xref: /aosp_15_r20/external/executorch/backends/qualcomm/builders/op_pow.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 Dict
7
8import executorch.backends.qualcomm.python.PyQnnWrapperAdaptor as PyQnnWrapper
9
10import torch
11from executorch.backends.qualcomm.utils.constants import QCOM_QUANT_ATTRS
12from executorch.exir.dialects._ops import ops as exir_ops
13
14from .node_visitor import NodeVisitor, register_node_visitor
15from .qnn_constants import OpElementWisePower, QNN_OP_PACKAGE_NAME_QTI_AISW
16
17
18# TODO Add more class Like PowTensorTensor if needed
19@register_node_visitor
20class PowTensorScalar(NodeVisitor):
21    target = ["aten.pow.Tensor_Scalar"]
22
23    def __init__(self, *args) -> None:
24        super().__init__(*args)
25
26    def define_node(
27        self,
28        node: torch.fx.Node,
29        nodes_to_wrappers: Dict[torch.fx.Node, PyQnnWrapper.TensorWrapper],
30    ) -> PyQnnWrapper.PyQnnOpWrapper:
31        out_tensor = self.get_tensor(node, node)
32        output_tensor_wrapper = self.define_tensor(
33            node,
34            out_tensor,
35            PyQnnWrapper.Qnn_TensorType_t.QNN_TENSOR_TYPE_NATIVE,
36            nodes_to_wrappers,
37            is_input_tensor=False,
38        )
39        pow_output_tensors = [output_tensor_wrapper]
40
41        # tensor input
42        input_node = node.args[0]
43        input_tensor = self.get_tensor(input_node, node)
44
45        tensor_type = PyQnnWrapper.Qnn_TensorType_t.QNN_TENSOR_TYPE_NATIVE
46
47        input_tensor_wrapper = self.define_tensor(
48            input_node,
49            input_tensor,
50            tensor_type,
51            nodes_to_wrappers,
52            is_input_tensor=True,
53        )
54
55        # scalar input
56        scalar = node.args[1]
57        scalar_tensor = torch.tensor(scalar).to(torch.float32)
58
59        # 'graph', 'name', 'op', 'target', 'args', and 'kwargs'
60        scalar_node = torch.fx.Node(
61            node.graph,
62            node.name + "_runtime_scalar",
63            "call_function",
64            exir_ops.edge.aten.scalar_tensor.default,
65            (),  # args
66            {},  # kwargs
67        )
68
69        if pow_quant_attrs := node.meta.get(QCOM_QUANT_ATTRS):
70            quant_attrs = pow_quant_attrs.copy()
71            quant_range = quant_attrs["quant_max"] - quant_attrs["quant_min"]
72            quant_attrs["zero_point"] = 0 if scalar >= 0 else quant_attrs["quant_max"]
73            quant_attrs["scale"] = (
74                scalar / quant_range if scalar >= 0 else -scalar / quant_range
75            )
76            scalar_node.meta[QCOM_QUANT_ATTRS] = quant_attrs
77
78        scalar_tensor_wrapper = self.define_tensor(
79            scalar_node,
80            scalar_tensor,
81            PyQnnWrapper.Qnn_TensorType_t.QNN_TENSOR_TYPE_STATIC,
82            nodes_to_wrappers,
83            is_input_tensor=False,
84        )
85
86        pow_input_tensors = [input_tensor_wrapper, scalar_tensor_wrapper]
87
88        pow_op = PyQnnWrapper.PyQnnOpWrapper(
89            node.name,
90            QNN_OP_PACKAGE_NAME_QTI_AISW,
91            OpElementWisePower.op_name,
92        )
93        pow_op.AddInputTensors(pow_input_tensors)
94        pow_op.AddOutputTensors(pow_output_tensors)
95
96        return pow_op
97