xref: /aosp_15_r20/external/ComputeLibrary/src/gpu/cl/operators/ClDirectConv2d.h (revision c217d954acce2dbc11938adb493fc0abd69584f3)
1 /*
2  * Copyright (c) 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_CL_DIRECT_CONV2D_H
25 #define ARM_COMPUTE_CL_DIRECT_CONV2D_H
26 
27 #include "src/gpu/cl/ClCompileContext.h"
28 #include "src/gpu/cl/IClKernel.h"
29 #include "src/gpu/cl/IClOperator.h"
30 
31 #include <memory>
32 
33 namespace arm_compute
34 {
35 namespace opencl
36 {
37 /** Basic function to simulate a directly convolution layer. This function calls the following OpenCL kernels:
38  *
39  * -# @ref CLFillBorderKernel (executed if padding size is different from zero)
40  * -# @ref opencl::ClDirectConv2d
41  */
42 class ClDirectConv2d : public IClOperator
43 {
44 public:
45     ClDirectConv2d() = default;
46     /** Set the src and dst tensors.
47      *
48      * @param[in]  compile_context The compile context to be used.
49      * @param[in]  src             Source tensor. 3 lower dimensions represent a single src [width, height, IFM],
50      *                             while every optional dimension from 4 and above represent a batch of srcs.
51      *                             Data types supported: QASYMM8_SIGNED/QASYMM8/F16/F32.
52      * @param[in]  weights         Weights tensor. Weights are 4D tensor with dimensions [kernel_x, kernel_y, IFM, OFM]. Data type supported:Same as @p src.
53      * @param[in]  biases          Biases tensor. Shared biases supported. Biases are 1D tensor with dimensions [OFM].
54      *                             Data type supported: Should match @p src data type, except for src of QASYMM8 and QASYMM8_SIGNED type where biases should be of S32 type.
55      * @param[out] dst             Destination tensor. 3 lower dimensions represent a single dst [width, height, OFM], while the rest represent batch of dsts.
56      *                             Data types supported: Same as @p src.
57      * @param[in]  conv_info       Contains padding and stride information described in @ref PadStrideInfo.
58      * @param[in]  act_info        (Optional) Activation layer information in case of a fused activation.
59      *
60      */
61     void configure(const CLCompileContext &compile_context, ITensorInfo *src, ITensorInfo *weights, ITensorInfo *biases, ITensorInfo *dst, const PadStrideInfo &conv_info,
62                    const ActivationLayerInfo &act_info = ActivationLayerInfo());
63     /** Static function to check if given info will lead to a valid configuration
64      *
65      * Similar to ClDirectConv2d::configure()
66      *
67      * @return a status
68      */
69     static Status validate(const ITensorInfo *src, const ITensorInfo *weights, const ITensorInfo *biases, const ITensorInfo *dst, const PadStrideInfo &conv_info,
70                            const ActivationLayerInfo &act_info = ActivationLayerInfo());
71 
72     // Inherited method overridden
73     void run(ITensorPack &tensors) override;
74 
75 private:
76     std::unique_ptr<IClKernel> _direct_conv_kernel{ nullptr };
77     std::unique_ptr<IClKernel> _src_border_handler{ nullptr };
78     std::unique_ptr<IClKernel> _activation_kernel{ nullptr };
79 };
80 } // namespace opencl
81 } // namespace arm_compute
82 #endif /* ARM_COMPUTE_CL_DIRECT_CONV2D_H */