Enum TensorWrite 

pub enum TensorWrite<'a> {
    Tensor(&'a mut Tensor),
    View(TensorViewMut<'a>),
}

Mutable tensor output accepted by synchronous eager kernels.

TensorWrite mirrors TensorRead for output dispatch: it can target an owned compact Tensor or a borrowed mutable TensorViewMut. The target is never resized.

Examples

use tenferro_tensor::{Tensor, TensorWrite};

let mut tensor = Tensor::from_vec_col_major(vec![1], vec![0.0_f64])?;
let write = TensorWrite::from_tensor(&mut tensor);
assert_eq!(write.shape(), &[1]);

Variants

Tensor(&'a mut Tensor)

View(TensorViewMut<'a>)

Implementations

impl<'a> TensorWrite<'a>

pub fn from_tensor(tensor: &'a mut Tensor) -> Self

pub fn from_view(view: TensorViewMut<'a>) -> Self

pub fn as_read(&self) -> TensorRead<'_>

Borrow this writable target as a read-only tensor input.

This is useful for explicit read-modify-write kernels such as accumulation updates. The returned view borrows through &self, so it cannot outlive the current read-only borrow of the writable target.

Examples

use tenferro_tensor::{DType, Tensor, TensorWrite};

let mut tensor = Tensor::from_vec_col_major(vec![1], vec![2.0_f64])?;
let write = TensorWrite::from_tensor(&mut tensor);
let read = write.as_read();
assert_eq!(read.dtype(), DType::F64);

pub fn dtype(&self) -> DType

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

pub fn strides(&self) -> Result<Vec<isize>>

pub fn offset(&self) -> isize

pub fn layout_linear_offset(&self, indices: &[usize]) -> Result<usize>

pub fn is_col_major_contiguous(&self) -> Result<bool>

pub fn layout_summary(&self) -> String

pub fn assert_col_major_contiguous(&self) -> Result<()>

pub fn copy_from_tensor(&mut self, src: &Tensor) -> Result<()>

Trait Implementations

impl<'a> Debug for TensorWrite<'a>

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

Formats the value using the given formatter.