For the complete documentation index, see llms.txt. Markdown versions of all pages are available by appending .md to any URL (e.g. /max/get-started.md).

Mojo function

get_conv_shape

def get_conv_shape[rank: Int, filter_packed: Bool](output: TileTensor[Storage=output.Storage, address_space=output.address_space, linear_idx_type=output.linear_idx_type], input: TileTensor[Storage=input.Storage, address_space=input.address_space, linear_idx_type=input.linear_idx_type], filter: TileTensor[Storage=filter.Storage, address_space=filter.address_space, linear_idx_type=filter.linear_idx_type], stride: IndexList[rank], dilation: IndexList[rank], pad_d: IndexList[Int(2)], pad_h: IndexList[Int(2)], pad_w: IndexList[Int(2)], num_groups: Int) -> ConvShape[rank]

Builds a ConvShape from the output, input, and filter tile tensors.

Parameters:

rank (Int): Spatial rank of the convolution (1, 2, or 3).
filter_packed (Bool): Whether the filter tensor stores spatial dimensions starting at index 1 (packed) rather than index 0.

Args:

output (TileTensor[Storage=output.Storage, address_space=output.address_space, linear_idx_type=output.linear_idx_type]): Output tile tensor in NHWC layout.
input (TileTensor[Storage=input.Storage, address_space=input.address_space, linear_idx_type=input.linear_idx_type]): Input tile tensor in NHWC layout.
filter (TileTensor[Storage=filter.Storage, address_space=filter.address_space, linear_idx_type=filter.linear_idx_type]): Filter tile tensor.
stride (IndexList[rank]): Per-dimension stride values.
dilation (IndexList[rank]): Per-dimension dilation values.
pad_d (IndexList[Int(2)]): Depth padding as (lower, upper).
pad_h (IndexList[Int(2)]): Height padding as (lower, upper).
pad_w (IndexList[Int(2)]): Width padding as (lower, upper).
num_groups (Int): Number of convolution groups.

Returns:

ConvShape[rank]: A populated ConvShape describing the convolution dimensions.