xref: /aosp_15_r20/external/ComputeLibrary/arm_compute/runtime/NEON/functions/NEDirectConvolutionLayer.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_NEDIRECTCONVOLUTIONLAYER_H
25 #define ARM_COMPUTE_NEDIRECTCONVOLUTIONLAYER_H
26 
27 #include "arm_compute/core/Types.h"
28 #include "arm_compute/runtime/IFunction.h"
29 #include "arm_compute/runtime/IMemoryManager.h"
30 #include "arm_compute/runtime/MemoryGroup.h"
31 
32 #include <memory>
33 
34 namespace arm_compute
35 {
36 class ITensor;
37 class ITensorInfo;
38 /** Function to run the direct convolution.
39  *
40  *  This function calls the following:
41  *
42  * -# @ref cpu::CpuDirectConv2d
43  */
44 class NEDirectConvolutionLayer : public IFunction
45 {
46 public:
47     /** Constructor */
48     NEDirectConvolutionLayer(std::shared_ptr<IMemoryManager> memory_manager = nullptr);
49     /** Prevent instances of this class from being copied (As this class contains pointers) */
50     NEDirectConvolutionLayer(const NEDirectConvolutionLayer &) = delete;
51     /** Prevent instances of this class from being copied (As this class contains pointers) */
52     NEDirectConvolutionLayer &operator=(const NEDirectConvolutionLayer &) = delete;
53     /** Prevent instances of this class from being moved (As this class contains non movable objects) */
54     NEDirectConvolutionLayer(NEDirectConvolutionLayer &&) = delete;
55     /** Prevent instances of this class from being moved (As this class contains non movable objects) */
56     NEDirectConvolutionLayer &operator=(NEDirectConvolutionLayer &&) = delete;
57     /** Default destructor */
58     ~NEDirectConvolutionLayer();
59     /** Set the input, weights, biases and output tensors.
60      *
61      * Valid data layouts:
62      * - NHWC
63      * - NCHW
64      *
65      * Valid data type configurations:
66      * |src0   |src1   |src2   |dst    |
67      * |:------|:------|:------|:------|
68      * |F16    |F16    |F16    |F16    |
69      * |F32    |F32    |F32    |F32    |
70      *
71      * @note: DirectConvolution only works in the following configurations:
72      *    1x1 convolution with stride_x = 1/2/3, stride_y = 1/2/3 data type = F16/F32
73      *    3x3 convolution with stride_x = 1/2/3, stride_y = 1/2/3 data type = F16/F32
74      *    5x5 convolution with stride_x = 1/2/3, stride_y = 1/2/3 data type = F32
75      *
76      * @param[in, out] input     Input tensor. Data types supported: F16/F32.
77      * @param[in]      weights   Set of kernels to convolve the input volume.
78      *                           Supported sizes: 1x1, 3x3 and 5x5.
79      *                           The 3rd dimension must be the same as the input's volume 3rd dimension.
80      *                           Data type supported: Same as @p input.
81      * @param[in]      bias      Set of biases. Can be nullptr. Data type supported: Same as @p input.
82      * @param[out]     output    Output tensor.
83      *                           The 3rd dimensions must be equal to the 4th dimension of the @p kernels tensor. Data types supported: Same as @p input.
84      * @param[in]      conv_info Contains padding and stride information described in @ref PadStrideInfo.
85      * @param[in]      act_info  (Optional) Activation layer information in case of a fused activation.
86      */
87     void configure(ITensor *input, const ITensor *weights, const ITensor *bias, ITensor *output, const PadStrideInfo &conv_info, const ActivationLayerInfo &act_info = ActivationLayerInfo());
88     /** Static function to check if given info will lead to a valid configuration of @ref NEDirectConvolutionLayer
89      *
90      * @note: DirectConvolution only works in the following configurations:
91      *    1x1 convolution with stride_x = 1/2/3, stride_y = 1/2/3 data type = F16/F32
92      *    3x3 convolution with stride_x = 1/2/3, stride_y = 1/2/3 data type = F16/F32
93      *    5x5 convolution with stride_x = 1/2/3, stride_y = 1/2/3 data type = F32
94      *
95      * @param[in] input     Input tensor. Data types supported: F16/F32.
96      * @param[in] weights   Set of kernels to convolve the input volume.
97      *                      Supported sizes: 1x1, 3x3 and 5x5.
98      *                      The 3rd dimension must be the same as the input's volume 3rd dimension.
99      *                      Data type supported: Same as @p input.
100      * @param[in] bias      Set of biases. Can be nullptr. Data type supported: Same as @p input.
101      * @param[in] output    Output tensor.
102      *                      The 3rd dimensions must be equal to the 4th dimension of the @p kernels tensor. Data types supported: Same as @p input.
103      * @param[in] conv_info Contains padding and stride information described in @ref PadStrideInfo.
104      * @param[in] act_info  (Optional) Activation layer information in case of a fused activation.
105      *
106      * @return a status
107      */
108     static Status validate(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *bias, const ITensorInfo *output, const PadStrideInfo &conv_info,
109                            const ActivationLayerInfo &act_info = ActivationLayerInfo());
110 
111     // Inherited methods overridden:
112     void run() override;
113 
114 private:
115     struct Impl;
116     std::shared_ptr<IMemoryManager> _memory_manager;
117     std::unique_ptr<Impl>           _impl;
118 };
119 } // namespace arm_compute
120 #endif /* ARM_COMPUTE_NEDIRECTCONVOLUTIONLAYER_H */
121