Struct DequantizePerTensorPlan 

pub struct DequantizePerTensorPlan<TIn: Element, TOut: IntElement> { /* private fields */ }

dequantize_per_tensor plan.

x = scale * (q - zero_point). Linear; exactly invertible (up to FW rounding) against QuantizePerTensorPlan.

When to use: FP recovery from a per-tensor-quantized buffer. Pair with DequantizePerTensorBackwardPlan for autograd through the dequant op.

Dtypes: input int {s8, u8} (= TOut); output FP {f32, f64, f16, bf16} (= TIn).

Shape limits: flat [numel].

Workspace: none.

Precision guarantee: deterministic, bit-stable.

Implementations

impl<TIn: Element, TOut: IntElement> DequantizePerTensorPlan<TIn, TOut>

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

Pick a kernel.

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

Validate args.

pub fn workspace_size(&self) -> usize

Workspace bytes.

pub fn sku(&self) -> KernelSku

Identity.

pub fn precision_guarantee(&self) -> PrecisionGuarantee

Numerical guarantees.

pub fn run( &self, stream: &Stream, _workspace: Workspace<'_>, args: DequantizePerTensorArgs<'_, TIn, TOut>, ) -> Result<()>

Launch.