Struct RollPlan 

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

roll plan.

y = x.roll(shifts, axes) — cyclic shift along axes (PyTorch torch.roll). Output shape equals input shape.

When to use: forward cyclic shift. Pair with RollBackwardPlan, which reuses the FW kernel with negated shifts.

Dtypes: {f32, f64, f16, bf16}. Pure index permutation.

Shape limits: rank in [1, 8]; per-axis shifts can be negative (kernel normalizes to [0, shape[d])).

Workspace: none.

Precision guarantee: deterministic, bit-stable, bit-exact.

Implementations

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

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

Pick a kernel for desc.

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

Validate args.

pub fn workspace_size(&self) -> usize

Workspace size in bytes.

pub fn sku(&self) -> KernelSku

Identity of the kernel this plan picked.

pub fn precision_guarantee(&self) -> PrecisionGuarantee

Numerical guarantees.

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

Launch.