xref: /aosp_15_r20/external/executorch/backends/qualcomm/builders/op_slice_copy.py (revision 523fa7a60841cd1ecfb9cc4201f1ca8b03ed023a)
1# Copyright (c) Qualcomm Innovation Center, Inc.
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.
6from typing import cast, Dict
7
8import executorch.backends.qualcomm.python.PyQnnWrapperAdaptor as PyQnnWrapper
9
10import numpy as np
11import torch
12
13from .node_visitor import NodeVisitor, register_node_visitor
14from .qnn_constants import OpStridedSlice, QNN_OP_PACKAGE_NAME_QTI_AISW
15
16
17@register_node_visitor
18class StrideSlice(NodeVisitor):
19    target = ["aten.slice_copy.Tensor"]
20
21    def __init__(self, *args) -> None:
22        super().__init__(*args)
23
24    def define_node(
25        self,
26        node: torch.fx.Node,
27        nodes_to_wrappers: Dict[torch.fx.Node, PyQnnWrapper.TensorWrapper],
28    ) -> PyQnnWrapper.PyQnnOpWrapper:
29        input_node = node.args[0]
30        input_tensor = self.get_tensor(input_node, node)
31        tensor_type = PyQnnWrapper.Qnn_TensorType_t.QNN_TENSOR_TYPE_NATIVE
32
33        input_tensor_wrapper = self.define_tensor(
34            input_node,
35            input_tensor,
36            tensor_type,
37            nodes_to_wrappers,
38            is_input_tensor=True,
39        )
40
41        output_tensor = self.get_tensor(node, node)
42        output_tensor_wrapper = self.define_tensor(
43            node,
44            output_tensor,
45            PyQnnWrapper.Qnn_TensorType_t.QNN_TENSOR_TYPE_NATIVE,
46            nodes_to_wrappers,
47            is_input_tensor=False,
48        )
49
50        dim = cast(int, node.args[1])
51        if dim < 0:
52            dim = dim % len(input_tensor.shape)
53        start = cast(int, node.args[2])
54        if start < 0:
55            start = start % input_tensor.shape[dim]
56        end = min(cast(int, node.args[3]), input_tensor.shape[dim])
57        if end < 0:
58            end = end % input_tensor.shape[dim]
59
60        input_tensor_rank = len(input_tensor.shape)
61        ranges = []
62        for i in range(input_tensor_rank):
63            if i == dim:
64                # find step
65                step = node.args[4] if len(node.args) > 4 else 1
66                ranges.extend([start, end, step])
67            else:
68                ranges.extend([0, input_tensor.shape[i], 1])
69
70        range_shape = [input_tensor_rank, 3]
71
72        stride_slice_op = PyQnnWrapper.PyQnnOpWrapper(
73            node.name,
74            QNN_OP_PACKAGE_NAME_QTI_AISW,
75            OpStridedSlice.op_name,
76        )
77        stride_slice_op.AddInputTensors([input_tensor_wrapper])
78        stride_slice_op.AddOutputTensors([output_tensor_wrapper])
79
80        stride_slice_op.AddTensorParam(
81            OpStridedSlice.param_ranges,
82            PyQnnWrapper.Qnn_DataType_t.QNN_DATATYPE_INT_32,
83            len(range_shape),
84            range_shape,
85            np.array(ranges, dtype=np.int32),
86            True,
87        )
88
89        return stride_slice_op
90