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

pool_shape_impl

def pool_shape_impl[input_type: DType, filter_type: DType, strides_type: DType, dilations_type: DType, paddings_type: DType, ceil_mode: Bool](input_buf: TileTensor[input_type, address_space=input_buf.address_space, linear_idx_type=input_buf.linear_idx_type, element_size=input_buf.element_size], filter_buf: TileTensor[filter_type, address_space=filter_buf.address_space, linear_idx_type=filter_buf.linear_idx_type, element_size=filter_buf.element_size], strides_buf: TileTensor[strides_type, address_space=strides_buf.address_space, linear_idx_type=strides_buf.linear_idx_type, element_size=strides_buf.element_size], dilations_buf: TileTensor[dilations_type, address_space=dilations_buf.address_space, linear_idx_type=dilations_buf.linear_idx_type, element_size=dilations_buf.element_size], paddings_buf: TileTensor[paddings_type, address_space=paddings_buf.address_space, linear_idx_type=paddings_buf.linear_idx_type, element_size=paddings_buf.element_size]) -> IndexList[TileTensor[input_type, input_buf.LayoutType, input_buf.origin, address_space=input_buf.address_space, linear_idx_type=input_buf.linear_idx_type, element_size=input_buf.element_size].rank]

Compute the output shape of a pooling operation, and assert the inputs are compatible. Works for 2D pool operations only in the NHWC format.

Parameters:

input_type (DType): Type of the input tensor.
filter_type (DType): Type of the filter tensor.
strides_type (DType): Type of the strides tensor.
dilations_type (DType): Type of the dilations tensor.
paddings_type (DType): Type of the paddings tensor.
ceil_mode (Bool): Define rounding mode for shape calculation.

Args:

input_buf (TileTensor[input_type, address_space=input_buf.address_space, linear_idx_type=input_buf.linear_idx_type, element_size=input_buf.element_size]): The input tensor.
filter_buf (TileTensor[filter_type, address_space=filter_buf.address_space, linear_idx_type=filter_buf.linear_idx_type, element_size=filter_buf.element_size]): The filter size buffer.
strides_buf (TileTensor[strides_type, address_space=strides_buf.address_space, linear_idx_type=strides_buf.linear_idx_type, element_size=strides_buf.element_size]): The strides size buffer.
dilations_buf (TileTensor[dilations_type, address_space=dilations_buf.address_space, linear_idx_type=dilations_buf.linear_idx_type, element_size=dilations_buf.element_size]): The dilations size buffer.
paddings_buf (TileTensor[paddings_type, address_space=paddings_buf.address_space, linear_idx_type=paddings_buf.linear_idx_type, element_size=paddings_buf.element_size]): The paddings size buffer.

Returns:

IndexList[TileTensor[input_type, input_buf.LayoutType, input_buf.origin, address_space=input_buf.address_space, linear_idx_type=input_buf.linear_idx_type, element_size=input_buf.element_size].rank]: The output shape.