Struct SpecializedConv 

pub struct SpecializedConv<TMM: TileMatmulFamily, L: AsyncPartialLoadingStrategy<RuntimeArgs>> { /* private fields */ }

Cmma convolution

Trait Implementations

impl<TMM: TileMatmulFamily<LhsTile = Strided, RhsTile = Strided, AccTile = CubeOption<Strided>, OutTile = Strided>, L: AsyncPartialLoadingStrategy<RuntimeArgs, TileKind = Strided>> Algorithm for SpecializedConv<TMM, L>

const IS_SPECIALIZED: bool = true

Whether to select specialized load flow in tests. Should replace with something cleaner eventually, but this is nice and simple.

type Routine = SpecializedAlgorithm<TMM, L, SyncBiasLoading>

type Args = TensorArgs<RuntimeArgs>

fn into_tensor_handle<R: Runtime>( client: &ComputeClient<R>, handle: &TensorHandleRef<'_, R>, dtype: StorageType, _operation: ConvolutionOperation, ) -> Result<TensorHandle<R>, LaunchError>

fn filter_line_sizes(line_sizes: AvailableLineSizes) -> AvailableLineSizes