Deeploy.TilingExtension.TilingCodegen.minimizeRectangle

Deeploy.TilingExtension.TilingCodegen.minimizeRectangle(rect: HyperRectangle, referenceShape: Sequence[int]) → Tuple[HyperRectangle, Tuple[int, ...]]

Minimize a hyperrectangle by collapsing dimensions where possible.

Reduces the dimensionality of a hyperrectangle by merging consecutive dimensions where the rectangle spans the entire reference shape. This optimization is useful for memory transfers and reduces complexity.

Parameters:

• rect (HyperRectangle) – The hyperrectangle to minimize.

• referenceShape (Sequence[int]) – The shape of the reference tensor that the rectangle is within.

Returns:

A tuple containing: - The minimized HyperRectangle with collapsed dimensions - The corresponding minimized reference shape

Return type:

Tuple[HyperRectangle, Tuple[int, …]]

Raises:

AssertionError – If the rectangle offset is non-zero when dimensions match the reference shape (indicating the rectangle spans the full dimension).

Notes

Dimensions are collapsed from right to left. When a rectangle dimension equals the reference dimension and has zero offset, it can be merged with adjacent dimensions to reduce the overall rank.

Example

>>> rect = HyperRectangle((0, 0), (2, 2))
>>> minimizeRectangle(rect, (4, 4))
    (HyperRectangle(offset=(0, 0), dims=(2, 2)), (2, 4))
>>> rect = HyperRectangle((0, 0), (2, 2))
>>> minimizeRectangle(rect, (4, 2))
    (HyperRectangle(offset=(0,), dims=(4,)), (8,))