xref: /aosp_15_r20/external/ComputeLibrary/src/core/CL/kernels/CLDepthwiseConvolutionLayerNativeKernel.h (revision c217d954acce2dbc11938adb493fc0abd69584f3)
1 /*
2  * Copyright (c) 2019-2022 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_CLDEPTHWISECONVOLUTIONLAYERNATIVEKERNEL_H
25 #define ARM_COMPUTE_CLDEPTHWISECONVOLUTIONLAYERNATIVEKERNEL_H
26 
27 #include "src/core/CL/ICLKernel.h"
28 
29 #include "arm_compute/core/KernelDescriptors.h"
30 
31 namespace arm_compute
32 {
33 class ICLTensor;
34 
35 /** Interface for the kernel to run a MxN depthwise convolution. M and N are respectively the rows and columns of the filter
36     This kernel assumes that tensor for the weights is NOT reshaped (Native version) */
37 class CLDepthwiseConvolutionLayerNativeKernel : public ICLKernel
38 {
39 public:
40     /** Default Constructor */
41     CLDepthwiseConvolutionLayerNativeKernel();
42     /** Prevent instances of this class from being copied (As this class contains pointers) */
43     CLDepthwiseConvolutionLayerNativeKernel(const CLDepthwiseConvolutionLayerNativeKernel &) = delete;
44     /** Prevent instances of this class from being copied (As this class contains pointers) */
45     CLDepthwiseConvolutionLayerNativeKernel &operator=(const CLDepthwiseConvolutionLayerNativeKernel &) = delete;
46     /** Allow instances of this class to be moved */
47     CLDepthwiseConvolutionLayerNativeKernel(CLDepthwiseConvolutionLayerNativeKernel &&) = default;
48     /** Allow instances of this class to be moved */
49     CLDepthwiseConvolutionLayerNativeKernel &operator=(CLDepthwiseConvolutionLayerNativeKernel &&) = default;
50 
51     /** Initialize the function's source, destination and parameters
52      *
53      * @param[in]  compile_context    The compile context to be used.
54      * @param[in]  input              Source tensor. Data type supported: QASYMM8/QASYMM8_SIGNED/FP32/FP16. Data layout supported: NHWC
55      * @param[in]  weights            Weights tensor. A 3D tensor with dimensions [IFM, N, M].
56      *                                Data type supported: Same as @p input or QASYMM8/QASYMM8_SIGNED/QSYMM8_PER_CHANNEL when @p input is QASYMM8.
57      * @param[in]  biases             Biases tensor. A 1D tensor with dimensions [IFM]. Must be nullptr if not needed.
58      *                                Data type supported: Same as @p input, S32 when input is QASYMM8/QASYMM8_SIGNED.
59      * @param[out] output             Destination tensor. Pass in nullptr or @p input for in-place operation. Data type supported: Same as @p input.
60      * @param[in]  dwc_info           Depthwise convolution layer info
61      * @param[in]  conv_info          Convolution info (padding, stride, dilation, ...)
62      * @param[in]  output_multipliers (Optional) Output multipliers tensor for quantized computations. In case of per-channel quantization,
63      *                                the number of multipliers must be equal to the number of filters (IFM). Supported data types: S32
64      * @param[in]  output_shifts      (Optional) Output shifts tensor for quantized computations. In case of per-channel quantization,
65      *                                the number of multipliers must be equal to the number of filters (IFM). Supported data types: S32
66      *
67      * @note: In-place is only supported when
68      *          * data layout: NHWC
69      *          * filter: 1x1
70      *          * @p depth_multiplier: 1
71      *          * strides: 1
72      *          * dilation: 1
73      *          * no padding
74      *          * no change of data layout after configure
75      */
76     void configure(const CLCompileContext &compile_context, ICLTensor *input, const ICLTensor *weights, const ICLTensor *biases, ICLTensor *output, const DWCComputeKernelInfo &dwc_info,
77                    const ConvolutionInfo &conv_info, const ICLTensor *output_multipliers = nullptr, const ICLTensor *output_shifts = nullptr);
78 
79     /** Static function to check if given info will lead to a valid configuration of @ref CLDepthwiseConvolutionLayerNativeKernel
80      *
81      * Similar to @ref CLDepthwiseConvolutionLayerNativeKernel::configure()
82      */
83     void configure(ICLTensor *input, const ICLTensor *weights, const ICLTensor *biases, ICLTensor *output, const DWCComputeKernelInfo &dwc_info,
84                    const ConvolutionInfo &conv_info, const ICLTensor *output_multipliers = nullptr, const ICLTensor *output_shifts = nullptr);
85 
86     /** Static function to check if given info will lead to a valid configuration of @ref CLDepthwiseConvolutionLayerNativeKernel
87      *
88      * Similar to @ref CLDepthwiseConvolutionLayerNativeKernel::configure()
89      *
90      * @return a status
91      */
92     static Status validate(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *biases, const ITensorInfo *output, const DWCComputeKernelInfo &dwc_info,
93                            const ConvolutionInfo &conv_info, const ITensorInfo *output_multipliers = nullptr, const ITensorInfo *output_shifts = nullptr);
94 
95     // Inherited methods overridden:
96     void run(const Window &window, cl::CommandQueue &queue) override;
97 
98 private:
99     const ICLTensor *_input {};
100     const ICLTensor *_weights{};
101     const ICLTensor *_biases{};
102     ICLTensor       *_output{};
103     unsigned int     _depth_multiplier{ 0 };
104     const ICLTensor *_output_multipliers{};
105     const ICLTensor *_output_shifts{};
106     bool             _export_input_to_cl_image{ false };
107     bool             _export_weights_to_cl_image{ true };
108     bool             _is_quantized{ false };
109 };
110 } // namespace arm_compute
111 #endif /*ARM_COMPUTE_CLDEPTHWISECONVOLUTIONLAYERNATIVEKERNEL_H */
112