Struct CastSubBytePlan 

pub struct CastSubBytePlan<TIn: DeviceRepr + Copy + 'static, TOut: DeviceRepr + Copy + 'static> { /* private fields */ }

Sub-byte cast plan.

TIn is the input element type, TOut is the output element type. Both are bounded by baracuda_types::DeviceRepr only (not baracuda_kernels_types::Element) so the dtype set can include S4, U4, Fp8E4M3, Fp8E5M2, and Bool alongside the classic fp / int element types.

Coverage: see the crate-level module docs for the full supported (TIn, TOut) pair list. A select-time check rejects any pair outside the explicit table with Error::Unsupported.

Workspace: none.

Implementations

impl<TIn: DeviceRepr + Copy + 'static, TOut: DeviceRepr + Copy + 'static> CastSubBytePlan<TIn, TOut>

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

Pick a kernel for desc.

pub fn can_implement(&self, args: &CastSubByteArgs<'_, TIn, TOut>) -> Result<()>

Validate args. Checks numel agreement and buffer sizing (the S4 / U4 packed-buffer accounting collapses to numel / 2 bytes, which is numel / 2 packed-slot elements at the buffer layer).

pub fn workspace_size(&self) -> usize

Workspace size in bytes. Always 0.

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: CastSubByteArgs<'_, TIn, TOut>, ) -> Result<()>

Launch.