Struct anira::v2::TensorShape#
-
struct TensorShape#
The tensor shapes of a 2.x configuration: universal (every backend) or of one backend, whose row must hold the universal shapes with their axes permuted or unit axes inserted (anira 3 keeps one shape per tensor and an axis layout per engine; a row that reshapes is refused with ANIRA_ERROR_JSON when the configuration is built).
Public Functions
-
TensorShape() = delete#
-
inline TensorShape(TensorShapeList input_shape, TensorShapeList output_shape)#
A universal row.
- Throws:
Error{ANIRA_ERROR_INVALID_ARGUMENT} – for an empty list.
-
inline TensorShape(TensorShapeList input_shape, TensorShapeList output_shape, InferenceBackend backend)#
A row of one backend.
- Throws:
Error{ANIRA_ERROR_INVALID_ARGUMENT} – for an empty list.
-
inline bool is_universal() const noexcept#
-
inline bool operator==(const TensorShape &other) const noexcept#
The 2.x rule: two universal rows by their shapes, two backend rows by their shapes and backend, never one of each.
-
inline bool operator!=(const TensorShape &other) const noexcept#
Public Members
-
TensorShapeList m_tensor_input_shape#
-
TensorShapeList m_tensor_output_shape#
-
InferenceBackend m_backend = CUSTOM#
a backend row’s backend; CUSTOM on a universal row
-
bool m_universal = false#
-
TensorShape() = delete#