xref: /aosp_15_r20/external/ComputeLibrary/src/gpu/cl/kernels/ClWinogradInputTransformKernel.h (revision c217d954acce2dbc11938adb493fc0abd69584f3)
1 /*
2  * Copyright (c) 2018-2022 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_CL_WINOGRAD_INPUT_TRANSFORM_KERNEL_H
25 #define ARM_COMPUTE_CL_WINOGRAD_INPUT_TRANSFORM_KERNEL_H
26 
27 #include "arm_compute/core/KernelDescriptors.h"
28 #include "src/core/common/Macros.h"
29 #include "src/gpu/cl/ClCompileContext.h"
30 #include "src/gpu/cl/IClKernel.h"
31 
32 namespace arm_compute
33 {
34 namespace opencl
35 {
36 namespace kernels
37 {
38 /** OpenCL kernel to perform Winograd input transform.*/
39 class ClWinogradInputTransformKernel : public IClKernel
40 {
41 public:
42     ClWinogradInputTransformKernel();
43     ARM_COMPUTE_DISALLOW_COPY_ALLOW_MOVE(ClWinogradInputTransformKernel);
44     /** Set the input and output of the kernel.
45      *
46      * @note Winograd input transform supports the following configurations for NCWH data layout
47      *       F(output tile, kernel size):F(2x2, 3x3), F(2x1, 3x1), F(1x2, 1x3),
48      *                                   F(4x4, 3x3), F(4x1, 3x1), F(1x4, 1x3),
49      *                                   F(4x4, 5x5), F(4x1, 5x1), F(1x4, 1x5)
50      *
51      * @note Winograd input transform supports the following configurations for NHWC data layout
52      *       F(output tile, kernel size):F(4x4, 3x3), F(4x1, 3x1), F(1x4, 1x3),
53      *                                   F(4x4, 5x5), F(4x1, 5x1), F(1x4, 1x5)
54      *
55      *       Strides: only unit strides
56      *
57      * @param[in] compile_context The compile context to be used.
58      * @param[in] src             The input tensor info to transform. Data types supported: F16/F32
59      * @param[in] dst             The output tensor info. The shape for this tensor can be calculated using the utility function @p compute_winograd_input_transform_shape. Data types supported: Same as @p input
60      * @param[in] winograd_info   Contains Winograd's information described in @ref WinogradInfo.
61      */
62     void configure(const ClCompileContext &compile_context, ITensorInfo *src, ITensorInfo *dst, const WinogradInfo &winograd_info);
63     /** Static function to check if given info will lead to a valid configuration
64      *
65      * Similar to ClWinogradInputTransformKernel::configure()
66      *
67      * @return a status
68      */
69     static Status validate(const ITensorInfo *src, const ITensorInfo *dst, const WinogradInfo &winograd_info);
70 
71     // Inherited methods overridden:
72     void run_op(ITensorPack &tensors, const Window &window, cl::CommandQueue &queue) override;
73     BorderSize border_size() const override;
74 
75 private:
76     using WinogradKey = std::pair<std::pair<int, int>, std::pair<int, int>>;
77 
78     BorderSize   _border_size{ 0 };
79     DataLayout   _data_layout{ DataLayout::UNKNOWN };
80     int          _num_tiles_x{ 0 };
81     int          _num_tiles_y{ 0 };
82     unsigned int _step_z{ 1 };
83     int32_t      _src_width{ 0 };
84     int32_t      _src_height{ 0 };
85 };
86 } // namespace kernels
87 } // namespace opencl
88 } // namespace arm_compute
89 #endif /* ARM_COMPUTE_CL_WINOGRAD_INPUT_TRANSFORM_KERNEL_H */
90