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_NEDIRECTCONVOLUTIONLAYERKERNEL_H 25 #define ARM_COMPUTE_NEDIRECTCONVOLUTIONLAYERKERNEL_H 26 27 #include "src/core/NEON/INEKernel.h" 28 29 namespace arm_compute 30 { 31 class ITensor; 32 33 /** NEON interface for Direct Convolution Layer kernel */ 34 class NEDirectConvolutionLayerKernel : public INEKernel 35 { 36 public: name()37 const char *name() const override 38 { 39 return "NEDirectConvolutionLayerKernel"; 40 } 41 /** Default constructor */ 42 NEDirectConvolutionLayerKernel(); 43 /** Prevent instances of this class from being copied (As this class contains pointers) */ 44 NEDirectConvolutionLayerKernel(const NEDirectConvolutionLayerKernel &) = delete; 45 /** Prevent instances of this class from being copied (As this class contains pointers) */ 46 NEDirectConvolutionLayerKernel &operator=(const NEDirectConvolutionLayerKernel &) = delete; 47 /** Allow instances of this class to be moved */ 48 NEDirectConvolutionLayerKernel(NEDirectConvolutionLayerKernel &&) = default; 49 /** Allow instances of this class to be moved */ 50 NEDirectConvolutionLayerKernel &operator=(NEDirectConvolutionLayerKernel &&) = default; 51 /** Default destructor */ 52 ~NEDirectConvolutionLayerKernel() = default; 53 /** Set the input, weights, and output tensors. 54 * 55 * @note: DirectConvolution only works in the following configurations: 56 * 1x1 convolution with stride_x = 1/2/3, stride_y = 1/2/3 57 * 3x3 convolution with stride_x = 1/2/3, stride_y = 1/2/3 58 * 59 * @param[in] input The input tensor to convolve. 3 lower dimensions represent a single input [width, height, IFM], 60 * while every optional dimension from 4 and above represent a batch of inputs. Data types supported: F16/F32. 61 * @param[in] weights Weights tensor. Weights are 4D tensor with dimensions [kernel_x, kernel_y, IFM, OFM]. 62 * The 3rd dimension must be the same as the input's volume 3rd dimension. 63 * Data type supported:Same as @p input. 64 * @param[out] output Output tensor. 65 * The 3rd dimensions must be equal to the 4th dimension of the @p kernels tensor. Data types supported: F16/F32 66 * @param[in] conv_info Contains padding and stride information described in @ref PadStrideInfo. 67 */ 68 void configure(const ITensor *input, const ITensor *weights, ITensor *output, const PadStrideInfo &conv_info); 69 /** Static function to check if given info will lead to a valid configuration of @ref NEDirectConvolutionLayerKernel 70 * 71 * @param[in] input The input tensor to convolve. 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: F16/F32. 73 * @param[in] weights Weights tensor. Weights are 4D tensor with dimensions [kernel_x, kernel_y, IFM, OFM]. 74 * The 3rd dimension must be the same as the input's volume 3rd dimension. 75 * Data type supported:Same as @p input. 76 * @param[in] output Output tensor. 77 * The 3rd dimensions must be equal to the 4th dimension of the @p kernels tensor. Data types supported: F16/F32 78 * @param[in] conv_info Contains padding and stride information described in @ref PadStrideInfo. 79 * 80 * @return a status 81 */ 82 static Status validate(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *output, const PadStrideInfo &conv_info); 83 84 // Inherited methods overridden: 85 void run(const Window &window, const ThreadInfo &info) override; 86 BorderSize border_size() const override; 87 88 private: 89 /* Template function for optimized convolution NHWC */ 90 template <typename T> 91 void convolve_nhwc_optimized(const Window &window); 92 93 /* Template function for convolution NHWC */ 94 template <typename T> 95 void convolve_nhwc(const Window &window); 96 97 const ITensor *_input; 98 const ITensor *_weights; 99 ITensor *_output; 100 PadStrideInfo _conv_info; 101 BorderSize _border_size; 102 unsigned int _kernel_size; 103 unsigned int _num_weight_elems_read_per_row; 104 unsigned int _num_elems_read_per_iteration; 105 unsigned int _num_elems_written_per_iteration; 106 }; 107 } // namespace arm_compute 108 #endif /*ARM_COMPUTE_NEDIRECTCONVOLUTIONLAYERKERNEL_H */ 109