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

reshape_shape

def reshape_shape[output_rank: Int, input_type: DType, target_shape_type: DType](input_buf: TileTensor[input_type, Storage=input_buf.Storage, address_space=input_buf.address_space, linear_idx_type=input_buf.linear_idx_type], target_shape_buf: TileTensor[target_shape_type, Storage=target_shape_buf.Storage, address_space=target_shape_buf.address_space, linear_idx_type=target_shape_buf.linear_idx_type]) -> IndexList[output_rank]

Reads a target shape from a buffer and returns it as a static IndexList, inferring any single -1 dimension.

Copies the rank-1 target_shape_buf into a static IndexList[output_rank], allowing at most one dimension to be specified as -1, which is then inferred from the input's element count. Raises an error if the constraints are violated or the inferred shape does not match the input's number of elements.

Parameters:

output_rank (Int): Expected rank of the target shape.
input_type (DType): Element type of the input buffer.
target_shape_type (DType): Element type of the target shape buffer.

Args:

input_buf (TileTensor[input_type, Storage=input_buf.Storage, address_space=input_buf.address_space, linear_idx_type=input_buf.linear_idx_type]): Source tensor whose element count constrains the inferred shape.
target_shape_buf (TileTensor[target_shape_type, Storage=target_shape_buf.Storage, address_space=target_shape_buf.address_space, linear_idx_type=target_shape_buf.linear_idx_type]): Rank-1 tensor holding the desired output dimensions, where at most one entry may be -1.

Returns:

IndexList[output_rank]: A static IndexList[output_rank] with any -1 dimension resolved.

Raises:

Error: If target_shape_buf is not rank 1, the rank does not match output_rank, more than one -1 is present, a negative value other than -1 appears, or the resulting element count does not match the input.