1 /* 2 * Copyright (c) 2017-2020 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_NEIM2COLKERNEL_H 25 #define ARM_COMPUTE_NEIM2COLKERNEL_H 26 27 #include "src/core/NEON/INEKernel.h" 28 29 namespace arm_compute 30 { 31 class ITensor; 32 class Size2D; 33 34 /** Interface for the im2col reshape kernel. 35 * 36 * Rearranges image blocks into columns. It is used to strip out each convolution block to a single column. 37 * It is used to transform a convolution to a plain matrix multiplication. 38 * 39 * For example taking into account the image below and assuming 3x3 image blocks with stride of 1 we have: 40 * 41 * @f[ 42 * \left( \begin{array}{cccc} 43 * a00 & a01 & a02 & a03 \\ 44 * a10 & a11 & a12 & a13 \\ 45 * a20 & a21 & a22 & a23 \\ 46 * a30 & a31 & a32 & a33 \\ 47 * \end{array} \right) 48 * \rightarrow 49 * \left( \begin{array}{ccccccccc} 50 * a00 & a01 & a02 & a10 & a11 & a12 & a20 & a21 & a22 \\ 51 * a01 & a02 & a03 & a11 & a12 & a13 & a21 & a22 & a23 \\ 52 * a10 & a11 & a12 & a20 & a21 & a22 & a30 & a31 & a32 \\ 53 * a11 & a12 & a13 & a21 & a22 & a23 & a31 & a32 & a33 \\ 54 * \end{array} \right) 55 * @f] 56 */ 57 class NEIm2ColKernel : public INEKernel 58 { 59 public: name()60 const char *name() const override 61 { 62 return "NEIm2ColKernel"; 63 } 64 /** Default constructor */ 65 NEIm2ColKernel(); 66 /** Prevent instances of this class from being copied (As this class contains pointers) */ 67 NEIm2ColKernel(const NEIm2ColKernel &) = delete; 68 /** Prevent instances of this class from being copied (As this class contains pointers) */ 69 NEIm2ColKernel &operator=(const NEIm2ColKernel &) = delete; 70 /** Allow instances of this class to be moved */ 71 NEIm2ColKernel(NEIm2ColKernel &&) = default; 72 /** Allow instances of this class to be moved */ 73 NEIm2ColKernel &operator=(NEIm2ColKernel &&) = default; 74 /** Default destructor */ 75 ~NEIm2ColKernel() = default; 76 77 /** Set the input and output of the kernel. 78 * 79 * @param[in] input The input tensor to convert. 3 lower dimensions represent a single input [width, height, IFM], 80 * while every optional dimension from 4 and above represent a batch of inputs. 81 * Data types supported: QASYMM8/QASYMM8_SIGNED/BFLOAT16/F16/F32 82 * Note: QASYMM8/QASYMM8_SIGNED works only for has_bias = false 83 * @param[out] output The output tensor. Data types supported: Same as @p input 84 * @param[in] kernel_dims The kernel dimensions (width and height). 85 * @param[in] conv_info Contains padding and stride information described in @ref PadStrideInfo. 86 * @param[in] has_bias In case biases are provided expands the matrix with 1. 87 * @param[in] dilation (Optional) Dilation, in elements, across x and y. Defaults to (1, 1). 88 * @param[in] num_groups (Optional) Number of groups when performing a grouped convolution. num_groups != 1 is not supported 89 */ 90 void configure(const ITensor *input, ITensor *output, const Size2D &kernel_dims, const PadStrideInfo &conv_info, 91 bool has_bias, const Size2D &dilation = Size2D(1U, 1U), unsigned int num_groups = 1); 92 /** Static function to check if given info will lead to a valid configuration of @ref NEIm2ColKernel 93 * 94 * @param[in] input The input tensor to convert. 3 lower dimensions represent a single input [width, height, IFM], 95 * while every optional dimension from 4 and above represent a batch of inputs. 96 * Data types supported: QASYMM8/QASYMM8_SIGNED/BFLOAT16/F16/F32 97 * Note: QASYMM8/QASYMM8_SIGNED works only for has_bias = false 98 * @param[in] output The output tensor. Data types supported: Same as @p input 99 * @param[in] kernel_dims The kernel dimensions (width and height). 100 * @param[in] conv_info Contains padding and stride information described in @ref PadStrideInfo. 101 * @param[in] has_bias In case biases are provided expands the matrix with 1. 102 * @param[in] dilation (Optional) Dilation, in elements, across x and y. Defaults to (1, 1). 103 * @param[in] num_groups (Optional) Number of groups when performing a grouped convolution. num_groups != 1 is not supported 104 * 105 * @return a status 106 */ 107 static Status validate(const ITensorInfo *input, const ITensorInfo *output, const Size2D &kernel_dims, const PadStrideInfo &conv_info, 108 bool has_bias, const Size2D &dilation = Size2D(1U, 1U), unsigned int num_groups = 1); 109 110 // Inherited methods overridden: 111 void run(const Window &window, const ThreadInfo &info) override; 112 113 private: 114 /** Template function to run im2col 115 * 116 * @param[in] window Region on which to execute the kernel. (Must be a valid region of the window returned by window()). 117 */ 118 template <typename T, bool has_pads, bool is_nchw> 119 void run_im2col(const Window &window); 120 121 /** Common signature for all the specialised im2col functions 122 * 123 * @param[in] window Region on which to execute the kernel. 124 */ 125 using Im2ColFunctionPtr = void (NEIm2ColKernel::*)(const Window &window); 126 127 Im2ColFunctionPtr _func; 128 const ITensor *_input; 129 ITensor *_output; 130 std::pair<unsigned int, unsigned int> _convolved_dims; 131 PadStrideInfo _conv_info; 132 unsigned int _kernel_width; 133 unsigned int _kernel_height; 134 bool _has_bias; 135 Size2D _dilation; 136 DataLayout _data_layout; 137 }; 138 } // namespace arm_compute 139 #endif /*ARM_COMPUTE_NEIM2COLKERNEL_H */ 140