Struct Storage 

pub struct Storage<T: Scalar> { /* private fields */ }

Raw memory storage for tensor data.

Storage is the fundamental building block for tensors. It manages a contiguous block of memory on a specific device and is reference-counted to allow efficient sharing between tensor views.

Implementations

impl<T: Scalar> Storage<T>

pub fn zeros(len: usize, device: Device) -> Self

Creates new storage with the given capacity, initialized to zero.

pub fn from_vec(data: Vec<T>, device: Device) -> Self

Creates storage from an existing vector (always on CPU).

pub fn from_slice(data: &[T], device: Device) -> Self

Creates storage from a slice by copying the data.

pub const fn len(&self) -> usize

Returns the number of elements in this storage view.

pub const fn is_empty(&self) -> bool

Returns true if the storage is empty.

pub const fn offset(&self) -> usize

Returns the offset into the underlying buffer.

pub fn device(&self) -> Device

Returns the device this storage is on.

pub fn is_cpu(&self) -> bool

Returns true if data is on CPU.

pub fn is_gpu(&self) -> bool

Returns true if data is on GPU.

pub fn size_bytes(&self) -> usize

Returns the size in bytes of this storage.

pub fn slice(&self, offset: usize, len: usize) -> Result<Self>

Creates a view into a portion of this storage.

pub fn is_unique(&self) -> bool

Returns true if this storage is uniquely owned (not shared).

pub fn as_slice(&self) -> StorageReadGuard<'_, T>

Returns an immutable reference to the CPU data.

Panics

Panics if the storage is on GPU. Use to_vec() for device-safe access.

pub fn as_slice_mut(&self) -> StorageWriteGuard<'_, T>

Returns a mutable reference to the CPU data.

Panics

Panics if the storage is on GPU.

pub fn copy_from(&self, other: &Self) -> Result<()>

Copies data from another storage into this one.

pub fn deep_copy(&self) -> Self

Makes a deep copy of this storage.

For GPU storage, this only works for Storage<f32>. Other types will panic on GPU storage.

pub fn to_vec(&self) -> Vec<T>

Returns the data as a Vec from CPU storage.

For GPU-resident f32 tensors, use the Tensor::to_vec() method which handles device-aware copies via Storage<f32>::to_vec_f32().

Panics

Panics if storage is on GPU (use to_vec_f32() on Storage<f32> instead).

pub fn to_device(&self, device: Device) -> Result<Self>

Transfers this storage to a different device.

For GPU transfers, only Storage<f32> is supported. Use the Storage<f32>::to_device() specialization for CPU↔GPU transfers.

Trait Implementations

impl<T: Scalar> Clone for Storage<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 + Scalar> Debug for Storage<T>

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

Formats the value using the given formatter.