Struct tension::HostTensor

pub struct HostTensor<T: Prm> { /* fields omitted */ }

Tensor structure. It consists of a contiguous one-dimensional array and a shape. Tensor tries to reuse resources as long as possible and implements copy-on-write mechanism.

Implementations

impl<T: Prm> HostTensor<T>

pub unsafe fn new_uninit(shape: &Shape) -> Self

Create unitialized tensor

pub fn new_filled(shape: &Shape, value: T) -> Self

Create tensor filled with value on the specified hardware

pub fn new_zeroed(shape: &Shape) -> Self

Create tensor filled with zeros on the specified hardware

pub fn iter<'a>(&'a self) -> HostTensorIter<'a, T>

Iterator over flatten tensor.

pub fn iter_mut<'a>(&'a mut self) -> HostTensorIterMut<'a, T>

Mutable iterator over flatten tensor.

Trait Implementations

impl<T: Prm> Tensor<T> for HostTensor<T>

type Buffer = HostBuffer<T>

Inner buffer type.

unsafe fn new_uninit_in(_: &(), shape: &Shape) -> Self

Create unitialized tensor

fn new_filled_in(_: &(), shape: &Shape, value: T) -> Self

Create tensor filled with value on the specified hardware

fn new_zeroed_in(_: &(), shape: &Shape) -> Self

Create tensor filled with zeros on the specified hardware

fn shape(&self) -> &Shape

Shape of the tensor - a slice containing all tensor dimensions.

fn reshape(&self, shape: &Shape) -> Self

Returns a new tensor that shares the same data but has other shape. Failed if the product of all shape dimensions is not equal to buffer size.

fn load(&self, dst: &mut [T])

Load flattened data from tensor to slice.

fn store(&mut self, src: &[T])

Store data from slice to a tensor in a flattened manner.