Module shape_utils 

A module contains shape utilities

Functions

compare_and_pad_shapes

pad shape to the shortter one, this is used for prepareing for matmul broadcast.

compare_and_pad_shapes_strides

pad shape and strides to the shortter one, this is used for prepareing for matmul broadcast.

get_broadcast_axes_from

Determines the axes along which broadcasting is required to match a desired result shape.

is_reshape_possible

Attempt to reshape an array without copying data. Translated from NumPy’s _attempt_nocopy_reshape function.

mt_intervals

Generates intervals for multi-threaded processing by dividing the outer loop into chunks.

mt_intervals_simd

Generates intervals for multi-threaded SIMD processing by dividing the outer loop into chunks.

predict_broadcast_shape

Predicts the broadcasted shape resulting from broadcasting two arrays.

try_pad_shape

Pads a shape with ones on the left to reach a specified length.

yield_one_after

Inserts a 1 into a shape vector immediately after a specified index.

yield_one_before

Inserts a dimension of size 1 before the specified index in a shape.