xref: /aosp_15_r20/external/ComputeLibrary/arm_compute/runtime/CL/functions/CLDirectConvolutionLayer.h (revision c217d954acce2dbc11938adb493fc0abd69584f3)
1 /*
2  * Copyright (c) 2017-2021 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_CLDIRECTCONVOLUTIONLAYER_H
25 #define ARM_COMPUTE_CLDIRECTCONVOLUTIONLAYER_H
26 
27 #include "arm_compute/core/Types.h"
28 #include "arm_compute/runtime/CL/functions/CLActivationLayer.h"
29 #include "arm_compute/runtime/IFunction.h"
30 
31 #include <memory>
32 
33 namespace arm_compute
34 {
35 class CLCompileContext;
36 class ICLTensor;
37 class ITensorInfo;
38 
39 /** Basic function to execute direct convolution function:
40  */
41 class CLDirectConvolutionLayer : public IFunction
42 {
43 public:
44     /** Constructor */
45     CLDirectConvolutionLayer();
46     /** Destructor */
47     ~CLDirectConvolutionLayer();
48     /** Prevent instances of this class from being copied (As this class contains pointers) */
49     CLDirectConvolutionLayer(const CLDirectConvolutionLayer &) = delete;
50     /** Default move constructor */
51     CLDirectConvolutionLayer(CLDirectConvolutionLayer &&);
52     /** Prevent instances of this class from being copied (As this class contains pointers) */
53     CLDirectConvolutionLayer &operator=(const CLDirectConvolutionLayer &) = delete;
54     /** Default move assignment operator */
55     CLDirectConvolutionLayer &operator=(CLDirectConvolutionLayer &&);
56     /** Set the input and output tensors.
57      *
58      * Valid data layouts:
59      * - NHWC
60      * - NCHW
61      *
62      * Valid data type configurations:
63      * |src0           |src1           |src2   |dst            |
64      * |:--------------|:--------------|:------|:--------------|
65      * |F16            |F16            |F16    |F16            |
66      * |F32            |F32            |F32    |F32            |
67      * |QASYMM8        |QASYMM8        |S32    |QASYMM8        |
68      * |QASYMM8_SIGNED |QASYMM8_SIGNED |S32    |QASYMM8_SIGNED |
69      *
70      * @param[in]  input     Source tensor. 3 lower dimensions represent a single input [width, height, IFM],
71      *                       while every optional dimension from 4 and above represent a batch of inputs.
72      *                       Data types supported: QASYMM8_SIGNED/QASYMM8/F16/F32.
73      * @param[in]  weights   Weights tensor. Weights are 4D tensor with dimensions [kernel_x, kernel_y, IFM, OFM]. Data type supported:Same as @p input.
74      * @param[in]  biases    Biases tensor. Shared biases supported. Biases are 1D tensor with dimensions [OFM].
75      *                       Data type supported: Should match @p input data type, except for input of QASYMM8 and QASYMM8_SIGNED type where biases should be of S32 type.
76      * @param[out] output    Destination tensor. 3 lower dimensions represent a single output [width, height, OFM], while the rest represent batch of outputs.
77      *                       Data types supported: Same as @p input.
78      * @param[in]  conv_info Contains padding and stride information described in @ref PadStrideInfo.
79      * @param[in]  act_info  (Optional) Activation layer information in case of a fused activation.
80      */
81     void configure(ICLTensor *input, const ICLTensor *weights, const ICLTensor *biases, ICLTensor *output, const PadStrideInfo &conv_info, const ActivationLayerInfo &act_info = ActivationLayerInfo());
82     /** Set the input and output tensors.
83      *
84      * @param[in]  compile_context The compile context to be used.
85      * @param[in]  input           Source tensor. 3 lower dimensions represent a single input [width, height, IFM],
86      *                             while every optional dimension from 4 and above represent a batch of inputs.
87      *                             Data types supported: QASYMM8_SIGNED/QASYMM8/F16/F32.
88      * @param[in]  weights         Weights tensor. Weights are 4D tensor with dimensions [kernel_x, kernel_y, IFM, OFM]. Data type supported:Same as @p input.
89      * @param[in]  biases          Biases tensor. Shared biases supported. Biases are 1D tensor with dimensions [OFM].
90      *                             Data type supported: Should match @p input data type, except for input of QASYMM8 and QASYMM8_SIGNED type where biases should be of S32 type.
91      * @param[out] output          Destination tensor. 3 lower dimensions represent a single output [width, height, OFM], while the rest represent batch of outputs.
92      *                             Data types supported: Same as @p input.
93      * @param[in]  conv_info       Contains padding and stride information described in @ref PadStrideInfo.
94      * @param[in]  act_info        (Optional) Activation layer information in case of a fused activation.
95      */
96     void configure(const CLCompileContext &compile_context, ICLTensor *input, const ICLTensor *weights, const ICLTensor *biases, ICLTensor *output, const PadStrideInfo &conv_info,
97                    const ActivationLayerInfo &act_info = ActivationLayerInfo());
98     /** Static function to check if given info will lead to a valid configuration of @ref CLDirectConvolutionLayer
99      *
100      * @param[in] input     Source tensor. 3 lower dimensions represent a single input [width, height, IFM],
101      *                      while every optional dimension from 4 and above represent a batch of inputs.
102      *                      Data types supported: QASYMM8_SIGNED/QASYMM8/F16/F32.
103      * @param[in] weights   Weights tensor. Weights are 4D tensor with dimensions [kernel_x, kernel_y, IFM, OFM]. Data type supported:Same as @p input.
104      * @param[in] biases    Biases tensor. Shared biases supported. Biases are 1D tensor with dimensions [OFM].
105      *                      Data type supported: Should match @p input data type, except for input of QASYMM8 and QASYMM8_SIGNED type where biases should be of S32 type.
106      * @param[in] output    Destination tensor. 3 lower dimensions represent a single output [width, height, OFM], while the rest represent batch of outputs.
107      *                      Data types supported: Same as @p input.
108      * @param[in] conv_info Contains padding and stride information described in @ref PadStrideInfo.
109      * @param[in] act_info  (Optional) Activation layer information in case of a fused activation.
110      *
111      * @return a status
112      */
113     static Status validate(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *biases, const ITensorInfo *output, const PadStrideInfo &conv_info,
114                            const ActivationLayerInfo &act_info = ActivationLayerInfo());
115 
116     // Inherited methods overridden:
117     void run() override;
118 
119 private:
120     struct Impl;
121     std::unique_ptr<Impl> _impl;
122 };
123 }
124 #endif /* ARM_COMPUTE_CLDIRECTCONVOLUTIONLAYER_H */
125