Struct TuContext 

pub struct TuContext<const T: Tu> { /* private fields */ }

Context for a specific tensor unit.

Implementations

impl<const T: Tu> TuContext<T>

pub fn begin<'l, D: Scalar, Chip: M, Cluster: M, Slice: M, Element: M>( &'l mut self, tensor: DmTensorView<'l, D, Chip, Cluster, Slice, Element>, ) -> BeginTensor<'l, T, D, Chip, Cluster, Slice, Identity, Element>

Begin a tensor unit operation in this context.

pub fn begin_interleaved<'l, I: AxisName, D: Scalar, Chip: M, Cluster: M, Slice: M, Element: M>( &'l mut self, lhs: DmTensorView<'l, D, Chip, Cluster, Slice, Element>, rhs: DmTensorView<'l, D, Chip, Cluster, Slice, Element>, ) -> BeginTensor<'l, T, D, Chip, Cluster, Slice, Symbol<I>, Element>

Begin a tensor unit operation in this context with interleaved tensors.

impl TuContext<{ Tu::Sub }>

pub fn parallel_copy_cluster_slice<'l, const CLUSTER_DIM: usize, AxisToSlice: M, AxisSlicedElement: M, Element2: M, D: Scalar, Chip: M, Cluster: M, Slice: M, Element: M>( &mut self, tensor: DmTensorView<'l, D, Chip, Cluster, Slice, Element>, slice_indices: &[usize; CLUSTER_DIM], ) -> DmTensor<D, Chip, Cluster, Slice, Element2>

Perform asymmetric cluster slice operation using ParallelCopy (stos command). This operation allows different clusters to select different slice positions.

Arguments

• tensor - Input tensor with cluster dimension
• slice_indices - Array of slice indices, one per cluster

Example

For Cluster=2 with slice_indices=[1,0]:

• Cluster 0 selects slice position 1
• Cluster 1 selects slice position 0

Restrictions

The AxisToSlice should be the outermost axis in Element.

pub fn parallel_copy_chip_slice<'l, const CHIP_DIM: usize, AxisToSlice: M, AxisSlicedElement: M, Element2: M, D: Scalar, Chip: M, Cluster: M, Slice: M, Element: M>( &mut self, tensor: DmTensorView<'l, D, Chip, Cluster, Slice, Element>, slice_indices: &[usize; CHIP_DIM], ) -> DmTensor<D, Chip, Cluster, Slice, Element2>

Perform asymmetric chip slice operation using ParallelCopy (stos command). This operation allows different chips to select different slice positions.

Arguments

• tensor - Input tensor with chip dimension
• slice_indices - Array of slice indices, one per chip

Example

For Chip=4 with slice_indices=[3,0,1,2]:

• Chip 0 selects slice position 3
• Chip 1 selects slice position 0
• Chip 2 selects slice position 1
• Chip 3 selects slice position 2

Trait Implementations

impl<const T: Tu> Debug for TuContext<T>

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

Formats the value using the given formatter.