xref: /aosp_15_r20/external/executorch/backends/xnnpack/operators/op_max_pool2d.py (revision 523fa7a60841cd1ecfb9cc4201f1ca8b03ed023a) 
1# Copyright (c) Meta Platforms, Inc. and affiliates.
2# All rights reserved.
3#
4# This source code is licensed under the BSD-style license found in the
5# LICENSE file in the root directory of this source tree.
6
7from typing import cast, Dict, List
8
9import torch
10from executorch.backends.xnnpack.operators.node_visitor import (
11    check_or_raise,
12    get_tensor_value,
13    NodeVisitor,
14    register_node_visitor,
15)
16from executorch.backends.xnnpack.serialization.xnnpack_graph_schema import (
17    XNNGraph,
18    XNNMaxPooling2d,
19    XNode,
20)
21from executorch.backends.xnnpack.utils.xnnpack_constants import XNN_FLAG_KEEP_DIMS
22
23
24@register_node_visitor
25class MaxPooling2d(NodeVisitor):
26    target = "aten.max_pool2d.default"
27
28    def __init__(self, *args) -> None:
29        super().__init__(*args)
30
31    def define_node(
32        self,
33        node: torch.fx.Node,
34        xnn_graph: XNNGraph,
35        vals_to_ids: Dict[torch.fx.Node, int],
36        debug_handle: int,
37    ) -> None:
38
39        self.define_nodes_tensor_inputs_outputs(
40            node, xnn_graph, vals_to_ids, convert_to_nhwc=True
41        )
42        kwargs = {}
43
44        kwargs["input_id"] = vals_to_ids[node.all_input_nodes[0]]
45
46        input_shape = get_tensor_value(xnn_graph.xvalues[kwargs["input_id"]]).dims
47        check_or_raise(len(input_shape) == 4, "Require input to be 4 dimensional")
48
49        # output
50        kwargs["input_id"] = vals_to_ids[node.all_input_nodes[0]]
51        kwargs["output_id"] = vals_to_ids[node]
52
53        # kernel info
54        kernal_shape = cast(List[int], node.args[1])
55        kwargs["pooling_height"] = kernal_shape[0]
56        kwargs["pooling_width"] = kernal_shape[1]
57
58        # stride info
59        stride = cast(List[int], node.args[2])
60        kwargs["stride_height"] = stride[0]
61        kwargs["stride_width"] = stride[1]
62
63        # padding info
64        kwargs["padding_top"] = 0
65        kwargs["padding_right"] = 0
66        kwargs["padding_bottom"] = 0
67        kwargs["padding_left"] = 0
68
69        if len(node.args) > 3:
70            padding_shape = cast(List[int], node.args[3])
71            kwargs["padding_top"] = padding_shape[0]
72            kwargs["padding_right"] = padding_shape[1]
73            kwargs["padding_bottom"] = padding_shape[0]
74            kwargs["padding_left"] = padding_shape[1]
75
76        # dilation info
77        kwargs["dilation_height"] = 1
78        kwargs["dilation_width"] = 1
79        if len(node._args) > 4:
80            dilation = cast(List[int], node.args[4])
81            kwargs["dilation_height"] = dilation[0]
82            kwargs["dilation_width"] = dilation[1]
83
84        kwargs["flags"] = XNN_FLAG_KEEP_DIMS
85
86        ser_node = XNode(
87            xnode_union=XNNMaxPooling2d(
88                **kwargs,
89            ),
90            debug_handle=debug_handle,
91        )
92        xnn_graph.xnodes.append(ser_node)
93