Struct TensorWriter

pub struct TensorWriter<EO: Numeric> {
    pub tensor: VirtualTensor<EO, ReadWrite>,
    pub x_offset: u32,
    pub y_offset: u32,
    pub stride_x: u32,
    pub stride_y: u32,
    pub shape_x: u32,
    pub shape_y: u32,
    pub batch_offset: u32,
}

A view of a tensor that starts reading data from a specified offset. Ensures safe access by preventing out-of-bounds errors. Includes pre-fetched shapes and strides for optimized performance.

Fields

tensor: VirtualTensor<EO, ReadWrite>

x_offset: u32

y_offset: u32

stride_x: u32

stride_y: u32

shape_x: u32

shape_y: u32

batch_offset: u32

Implementations

impl<EG: Numeric> TensorWriter<EG>

pub fn new( tensor: VirtualTensor<EG, ReadWrite>, x_offset: u32, y_offset: u32, batch_offset: u32, ) -> Self

Instantiate a write view over the given tensor, pre-fetching needed strides and shapes

pub fn write_coalesced<ES: Numeric, G: GlobalConfig>( &mut self, tile_x: u32, tile_y: u32, unit_id: u32, value: Line<ES>, config: G, )

Writes data into the tensor view at the specified coordinates (tile_x, tile_y).

Each unit writes one line in a coalesced manner for improved efficiency, assuming row-major layout.

pub fn __expand_new( context: &mut Scope, tensor: <VirtualTensor<EG, ReadWrite> as CubeType>::ExpandType, x_offset: <u32 as CubeType>::ExpandType, y_offset: <u32 as CubeType>::ExpandType, batch_offset: <u32 as CubeType>::ExpandType, ) -> <Self as CubeType>::ExpandType

pub fn __expand_write_coalesced<ES: Numeric, G: GlobalConfig>( context: &mut Scope, this: <Self as CubeType>::ExpandType, tile_x: <u32 as CubeType>::ExpandType, tile_y: <u32 as CubeType>::ExpandType, unit_id: <u32 as CubeType>::ExpandType, value: <Line<ES> as CubeType>::ExpandType, config: G, ) -> <() as CubeType>::ExpandType

Trait Implementations

impl<EO: Numeric> CubeType for TensorWriter<EO>

type ExpandType = TensorWriterExpand<EO>

fn init(scope: &mut Scope, expand: Self::ExpandType) -> Self::ExpandType

Wrapper around the init method, necessary to type inference.

impl<EG: Numeric> Send for TensorWriter<EG>

impl<EG: Numeric> Sync for TensorWriter<EG>