriminant — its interpretation depends on category. For OpCategory::BinaryElementwise it’s a BinaryKind as u16; for OpCategory::UnaryElementwise it’s a UnaryKind as u16; etc. Surfacing it as a flat u16 (rather than a per-category enum) keeps the struct shape stable across categories so it can be hashed into autotuner caches uniformly.

Fields

category: OpCategory

Op category. Primary axis for telemetry / selector dispatch.

op: u16

Category-local op discriminant. Interpret against the per-category *Kind enum that the category’s Plan type uses.

element: ElementKind

Primary element type the kernel operates on.

aux_element: Option<ElementKind>

Auxiliary element type, when meaningful (bias element for GEMM bias epilogues, index element for gather / scatter, output cast type for ops that produce a different dtype than the input). None when the op has no auxiliary element.

layout: Option<LayoutSku>

Layout discriminant for matrix-multiply-shaped kernels. None for op categories that don’t have a row/col layout dimension (elementwise, reduce, scan, …).

epilogue: Option<EpilogueKind>

Epilogue discriminant for matrix-multiply-shaped kernels. None for op categories that don’t fuse a GEMM epilogue chain.

arch: ArchSku

Compute capability the selected kernel was compiled for.

backend: BackendKind

Which underlying compute path served this SKU.

precision_guarantee: PrecisionGuarantee

Numerical guarantees this SKU provides.

Trait Implementations

impl Clone for KernelSku

fn clone(&self) -> KernelSku

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Copy for KernelSku

impl Debug for KernelSku

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

Formats the value using the given formatter.

impl Eq for KernelSku

impl Hash for KernelSku

fn hash<__H>(&self, state: &mut __H)

where __H: Hasher,

Feeds this value into the given Hasher.

fn hash_slice<H>(data: &[Self], state: &mut H)

where H: Hasher, Self: Sized,

Feeds a slice of this type into the given Hasher.

impl PartialEq for KernelSku

fn eq(&self, other: &KernelSku) -> bool

Tests for self and other values to be equal, and is used by ==.

fn ne(&self, other: &Rhs) -> bool

Tests for !=. The default implementation is almost always sufficient, and should not be overridden without very good reason.

impl StructuralPartialEq for KernelSku