Accera v1.2 Reference

accera.Array(role[, data, element_type, layout, offset, shape])

Constructs an array.

Arguments

argument	description	type/default
role	The role of the array determines if the array scope is internal or external, if the array is mutable or immutable, and if the array memory is dynamically allocated.	accera.Role
data	The contents of a constant array. Required for accera.Role.CONST arrays but should not be specified for other roles.	Python buffer or numpy.ndarray.
element_type	The array element type.	accera.ScalarType, default: accera.ScalarType.float32.
layout	The affine memory map.	accera.Array.Layout, or tuple of (integers or accera.Dimension), default: accera.Array.Layout.FIRST_MAJOR.
offset	The offset of the affine memory map	integer (positive, zero, or negative), default: 0.
shape	The array shape. Required for roles other than accera.Role.CONST, should not be specified for accera.Role.CONST.	tuple of (integers or accera.Dimension).

Examples

Construct an input array:

import accera as acc
A = acc.Array(role=acc.Role.INPUT, element_type=acc.ScalarType.float32, shape=(10, 20))  # the default layout is acc.Array.Layout.FIRST_MAJOR

Construct an input array with an explicit standard layout:

A = acc.Array(role=acc.Role.INPUT, element_type=acc.ScalarType.float32, shape=(10, 20), layout=acc.Array.Layout.LAST_MAJOR)

Construct an input array with an explicit affine memory map:

A = acc.Array(role=acc.Role.INPUT, element_type=acc.ScalarType.float32, shape=(10, 20), layout=(1, 10))

Construct an input array with an infinite (undefined) major dimension:

A = acc.Array(role=acc.Role.INPUT, element_type=acc.ScalarType.float32, shape=(10, acc.inf), layout=acc.Array.Layout.LAST_MAJOR)

Construct a input array with both runtime and compile-time dimension sizes:

M = acc.create_dimensions()
A = acc.Array(role=acc.Role.INPUT, element_type=acc.ScalarType.float32, shape=(M, 20))

Construct an input/output array:

A = acc.Array(role=acc.Role.INPUT_OUTPUT, element_type=acc.ScalarType.float32, shape=(10, 20))

Construct an input/output array with runtime input dimension sizes:

M, N = acc.create_dimensions()
A = acc.Array(role=acc.Role.INPUT_OUTPUT, element_type=acc.ScalarType.float32, shape=(M, N))

Construct an output array with runtime output dimension sizes:

M, N = acc.create_dimensions(role=acc.Role.OUTPUT)
A = acc.Array(role=acc.Role.OUTPUT, element_type=acc.ScalarType.float32, shape=(M, N))

Construct an output array with explicit affine memory map:

M, N = acc.create_dimensions(role=acc.Role.OUTPUT)
A = acc.Array(role=acc.Role.OUTPUT, element_type=acc.ScalarType.float32, shape=(M, N), layout=(1, M))

Construct a constant array:

D = np.random.rand(10, 16)
A = acc.Array(role=acc.Role.CONST, data=D)

Construct a constant array with an explicit element type and layout, which does not necessarily match the input data:

D = np.random.rand(10, 16)
A = acc.Array(role=acc.Role.CONST, element_type=acc.ScalarType.float32, layout=acc.Array.Layout.LAST_MAJOR, data=D)

Construct a temporary array:

A = acc.Array(role=acc.Role.TEMP, element_type=acc.ScalarType.float32, shape=(10, 20), layout=acc.Array.Layout.LAST_MAJOR)

---

Last update: 2023-04-17