How to specify indexing_maps in linalg if Op has padding

For example, If the operation
linalg.pooling_nhwc_max
got an extra attributes padding, which contains two number specifying left padding in W direction,
and top padding in H directions of the input tensor.

I do not know how to write indexing_maps for this, because it seems the iteration space of indexing_maps variable only depend on dims of inputs/outputs, without relation to the attributes.

Any example how to write linalg.generic/linalg.indexed_generic for this operation,
including indexing_maps.
For simplicity, do not consider dilations and strides.

Thanks a lot