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_CLIM2COLKERNEL_H 25 #define ARM_COMPUTE_CLIM2COLKERNEL_H 26 27 #include "arm_compute/core/Size2D.h" 28 #include "src/core/CL/ICLKernel.h" 29 30 namespace arm_compute 31 { 32 class ICLTensor; 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 * @f[ 41 * \left( \begin{array}{cccc} 42 * a00 & a01 & a02 & a03 \\ 43 * a10 & a11 & a12 & a13 \\ 44 * a20 & a21 & a22 & a23 \\ 45 * a30 & a31 & a32 & a33 \\ 46 * \end{array} \right) 47 * = 48 * \left( \begin{array}{ccccccccc} 49 * a00 & a01 & a02 & a10 & a11 & a12 & a20 & a21 & a22 \\ 50 * a01 & a02 & a03 & a11 & a12 & a13 & a21 & a22 & a23 \\ 51 * a10 & a11 & a12 & a20 & a21 & a22 & a30 & a31 & a32 \\ 52 * a11 & a12 & a13 & a21 & a22 & a23 & a31 & a32 & a33 \\ 53 * \end{array} \right) 54 * @f] 55 */ 56 class CLIm2ColKernel : public ICLKernel 57 { 58 public: 59 /** Default constructor */ 60 CLIm2ColKernel(); 61 /** Prevent instances of this class from being copied (As this class contains pointers) */ 62 CLIm2ColKernel(const CLIm2ColKernel &) = delete; 63 /** Prevent instances of this class from being copied (As this class contains pointers) */ 64 CLIm2ColKernel &operator=(const CLIm2ColKernel &) = delete; 65 /** Allow instances of this class to be moved */ 66 CLIm2ColKernel(CLIm2ColKernel &&) = default; 67 /** Allow instances of this class to be moved */ 68 CLIm2ColKernel &operator=(CLIm2ColKernel &&) = default; 69 /** Set the input and output of the kernel. 70 * 71 * @param[in] input The input tensor to convert. 3 lower dimensions represent a single input [width, height, IFM], 72 * while every optional dimension from 4 and above represent a batch of inputs. Data types supported: QASYMM8/QASYMM8_SIGNED/F16/F32 73 * @param[out] output The output tensor. First 2 lower dimensions represent a transform of each 3D input, 74 * while every dimension above represents a batch. Data types supported: Same as @p input 75 * @param[in] kernel_dims The kernel dimensions (width and height). 76 * @param[in] conv_info Contains padding and stride information described in @ref PadStrideInfo. 77 * @param[in] has_bias In case biases are provided expands the matrix with 1. 78 * This is valid only for non-quantized inputs. 79 * @param[in] dilation (Optional) Dilation, in elements, across x and y. Defaults to (1, 1). 80 * @param[in] num_groups (Optional) Number of groups when performing a grouped convolution. 81 * Number of groups other than 1 is only supported for NCHW data layout. 82 * Number of groups should be multiple to the number of channels. 83 */ 84 void configure(const ICLTensor *input, ICLTensor *output, const Size2D &kernel_dims, const PadStrideInfo &conv_info, bool has_bias, const Size2D &dilation = Size2D(1U, 1U), 85 unsigned int num_groups = 1); 86 /** Set the input and output of the kernel. 87 * 88 * @param[in] compile_context The compile context to be used. 89 * @param[in] input The input tensor to convert. 3 lower dimensions represent a single input [width, height, IFM], 90 * while every optional dimension from 4 and above represent a batch of inputs. Data types supported: QASYMM8/QASYMM8_SIGNED/F16/F32 91 * @param[out] output The output tensor. First 2 lower dimensions represent a transform of each 3D input, 92 * while every dimension above represents a batch. Data types supported: Same as @p input 93 * @param[in] kernel_dims The kernel dimensions (width and height). 94 * @param[in] conv_info Contains padding and stride information described in @ref PadStrideInfo. 95 * @param[in] has_bias In case biases are provided expands the matrix with 1. 96 * @param[in] dilation (Optional) Dilation, in elements, across x and y. Defaults to (1, 1). 97 * @param[in] num_groups (Optional) Number of groups when performing a grouped convolution. num_groups != 1 is only supported for NCHW data layout 98 */ 99 void configure(const CLCompileContext &compile_context, const ICLTensor *input, ICLTensor *output, const Size2D &kernel_dims, const PadStrideInfo &conv_info, bool has_bias, 100 const Size2D &dilation = Size2D(1U, 1U), 101 unsigned int num_groups = 1); 102 /** Static function to check if given info will lead to a valid configuration of @ref CLIm2ColKernel 103 * 104 * @param[in] input The input tensor to convert. 3 lower dimensions represent a single input [width, height, IFM], 105 * while every optional dimension from 4 and above represent a batch of inputs. Data types supported: QASYMM8/QASYMM8_SIGNED/F16/F32 106 * @param[in] output The output tensor. First 2 lower dimensions represent a transform of each 3D input, 107 * while every dimension above represents a batch. Data types supported: Same as @p input 108 * @param[in] kernel_dims The kernel dimensions (width and height). 109 * @param[in] conv_info Contains padding and stride information described in @ref PadStrideInfo. 110 * @param[in] has_bias In case biases are provided expands the matrix with 1. 111 * This is valid only for non-quantized inputs. 112 * @param[in] dilation (Optional) Dilation, in elements, across x and y. Defaults to (1, 1). 113 * @param[in] num_groups (Optional) Number of groups when performing a grouped convolution. 114 * Number of groups other than 1 is only supported for NCHW data layout. 115 * Number of groups should be multiple to the number of channels. 116 * 117 * @return a status 118 */ 119 static Status validate(const ITensorInfo *input, const ITensorInfo *output, const Size2D &kernel_dims, const PadStrideInfo &conv_info, bool has_bias, const Size2D &dilation = Size2D(1U, 1U), 120 unsigned int num_groups = 1); 121 122 // Inherited methods overridden: 123 void run(const Window &window, cl::CommandQueue &queue) override; 124 125 public: 126 const ICLTensor *_input; 127 ICLTensor *_output; 128 DataLayout _data_layout; 129 std::pair<unsigned int, unsigned int> _convolved_dims; 130 unsigned int _num_elems_processed_per_iteration; 131 Size2D _kernel_dims; 132 PadStrideInfo _conv_info; 133 unsigned int _num_groups; 134 }; 135 } // namespace arm_compute 136 #endif /*ARM_COMPUTE_CLIM2COLKERNEL_H */ 137