Struct CpuTensor 

pub struct CpuTensor(/* private fields */);

f32-dense CPU-resident tensor handle. Arc-shared so intra-Node clones (FedAvg’s per-peer buffer insert, slot-table writes, etc.) are refcount bumps rather than Vec<f32> deep copies. The underlying CpuBackendBuffer is owned by the backend, which is free to pool / reuse / free the storage at a later milestone without API churn (the handle shape stays identical).

Implementations

impl CpuTensor

pub fn from_array(data: ArrayD<f32>) -> Self

Wrap an existing ArrayD<f32> in a fresh backend buffer. charged_bytes = 0 — kernel outputs and test fixtures don’t arrive via the wire and therefore don’t hold an ingress charge. The wire path uses the crate-private from_wire_buffer helper from CpuBackend::materialize_from_wire.

pub fn from_vec(shape: Vec<i64>, data: Vec<f32>) -> Self

Construct from ONNX-signed shape + flat row-major data via ndarray::ArrayD::from_shape_vec. Equivalent to Self::new but spelled out for callers preferring the builder-style name.

pub fn as_array(&self) -> &ArrayD<f32>

Borrow the underlying ndarray.

pub fn into_array(self) -> ArrayD<f32>

Clone the backend buffer into an owned ArrayD<f32>. The Arc shape means the buffer cannot be unwrapped in-place (other handles may share it), so this always pays the ndarray deep copy. Test-only callers needing flat data should prefer Self::flat_data.

pub fn new(dims: Vec<i64>, data: Vec<f32>) -> Self

Construct from ONNX-signed shape + flat row-major data. Panics if the dims product doesn’t match the data length — callers needing strict checking use Self::new_checked.

pub fn new_checked( dims: Vec<i64>, data: Vec<f32>, ) -> Result<Self, CpuTensorError>

Construct + validate dims_product(&dims) == data.len().

pub fn zeros(dims: Vec<i64>) -> Self

Construct a zero-filled tensor with the given shape.

pub fn ones(dims: Vec<i64>) -> Self

Construct a ones-filled tensor with the given shape.

pub fn strong_count(&self) -> usize

Observe the underlying Arc<CpuBackendBuffer> strong-refcount. One strong holder means the caller holds the only handle to the buffer; future pooling implementations (v2) read this to decide whether to return the buffer to the pool on drop. Tests use this to prove the wire-decode path lands a single carrier in the slot table with no spurious clones.

Trait Implementations

impl Clone for CpuTensor

fn clone(&self) -> CpuTensor

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for CpuTensor

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

Formats the value using the given formatter.

impl<'de> Deserialize<'de> for CpuTensor

fn deserialize<D>(deserializer: D) -> Result<Self, D::Error>

where D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Display for CpuTensor

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

Formats the value using the given formatter.

impl Serialize for CpuTensor

fn serialize<S>(&self, serializer: S) -> Result<S::Ok, S::Error>

where S: Serializer,

Serialize this value into the given Serde serializer.

impl Storage for CpuTensor

CpuTensor → TYPE_TENSOR_F32 in the polymorphism tree.

const TYPE: &'static TypeNode

Position-in-tree declaration. The TypeNode static this constant points at decides what other storage types unify with this one during the type-solver walk.

impl Tensor for CpuTensor

type Scalar = f32

The scalar element type this tensor holds.

fn dims(&self) -> &[i64]

Tensor shape. ONNX-compatible signed-dim convention; -1 for dynamic dims.

fn len(&self) -> usize

Total element count across all dims. For dynamic-dim tensors callers must resolve concrete dims before consulting.

fn to_proto(&self) -> TensorProto

Serialize to canonical ONNX TensorProto. The result is portable across backends declaring the same scalar type.

fn from_proto(proto: TensorProto) -> Result<Self, TensorSerializationError>

Deserialize from canonical ONNX TensorProto. Returns an error if the proto’s elem_type / shape doesn’t match Self’s expectations.

fn is_empty(&self) -> bool

true when the tensor holds zero elements.