skainet-lang-core/sk.ainet.lang.tensor.storage/TensorStorage

TensorStorage

data class TensorStorage(val shape: Shape, val logicalType: LogicalDType, val encoding: TensorEncoding, val buffer: BufferHandle, val placement: Placement = Placement.CPU_HEAP, val byteOffset: Long = 0, val strides: LongArray? = null, val isContiguous: Boolean = true)(source)

Runtime descriptor for a tensor's backing memory.

TensorStorage is the main architectural type that replaces ad-hoc array passing between loaders, planners, and backends. It carries enough information to handle a tensor without inspecting its bytes:

What the values mean: logicalType
How they are stored: encoding
Where the bytes live: buffer + placement
Layout: shape, byteOffset, strides, isContiguous
Ownership: via buffer's BufferHandle subtype

Existing sk.ainet.lang.tensor.data.TensorData remains as a compatibility façade. New loaders, planners, and backends should target TensorStorage directly.

Constructors

constructor(shape: Shape, logicalType: LogicalDType, encoding: TensorEncoding, buffer: BufferHandle, placement: Placement = Placement.CPU_HEAP, byteOffset: Long = 0, strides: LongArray? = null, isContiguous: Boolean = true)

Properties

val buffer: BufferHandle

val byteOffset: Long

val elementCount: Long

Number of logical elements in this tensor.

val encoding: TensorEncoding

val isAlias: Boolean

Whether this storage is an alias (view) into another buffer.

val isContiguous: Boolean

val isFileBacked: Boolean

Whether this storage is backed by a memory-mapped file.

val isMutable: Boolean

Whether this storage is mutable.

val logicalBytes: Long

Logical size: number of elements x logical element size.

val logicalType: LogicalDType

val ownership: Ownership

Ownership mode of the backing buffer.

val physicalBytes: Long

Physical size: actual bytes consumed in the buffer for this tensor.

val placement: Placement

val shape: Shape

val strides: LongArray?

Functions

copyMaterialize

fun copyMaterialize(): TensorStorage

Create a new TensorStorage with an owned copy of this storage's data. The returned storage is independent of the original buffer.

fun copyToDevice(device: DeviceKind): TensorStorage

Request a copy of this storage on the specified device. Currently only CPU is supported — GPU/NPU backends will override.

fun copyToHost(): TensorStorage

Ensure this storage resides on the host (CPU heap). If already on host, returns this. Otherwise copies to host.

open operator override fun equals(other: Any?): Boolean

open override fun hashCode(): Int

fun memoryReport(): StorageMemoryReport

Memory report for this single tensor, useful for diagnostics and regression testing.

fun repackTo(targetEncoding: TensorEncoding): TensorStorage

Re-encode this storage into a different physical encoding. Currently a stub — actual transcoding requires backend kernels.