irx.builtins.collections.tensor

`irx.builtins.collections.tensor`

Source: packages/irx/src/irx/builtins/collections/tensor.py

title: Tensor layout helpers layered on the builtin tensor runtime.
summary: >-
  Define IRx's backend-neutral Tensor metadata helpers on top of the canonical
  buffer/view substrate and the Arrow C++ backed tensor runtime.

Functions

`tensor_element_count(layout: TensorLayout) -> int`

title: Return the logical element count for one layout.
parameters:
  layout:
    type: TensorLayout
returns:
  type: int

`tensor_default_strides(shape: tuple[int, ...], item_size_bytes: int, *, order: TensorOrder=TensorOrder.C) -> tuple[int, ...]`

title: Return canonical byte strides for one contiguous Tensor shape.
parameters:
  shape:
    type: tuple[int, Ellipsis]
  item_size_bytes:
    type: int
  order:
    type: TensorOrder
returns:
  type: tuple[int, Ellipsis]

`tensor_is_c_contiguous(layout: TensorLayout, item_size_bytes: int) -> bool`

title: Return whether one layout matches canonical C-order strides.
parameters:
  layout:
    type: TensorLayout
  item_size_bytes:
    type: int
returns:
  type: bool

`tensor_is_f_contiguous(layout: TensorLayout, item_size_bytes: int) -> bool`

title: Return whether one layout matches canonical Fortran-order strides.
parameters:
  layout:
    type: TensorLayout
  item_size_bytes:
    type: int
returns:
  type: bool

`validate_tensor_layout(layout: TensorLayout) -> tuple[str, ...]`

title: Validate one static Tensor layout.
parameters:
  layout:
    type: TensorLayout
returns:
  type: tuple[str, Ellipsis]

`tensor_byte_bounds(layout: TensorLayout) -> tuple[int, int] | None`

title: Return the minimum and maximum element-start byte offsets.
summary: >-
  The result is relative to the underlying data pointer. None means the
  logical layout has zero extent and therefore addresses no elements.
parameters:
  layout:
    type: TensorLayout
returns:
  type: tuple[int, int] | None

`tensor_primitive_type_name(type_: astx.DataType | None) -> str | None`

title: Return the builtin primitive storage name for one Tensor element.
parameters:
  type_:
    type: astx.DataType | None
returns:
  type: str | None

`tensor_element_size_bytes(type_: astx.DataType | None) -> int | None`

title: Return the byte width for one Tensor element type.
parameters:
  type_:
    type: astx.DataType | None
returns:
  type: int | None

`tensor_buffer_dtype(type_: astx.DataType | None) -> BufferHandle | None`

title: Return the canonical buffer dtype handle for one Tensor element.
parameters:
  type_:
    type: astx.DataType | None
returns:
  type: BufferHandle | None

`tensor_buffer_view_metadata(*, data: BufferHandle, owner: BufferHandle, dtype: BufferHandle, layout: TensorLayout, ownership: BufferOwnership, mutability: BufferMutability, has_validity_bitmap: bool=False) -> BufferViewMetadata`

title: Bridge one Tensor layout into canonical buffer/view metadata.
parameters:
  data:
    type: BufferHandle
  owner:
    type: BufferHandle
  dtype:
    type: BufferHandle
  layout:
    type: TensorLayout
  ownership:
    type: BufferOwnership
  mutability:
    type: BufferMutability
  has_validity_bitmap:
    type: bool
returns:
  type: BufferViewMetadata

`tensor_element_size_bytes_from_dtype(dtype: BufferHandle) -> int | None`

title: Return the byte width for one canonical primitive dtype handle.
parameters:
  dtype:
    type: BufferHandle
returns:
  type: int | None

Classes

`TensorOrder(str, Enum)`

title: Canonical contiguous layout order for Tensor helpers.

`TensorLayout`

title: Static Tensor layout metadata.
summary: >-
  Represent the logical rank, shape, strides, and byte offset of one Tensor
  value without duplicating the lower-level storage machinery.
attributes:
  shape:
    type: tuple[int, Ellipsis]
  strides:
    type: tuple[int, Ellipsis]
  offset_bytes:
    type: int

Methods

`ndim(self) -> int`

title: Return the rank encoded by the layout.
returns:
  type: int