xref: /aosp_15_r20/external/ComputeLibrary/arm_compute/runtime/CL/functions/CLIndirectConvolutionLayer.h (revision c217d954acce2dbc11938adb493fc0abd69584f3)
1 /*
2  * Copyright (c) 2022 Arm Limited.
3  *
4  * SPDX-License-Identifier: MIT
5  *
6  * Permission is hereby granted, free of charge, to any person obtaining a copy
7  * of this software and associated documentation files (the "Software"), to
8  * deal in the Software without restriction, including without limitation the
9  * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
10  * sell copies of the Software, and to permit persons to whom the Software is
11  * furnished to do so, subject to the following conditions:
12  *
13  * The above copyright notice and this permission notice shall be included in all
14  * copies or substantial portions of the Software.
15  *
16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
19  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
22  * SOFTWARE.
23  */
24 #ifndef ARM_COMPUTE_CLINDIRECTCONVOLUTIONLAYER_H
25 #define ARM_COMPUTE_CLINDIRECTCONVOLUTIONLAYER_H
26 
27 #include "arm_compute/core/Types.h"
28 #include "arm_compute/runtime/IFunction.h"
29 
30 #include <memory>
31 
32 namespace arm_compute
33 {
34 class CLCompileContext;
35 class ICLTensor;
36 class ITensorInfo;
37 
38 /** Basic function to run the indirect convolution function
39  */
40 class CLIndirectConvolutionLayer : public IFunction
41 {
42 public:
43     /** Constructor */
44     CLIndirectConvolutionLayer();
45     /** Destructor */
46     ~CLIndirectConvolutionLayer();
47     /** Prevent instances of this class from being copied (As this class contains pointers) */
48     CLIndirectConvolutionLayer(const CLIndirectConvolutionLayer &) = delete;
49     /** Default move constructor */
50     CLIndirectConvolutionLayer(CLIndirectConvolutionLayer &&);
51     /** Prevent instances of this class from being copied (As this class contains pointers) */
52     CLIndirectConvolutionLayer &operator=(const CLIndirectConvolutionLayer &) = delete;
53     /** Default move assignment operator */
54     CLIndirectConvolutionLayer &operator=(CLIndirectConvolutionLayer &&);
55     /** Set the input and output tensors.
56      *
57      * Valid data layouts:
58      * - NHWC
59      *
60      * Valid data type configurations:
61      * |src0           |src1           |src2   |dst            |
62      * |:--------------|:--------------|:------|:--------------|
63      * |F16            |F16            |F16    |F16            |
64      * |F32            |F32            |F32    |F32            |
65      *
66      * @param[in]  input     Source tensor. 3 lower dimensions represent a single input,
67      *                       while every optional dimension from 4 and above represent a batch of inputs.
68      *                       Data types supported: F16/F32.
69      * @param[in]  weights   Weights tensor. Weights are 4D tensor. Data type supported:Same as @p input.
70      * @param[in]  biases    Biases tensor. Shared biases supported. Biases are 1D tensor with dimensions [OFM].
71      *                       Data type supported: Should match @p input data type.
72      * @param[out] output    Destination tensor. 3 lower dimensions represent a single output, while the rest represent batch of outputs.
73      *                       Data types supported: Same as @p input.
74      * @param[in]  conv_info Contains padding and stride information described in @ref PadStrideInfo.
75      * @param[in]  act_info  (Optional) Activation layer information in case of a fused activation.
76      */
77     void configure(ICLTensor *input, const ICLTensor *weights, const ICLTensor *biases, ICLTensor *output, const PadStrideInfo &conv_info, const ActivationLayerInfo &act_info = ActivationLayerInfo());
78     /** Set the input and output tensors.
79      *
80      * @param[in]  compile_context The compile context to be used.
81      * @param[in]  input           Source tensor. 3 lower dimensions represent a single input,
82      *                             while every optional dimension from 4 and above represent a batch of inputs.
83      *                             Data types supported: F16/F32.
84      * @param[in]  weights         Weights tensor. Weights are 4D tensor with dimensions. Data type supported:Same as @p input.
85      * @param[in]  biases          Biases tensor. Shared biases supported. Biases are 1D tensor with dimensions [OFM].
86      *                             Data type supported: Should match @p input data type.
87      * @param[out] output          Destination tensor. 3 lower dimensions represent a single output, while the rest represent batch of outputs.
88      *                             Data types supported: Same as @p input.
89      * @param[in]  conv_info       Contains padding and stride information described in @ref PadStrideInfo.
90      * @param[in]  act_info        (Optional) Activation layer information in case of a fused activation.
91      */
92     void configure(const CLCompileContext &compile_context, ICLTensor *input, const ICLTensor *weights, const ICLTensor *biases, ICLTensor *output, const PadStrideInfo &conv_info,
93                    const ActivationLayerInfo &act_info = ActivationLayerInfo());
94     /** Static function to check if given info will lead to a valid configuration of @ref CLIndirectConvolutionLayer
95      *
96      * @param[in] input     Source tensor. 3 lower dimensions represent a single input,
97      *                      while every optional dimension from 4 and above represent a batch of inputs.
98      *                      Data types supported: F16/F32.
99      * @param[in] weights   Weights tensor. Weights are 4D tensor with dimensions. Data type supported:Same as @p input.
100      * @param[in] biases    Biases tensor. Shared biases supported. Biases are 1D tensor with dimensions [OFM].
101      *                      Data type supported: Should match @p input data type.
102      * @param[in] output    Destination tensor. 3 lower dimensions represent a single output, while the rest represent batch of outputs.
103      *                      Data types supported: Same as @p input.
104      * @param[in] conv_info Contains padding and stride information described in @ref PadStrideInfo.
105      * @param[in] act_info  (Optional) Activation layer information in case of a fused activation.
106      *
107      * @return a status
108      */
109     static Status validate(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *biases, const ITensorInfo *output, const PadStrideInfo &conv_info,
110                            const ActivationLayerInfo &act_info = ActivationLayerInfo());
111 
112     // Inherited methods overridden:
113     void run() override;
114 
115 private:
116     struct Impl;
117     std::unique_ptr<Impl> _impl;
118 };
119 }
120 #endif /* ARM_COMPUTE_CLINDIRECTCONVOLUTIONLAYER_H */
121