Struct TensorArena 

pub struct TensorArena { /* private fields */ }

Tensor Arena allocator using bumpalo

Provides shape-aware memory reuse with bump allocation strategy. Memory is allocated from the arena and can be reused for tensors with the same shape, avoiding repeated allocations.

Implementations

impl TensorArena

pub fn new() -> Self

Create a new tensor arena with default capacity (16 MB)

pub fn with_capacity(capacity: usize) -> Self

Create a new tensor arena with specified capacity

pub fn allocate(&mut self, shape: &[usize]) -> Result<ArenaTensor, TensorError>

Allocate a tensor with the given shape

Tries to reuse memory from the free list if a matching shape exists. Otherwise allocates new memory from the bump arena.

pub fn deallocate(&mut self, tensor: ArenaTensor)

Deallocate a tensor and return its memory to the free list

pub fn reset(&mut self)

Reset the arena, clearing all free lists

This releases all memory back to the system.

pub fn stats(&self) -> &ArenaStats

Get statistics

pub fn capacity(&self) -> usize

Get the current capacity in bytes

pub fn bytes_in_use(&self) -> usize

Get the current bytes in use

pub fn allocate_fresh( &mut self, shape: &[usize], ) -> Result<ArenaTensor, TensorError>

Force allocate without reuse (for benchmarking)

Trait Implementations

impl Debug for TensorArena

Available on crate feature tensor-pool only.

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

Formats the value using the given formatter.

impl Default for TensorArena

Available on crate feature tensor-pool only.

fn default() -> Self

Returns the “default value” for a type.