Struct TensorMeta 

pub struct TensorMeta {
    pub dtype: DType,
    pub extents: Vec<ShapeExtent<SymDim>>,
}

Per-value tensor metadata used by AD rules.

Shape information is stored as per-axis ShapeExtent values. Callers must explicitly choose whether they need an exact shape or only a known bound.

Examples

use tenferro_ops::{SymDim, TensorMeta};
use tenferro_tensor::DType;

let meta = TensorMeta::exact(DType::F64, vec![SymDim::from(2usize), SymDim::from(3usize)]);
assert_eq!(meta.rank(), 2);

Fields

dtype: DType

Element dtype of the tensor value.

extents: Vec<ShapeExtent<SymDim>>

Per-axis shape guarantees.

Implementations

impl TensorMeta

pub fn exact(dtype: DType, shape: Vec<SymDim>) -> Self

Construct metadata whose every axis is exact.

Examples

use tenferro_ops::{SymDim, TensorMeta};
use tenferro_tensor::DType;

let meta = TensorMeta::exact(DType::F64, vec![SymDim::from(4usize)]);
assert_eq!(meta.exact_shape(), Some(vec![SymDim::from(4usize)]));

pub fn with_extents(dtype: DType, extents: Vec<ShapeExtent<SymDim>>) -> Self

Construct metadata from per-axis extents.

Examples

use tenferro_ops::{ShapeExtent, SymDim, TensorMeta};
use tenferro_tensor::DType;

let meta = TensorMeta::with_extents(
    DType::F64,
    vec![ShapeExtent::upper_bound(SymDim::from(8usize))],
);
assert_eq!(meta.exact_shape(), None);

pub fn rank(&self) -> usize

Return the tensor rank known by this metadata record.

pub fn extents(&self) -> &[ShapeExtent<SymDim>]

Return the per-axis shape guarantees.

Examples

use tenferro_ops::{SymDim, TensorMeta};
use tenferro_tensor::DType;

let meta = TensorMeta::exact(DType::F64, vec![SymDim::from(4usize)]);
assert_eq!(meta.extents().len(), 1);

pub fn exact_shape(&self) -> Option<Vec<SymDim>>

Return the shape only when every axis is exact.

Examples

use tenferro_ops::{ShapeExtent, SymDim, TensorMeta};
use tenferro_tensor::DType;

let meta = TensorMeta::with_extents(
    DType::F64,
    vec![ShapeExtent::upper_bound(SymDim::from(8usize))],
);
assert_eq!(meta.exact_shape(), None);

pub fn bound_shape(&self) -> Option<Vec<SymDim>>

Return one known bound per axis when every axis has a bound.

This is intentionally separate from TensorMeta::exact_shape: a bound is not proof of the runtime size.

Trait Implementations

impl Clone for TensorMeta

fn clone(&self) -> TensorMeta

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for TensorMeta

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

Formats the value using the given formatter.

impl PartialEq for TensorMeta

fn eq(&self, other: &TensorMeta) -> bool

Tests for self and other values to be equal, and is used by ==.

fn ne(&self, other: &Rhs) -> bool

Tests for !=. The default implementation is almost always sufficient, and should not be overridden without very good reason.

impl Eq for TensorMeta

impl StructuralPartialEq for TensorMeta