Struct SimpleConvolution

pub struct SimpleConvolution<MP: MatmulPrecision, SMM: StageMatmul<MP>> { /* private fields */ }

Performs matrix multiplication at the global level, with each plane sharing the same responsibilities

• All planes load data to the stage
• All planes are used in the stage matmul computation

Trait Implementations

impl<MP: MatmulPrecision, SMM> Convolution<MP> for SimpleConvolution<MP, SMM>

where SMM: StageMatmul<MP, LhsReader = FullReader<MP::ES, ConvTilingLayout>, RhsReader = FullReader<MP::ES, ConvTilingLayout>>,

type LhsLoader = SimpleIm2colLoader<MP, <SimpleConvolution<MP, SMM> as Convolution<MP>>::Config>

type Config = HomogeneousConfig<Config<<SMM as StageMatmul<MP>>::Config>>

type RhsLoader = SyncFullLoader<MP, <SMM as StageMatmul<MP>>::Config, LoadingStrategy<RowMajorTilingOrder>>

type AccumulatorLoader = BiasLoader<MP>

type Out = Unloader<<MP as MatmulPrecision>::EO>

type Accumulator = <SMM as StageMatmul<MP>>::Accumulator

fn execute( lhs_loader: Self::LhsLoader, rhs_loader: Self::RhsLoader, acc_loader: Self::AccumulatorLoader, out_unloader: Self::Out, acc: &mut Self::Accumulator, k_range: (u32, u32), config: Self::Config, )

Performs the convolution over data loaded by the LHS and RHS loaders, over the range given for K, and stores with using the output unloader.

fn init_lhs_loader( lhs: VirtualTensor<MP::EI>, x_offset: u32, y_offset: u32, runtime_args: &RuntimeArgs, config: Self::Config, ) -> Self::LhsLoader

fn init_rhs_loader( rhs: VirtualTensor<MP::EI>, x_offset: u32, y_offset: u32, _runtime_args: &RuntimeArgs, config: Self::Config, ) -> Self::RhsLoader

fn init_bias_loader( bias: CubeOption<VirtualTensor<MP::EO>>, n_offset: u32, config: Self::Config, ) -> Self::AccumulatorLoader

fn init_unloader( out: VirtualTensor<MP::EO, ReadWrite>, x_offset: u32, y_offset: u32, ) -> Self::Out

fn init_accumulator(config: Self::Config) -> Self::Accumulator

fn __expand_execute( context: &mut Scope, lhs_loader: <Self::LhsLoader as CubeType>::ExpandType, rhs_loader: <Self::RhsLoader as CubeType>::ExpandType, acc_loader: <Self::AccumulatorLoader as CubeType>::ExpandType, out_unloader: <Self::Out as CubeType>::ExpandType, acc: <Self::Accumulator as CubeType>::ExpandType, k_range: <(u32, u32) as CubeType>::ExpandType, config: Self::Config, ) -> <() as CubeType>::ExpandType

fn __expand_init_lhs_loader( context: &mut Scope, lhs: <VirtualTensor<MP::EI> as CubeType>::ExpandType, x_offset: <u32 as CubeType>::ExpandType, y_offset: <u32 as CubeType>::ExpandType, runtime_args: <RuntimeArgs as CubeType>::ExpandType, config: Self::Config, ) -> <Self::LhsLoader as CubeType>::ExpandType

fn __expand_init_rhs_loader( context: &mut Scope, rhs: <VirtualTensor<MP::EI> as CubeType>::ExpandType, x_offset: <u32 as CubeType>::ExpandType, y_offset: <u32 as CubeType>::ExpandType, _runtime_args: <RuntimeArgs as CubeType>::ExpandType, config: Self::Config, ) -> <Self::RhsLoader as CubeType>::ExpandType

fn __expand_init_bias_loader( context: &mut Scope, bias: <CubeOption<VirtualTensor<MP::EO>> as CubeType>::ExpandType, n_offset: <u32 as CubeType>::ExpandType, config: Self::Config, ) -> <Self::AccumulatorLoader as CubeType>::ExpandType

fn __expand_init_unloader( context: &mut Scope, out: <VirtualTensor<MP::EO, ReadWrite> as CubeType>::ExpandType, x_offset: <u32 as CubeType>::ExpandType, y_offset: <u32 as CubeType>::ExpandType, ) -> <Self::Out as CubeType>::ExpandType

fn __expand_init_accumulator( context: &mut Scope, config: Self::Config, ) -> <Self::Accumulator as CubeType>::ExpandType