Struct TensorView 

pub struct TensorView<T> { /* private fields */ }

A view into a contiguous memory region with shape and stride information.

TensorView does not own the data - it provides a structured view over existing memory, enabling zero-copy operations like slicing and transposition.

Type Parameters

• T - Element type (typically f32 for GPU compute)

cuda-tile-behavior.md References

• Falsification test #31: TensorView preserves data integrity
• Falsification test #32: Slicing produces correct views
• Falsification test #33: Transpose swaps dimensions correctly

Implementations

impl<T> TensorView<T>

pub fn new(shape: [usize; 4]) -> Self

Create a new TensorView with the given shape.

Strides are computed automatically based on row-major layout.

Arguments

• shape - Shape of the tensor (unused dimensions should be 1)

Examples

let view = TensorView::<f32>::new([2, 3, 4, 1]); // 2x3x4 tensor
assert_eq!(view.numel(), 24);

pub fn with_strides(shape: [usize; 4], strides: [usize; 4]) -> Self

Create a TensorView with explicit strides.

Arguments

• shape - Shape of the tensor
• strides - Strides for each dimension (in elements)

pub fn new_1d(len: usize) -> Self

Create a 1D TensorView.

pub fn new_2d(rows: usize, cols: usize) -> Self

Create a 2D TensorView (matrix).

pub fn new_3d(d0: usize, d1: usize, d2: usize) -> Self

Create a 3D TensorView.

pub fn new_4d(d0: usize, d1: usize, d2: usize, d3: usize) -> Self

Create a 4D TensorView.

pub fn shape(&self) -> &[usize; 4]

Get the shape of the tensor.

pub fn strides(&self) -> &[usize; 4]

Get the strides of the tensor.

pub fn offset(&self) -> usize

Get the offset from the start of the buffer.

pub fn layout(&self) -> MemoryLayout

Get the memory layout.

pub fn ndim(&self) -> usize

Get the number of active dimensions.

pub fn numel(&self) -> usize

Get the total number of elements.

pub fn is_contiguous(&self) -> bool

Check if the tensor is contiguous in memory.

A tensor is contiguous if elements are stored without gaps in row-major order.

pub fn is_empty(&self) -> bool

Check if the tensor is empty (has zero elements).

pub fn dim(&self, dim: usize) -> usize

Get dimension size at the given index.

Panics

Panics if dim >= 4.

pub fn stride(&self, dim: usize) -> usize

Get stride at the given dimension.

Panics

Panics if dim >= 4.

pub fn slice(&self, range: Range<usize>) -> Self

Create a slice of this tensor along the first dimension.

Arguments

• range - Range of indices to include

Returns

A new TensorView representing the slice.

cuda-tile-behavior.md References

• Falsification test #32: Slicing produces correct views

pub fn slice_dim(&self, dim: usize, range: Range<usize>) -> Self

Create a slice along a specific dimension.

Arguments

• dim - Dimension to slice along
• range - Range of indices to include

pub fn transpose(&self, dim0: usize, dim1: usize) -> Self

Transpose the tensor by swapping two dimensions.

Arguments

• dim0 - First dimension to swap
• dim1 - Second dimension to swap

Returns

A new TensorView with swapped dimensions.

cuda-tile-behavior.md References

• Falsification test #33: Transpose swaps dimensions correctly

pub fn reshape(&self, new_shape: [usize; 4]) -> Option<Self>

Reshape the tensor to a new shape.

Arguments

• new_shape - New shape (must have same number of elements)

Returns

A new TensorView with the new shape, or None if reshape is invalid.

pub fn squeeze(&self) -> Self

Squeeze dimensions of size 1.

Returns a view with all size-1 dimensions removed.

pub fn unsqueeze(&self, dim: usize) -> Option<Self>

Unsqueeze: add a dimension of size 1 at the specified position.

Arguments

• dim - Position to insert the new dimension

pub fn with_layout(self, layout: MemoryLayout) -> Self

Set the memory layout hint.

pub fn linear_index(&self, indices: [usize; 4]) -> usize

Compute linear index from multi-dimensional indices.

Arguments

• indices - Array of indices for each dimension

Returns

Linear offset into the underlying buffer.

Trait Implementations

impl<T> Clone for TensorView<T>

fn clone(&self) -> Self

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl<T: Debug> Debug for TensorView<T>

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

Formats the value using the given formatter.

impl<T> Default for TensorView<T>

fn default() -> Self

Returns the “default value” for a type.