1 /* 2 * Copyright (c) 2018-2020 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_CLWINOGRADINPUTTRANSFORMKERNEL_H 25 #define ARM_COMPUTE_CLWINOGRADINPUTTRANSFORMKERNEL_H 26 27 #include "src/core/CL/ICLKernel.h" 28 29 namespace arm_compute 30 { 31 class ICLTensor; 32 33 /** OpenCL kernel to perform Winograd input transform.*/ 34 class CLWinogradInputTransformKernel : public ICLKernel 35 { 36 public: 37 /** Default constructor */ 38 CLWinogradInputTransformKernel(); 39 /** Prevent instances of this class from being copied (As this class contains pointers) */ 40 CLWinogradInputTransformKernel(const CLWinogradInputTransformKernel &) = delete; 41 /** Prevent instances of this class from being copied (As this class contains pointers) */ 42 CLWinogradInputTransformKernel &operator=(const CLWinogradInputTransformKernel &) = delete; 43 /** Allow instances of this class to be moved */ 44 CLWinogradInputTransformKernel(CLWinogradInputTransformKernel &&) = default; 45 /** Allow instances of this class to be moved */ 46 CLWinogradInputTransformKernel &operator=(CLWinogradInputTransformKernel &&) = default; 47 /** Set the input and output of the kernel. 48 * 49 * @note Winograd input transform supports the following configurations for NCWH data layout 50 * F(output tile, kernel size):F(2x2, 3x3), F(2x1, 3x1), F(1x2, 1x3), 51 * F(4x4, 3x3), F(4x1, 3x1), F(1x4, 1x3), 52 * F(4x4, 5x5), F(4x1, 5x1), F(1x4, 1x5) 53 * 54 * @note Winograd input transform supports the following configurations for NHWC data layout 55 * F(output tile, kernel size):F(4x4, 3x3), F(4x1, 3x1), F(1x4, 1x3), 56 * F(4x4, 5x5), F(4x1, 5x1), F(1x4, 1x5) 57 * 58 * Strides: only unit strides 59 * 60 * @param[in] input The input tensor to transform. Data types supported: F16/F32 61 * @param[in] output The output tensor. 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 62 * @param[in] winograd_info Contains Winograd's information described in @ref WinogradInfo. 63 */ 64 void configure(const ICLTensor *input, ICLTensor *output, const WinogradInfo &winograd_info); 65 /** Set the input and output of the kernel. 66 * 67 * @note Winograd input transform supports the following configurations for NCWH data layout 68 * F(output tile, kernel size):F(2x2, 3x3), F(2x1, 3x1), F(1x2, 1x3), 69 * F(4x4, 3x3), F(4x1, 3x1), F(1x4, 1x3), 70 * F(4x4, 5x5), F(4x1, 5x1), F(1x4, 1x5) 71 * 72 * @note Winograd input transform supports the following configurations for NHWC data layout 73 * F(output tile, kernel size):F(4x4, 3x3), F(4x1, 3x1), F(1x4, 1x3), 74 * F(4x4, 5x5), F(4x1, 5x1), F(1x4, 1x5) 75 * 76 * Strides: only unit strides 77 * 78 * @param[in] compile_context The compile context to be used. 79 * @param[in] input The input tensor to transform. Data types supported: F16/F32 80 * @param[in] output The output tensor. 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 81 * @param[in] winograd_info Contains Winograd's information described in @ref WinogradInfo. 82 */ 83 void configure(const CLCompileContext &compile_context, const ICLTensor *input, ICLTensor *output, const WinogradInfo &winograd_info); 84 /** Static function to check if given info will lead to a valid configuration of @ref CLWinogradInputTransformKernel 85 * 86 * @note Winograd input transform supports the following configurations for NCWH data layout 87 * F(output tile, kernel size):F(2x2, 3x3), F(2x1, 3x1), F(1x2, 1x3), 88 * F(4x4, 3x3), F(4x1, 3x1), F(1x4, 1x3), 89 * F(4x4, 5x5), F(4x1, 5x1), F(1x4, 1x5) 90 * 91 * @note Winograd input transform supports the following configurations for NHWC data layout 92 * F(output tile, kernel size):F(4x4, 3x3), F(4x1, 3x1), F(1x4, 1x3), 93 * F(4x4, 5x5), F(4x1, 5x1), F(1x4, 1x5) 94 * 95 * Strides: only unit strides 96 * 97 * @param[in] input The input tensor to transform. Data types supported: F16/F32 98 * @param[in] output The output tensor. 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 99 * @param[in] winograd_info Contains Winograd's information described in @ref WinogradInfo. 100 * 101 * @return a status 102 */ 103 static Status validate(const ITensorInfo *input, const ITensorInfo *output, const WinogradInfo &winograd_info); 104 105 // Inherited methods overridden: 106 void run(const Window &window, cl::CommandQueue &queue) override; 107 BorderSize border_size() const override; 108 109 private: 110 using WinogradKey = std::pair<std::pair<int, int>, std::pair<int, int>>; 111 112 BorderSize _border_size; 113 const ICLTensor *_input; 114 ICLTensor *_output; 115 DataLayout _data_layout; 116 int _num_tiles_x; 117 int _num_tiles_y; 118 unsigned int _step_z; 119 }; 120 } // arm_compute 121 #endif /*ARM_COMPUTE_CLWINOGRADINPUTTRANSFORMKERNEL_H */ 122