Skip to content

Unfold / Fold

Differentiable block extraction (unfold) and block reassembly (fold) operations, supporting 1D/2D/3D with optional masking and Triton backends.

array_to_blocks and blocks_to_array are the supported public API. The raw unfold / fold kernels are internal implementation details; use the wrappers when you need autograd support.

torchlinops.functional.array_to_blocks

array_to_blocks(
    input,
    block_shape: tuple[int, ...],
    stride: tuple[int, ...] | None = None,
    mask: Bool[Tensor, ...] | None = None,
    out: Tensor | None = None,
)

Wrapper for default arguments

Source code in src/torchlinops/functional/_unfold/array_to_blocks.py
def array_to_blocks(
    input,
    block_shape: tuple[int, ...],
    stride: tuple[int, ...] | None = None,
    mask: Bool[Tensor, "..."] | None = None,
    out: Tensor | None = None,
):
    """Wrapper for default arguments"""
    return ArrayToBlocksFn.apply(input, block_shape, stride, mask, out)

torchlinops.functional.blocks_to_array

blocks_to_array(
    input,
    im_size: tuple,
    block_shape: tuple,
    stride: tuple | None = None,
    mask: Bool[Tensor, ...] | None = None,
    out: Tensor | None = None,
)

Wrapper for default arguments

Source code in src/torchlinops/functional/_unfold/array_to_blocks.py
def blocks_to_array(
    input,
    im_size: tuple,
    block_shape: tuple,
    stride: tuple | None = None,
    mask: Bool[Tensor, "..."] | None = None,
    out: Tensor | None = None,
):
    """Wrapper for default arguments"""
    return BlocksToArrayFn.apply(input, im_size, block_shape, stride, mask, out)