xref: /aosp_15_r20/external/ComputeLibrary/arm_compute/runtime/CL/functions/CLDeconvolutionLayer.h (revision c217d954acce2dbc11938adb493fc0abd69584f3)
1 /*
2  * Copyright (c) 2017-2022 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_CLDECONVOLUTIONLAYER_H
25 #define ARM_COMPUTE_CLDECONVOLUTIONLAYER_H
26 
27 #include "arm_compute/runtime/CL/functions/CLDirectDeconvolutionLayer.h"
28 #include "arm_compute/runtime/CL/functions/CLGEMMDeconvolutionLayer.h"
29 #include "arm_compute/runtime/IFunction.h"
30 #include "arm_compute/runtime/IMemoryManager.h"
31 
32 #include <memory>
33 
34 namespace arm_compute
35 {
36 /** Basic function to compute the deconvolution layer. This function calls the following OpenCL kernels/functions:
37  *
38  * -# @ref CLGEMMDeconvolutionLayer
39  * -# @ref CLDirectDeconvolutionLayer
40  */
41 class CLDeconvolutionLayer : public IFunction
42 {
43 public:
44     /** Default constructor */
45     CLDeconvolutionLayer(std::shared_ptr<IMemoryManager> memory_manager = nullptr);
46 
47     ~CLDeconvolutionLayer();
48 
49     /** Set the input, weights, biases and output tensors.
50      *
51      * Valid data layouts:
52      * - NHWC
53      * - NCHW
54      *
55      * Valid data type configurations:
56      * |src0           |src1               |src2   |dst            |
57      * |:--------------|:------------------|:------|:--------------|
58      * |F16            |F16                |F16    |F16            |
59      * |F32            |F32                |F32    |F32            |
60      * |QASYMM8        |QASYMM8            |S32    |QASYMM8        |
61      * |QASYMM8        |QSYMM8_PER_CHANNEL |S32    |QASYMM8        |
62      * |QASYMM8_SIGNED |QASYMM8_SIGNED     |S32    |QASYMM8_SIGNED |
63      * |QASYMM8_SIGNED |QSYMM8_PER_CHANNEL |S32    |QASYMM8_SIGNED |
64      *
65      * @param[in,out] input        Input tensor. 3 lower dimensions represent a single input, and an optional 4th dimension for batch of inputs. Data types supported: QASYMM8_SIGNED/QASYMM8/F16/F32.
66      * @param[in]     weights      The 4d weights with dimensions [width, height, IFM, OFM]. Data type supported: Same as @p input or QSYMM8_PER_CHANNEL if @p input is QASYMM8/QASYMM8_SIGNED.
67      * @param[in]     bias         (Optional) The biases have one dimension. 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
68      * @param[out]    output       Output tensor. The output has the same number of dimensions as the @p input.
69      * @param[in]     deconv_info  Contains padding and policies to be used in the deconvolution, this is described in @ref PadStrideInfo.
70      * @param[in]     weights_info (Optional) Weights information needed for @ref CLConvolutionLayer, specifies if the weights tensor has been reshaped with @ref opencl::kernels::ClWeightsReshapeKernel.
71      *
72      */
73     void configure(ICLTensor *input, ICLTensor *weights, const ICLTensor *bias, ICLTensor *output, const PadStrideInfo &deconv_info, const WeightsInfo &weights_info = WeightsInfo());
74     /** Set the input, weights, biases and output tensors.
75      *
76      * @param[in]     compile_context The compile context to be used.
77      * @param[in,out] input           Input tensor. 3 lower dimensions represent a single input, and an optional 4th dimension for batch of inputs. Data types supported: QASYMM8_SIGNED/QASYMM8/F16/F32.
78      * @param[in]     weights         The 4d weights with dimensions [width, height, IFM, OFM]. Data type supported: Same as @p input or QSYMM8_PER_CHANNEL if @p input is QASYMM8/QASYMM8_SIGNED.
79      * @param[in]     bias            (Optional) The biases have one dimension. 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
80      * @param[out]    output          Output tensor. The output has the same number of dimensions as the @p input.
81      * @param[in]     deconv_info     Contains padding and policies to be used in the deconvolution, this is described in @ref PadStrideInfo.
82      * @param[in]     weights_info    (Optional) Weights information needed for @ref CLConvolutionLayer, specifies if the weights tensor has been reshaped with @ref opencl::kernels::ClWeightsReshapeKernel.
83      *
84      */
85     void configure(const CLCompileContext &compile_context, ICLTensor *input, ICLTensor *weights, const ICLTensor *bias, ICLTensor *output, const PadStrideInfo &deconv_info,
86                    const WeightsInfo &weights_info = WeightsInfo());
87     /** Static function to check if given info will lead to a valid configuration of @ref CLDeconvolutionLayer
88      *
89      * @param[in] input        Input tensor info. 3 lower dimensions represent a single input, and an optional 4th dimension for batch of inputs. Data types supported: QASYMM8_SIGNED/QASYMM8/F16/F32.
90      * @param[in] weights      The 4d weights info with dimensions [width, height, IFM, OFM]. Data type supported: Same as @p input or QSYMM8_PER_CHANNEL if @p input is QASYMM8/QASYMM8_SIGNED.
91      * @param[in] bias         (Optional) The biases have one dimension. 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
92      * @param[in] output       Output tensor info. The output has the same number of dimensions as the @p input.
93      * @param[in] deconv_info  Contains padding and policies to be used in the deconvolution, this is described in @ref PadStrideInfo.
94      * @param[in] weights_info (Optional) Weights information needed for @ref CLConvolutionLayer, specifies if the weights tensor has been reshaped with @ref opencl::kernels::ClWeightsReshapeKernel.
95      *
96      * @return a status
97      */
98     static Status validate(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *bias, ITensorInfo *output, const PadStrideInfo &deconv_info,
99                            const WeightsInfo &weights_info = WeightsInfo());
100 
101     static DeconvolutionMethod get_deconvolution_method(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *bias, ITensorInfo *output, const PadStrideInfo &deconv_info,
102                                                         const WeightsInfo &weights_info);
103     // Inherited methods overridden:
104     void run() override;
105     void prepare() override;
106 
107 private:
108     std::shared_ptr<IMemoryManager> _memory_manager;
109     std::unique_ptr<IFunction>      _function;
110 
111     struct Impl;
112     std::unique_ptr<Impl> _impl;
113 };
114 } // namespace arm_compute
115 #endif /* ARM_COMPUTE_CLDECONVOLUTIONLAYER_H */
116