Enum coaster_nn::ConvBackwardDataAlgo

pub enum ConvBackwardDataAlgo {
    Auto,
    ImplicitGEMM,
    ImplicitGEMMSum,
    FFT,
    FFTTiling,
    Winograd,
    WinogradNonFused,
}

Different algorithms to compute the gradient with respect to the filter.

Variants

Auto

Attempt to automatically find the best algorithm of all the other available ones.

ImplicitGEMM

Compute the convolution as matrix product without forming the matrix that holds the input data.

Does not need any memory workspace.

The results are deterministic.

ImplicitGEMMSum

Compute the convolution as sum of matrix product without forming the matrix that holds the input data.

Does not need any memory workspace.

The results are non-deterministic.

FFT

Compute the convolution as Fast-Fourier Transform.

Needs a significant memory workspace.

The results are deterministic.

FFTTiling

Compute the convolution as Fast-Fourier Transform with 32x32 tiles.

Needs a significant memory workspace.

The results are deterministic.

Winograd

Winograd Transform

WinogradNonFused

Winograd Transform Non-Fused

Implementations

impl ConvBackwardDataAlgo

pub fn is_auto(&self) -> bool

Check if algorithim should be chosen automatically.

Trait Implementations

impl Clone for ConvBackwardDataAlgo

fn clone(&self) -> ConvBackwardDataAlgo

Returns a copy of the value.

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

Performs copy-assignment from source.

impl Debug for ConvBackwardDataAlgo

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

Formats the value using the given formatter.

impl PartialEq for ConvBackwardDataAlgo

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

This method tests for self and other values to be equal, and is used by ==.

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

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

impl Copy for ConvBackwardDataAlgo

impl Eq for ConvBackwardDataAlgo

impl StructuralPartialEq for ConvBackwardDataAlgo