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_NELOCALLYCONNECTEDLAYER_H 25 #define ARM_COMPUTE_NELOCALLYCONNECTEDLAYER_H 26 27 #include "arm_compute/runtime/IFunction.h" 28 29 #include "arm_compute/core/Types.h" 30 #include "arm_compute/runtime/IMemoryManager.h" 31 #include "arm_compute/runtime/MemoryGroup.h" 32 #include "arm_compute/runtime/NEON/functions/NECol2Im.h" 33 #include "arm_compute/runtime/NEON/functions/NEIm2Col.h" 34 #include "arm_compute/runtime/Tensor.h" 35 36 #include <memory> 37 38 namespace arm_compute 39 { 40 class INETensor; 41 class NEWeightsReshapeKernel; 42 class NELocallyConnectedMatrixMultiplyKernel; 43 44 /** Basic function to compute the locally connected layer. This function calls the following NEON kernels: 45 * 46 * -# @ref NEWeightsReshapeKernel (executed only once for each configuration) 47 * -# @ref NEIm2ColKernel 48 * -# @ref NELocallyConnectedMatrixMultiplyKernel 49 * -# @ref NECol2ImKernel 50 */ 51 class NELocallyConnectedLayer : public IFunction 52 { 53 public: 54 /** Default constructor */ 55 NELocallyConnectedLayer(std::shared_ptr<IMemoryManager> memory_manager = nullptr); 56 /** Prevent instances of this class from being copied (As this class contains pointers) */ 57 NELocallyConnectedLayer(const NELocallyConnectedLayer &) = delete; 58 /** Prevent instances of this class from being moved (As this class contains pointers) */ 59 NELocallyConnectedLayer(NELocallyConnectedLayer &&) = delete; 60 /** Prevent instances of this class from being copied (As this class contains pointers) */ 61 NELocallyConnectedLayer &operator=(const NELocallyConnectedLayer &) = delete; 62 /** Prevent instances of this class from being moved (As this class contains pointers) */ 63 NELocallyConnectedLayer &operator=(NELocallyConnectedLayer &&) = delete; 64 /** Default destructor */ 65 ~NELocallyConnectedLayer(); 66 /** Set the input and output tensors. 67 * 68 * @param[in] input Source tensor. 3 lower dimensions represent a single input [width, height, IFM], 69 * while every optional dimension from 4 and above represent a batch of inputs. 70 * Data types supported: F16, F32. 71 * @param[in] weights Weights tensor. Weights are 5D tensor with dimensions [kernel_x, kernel_y, IFM, OFM, num_patches]. Data type supported:Same as @p input. 72 * @param[in] biases Biases tensor. Shared biases supported. Biases are 2D tensor with dimensions [OFM, num_patches]. Data type supported:Same as @p input. 73 * @param[out] output Destination tensor. 3 lower dimensions represent a single output [width, height, OFM], while the rest represent batch of outputs. 74 * Data types supported: Same as @p input. 75 * @param[in] conv_info Contains padding and stride information described in @ref PadStrideInfo. 76 */ 77 ARM_COMPUTE_DEPRECATED_REL(20.11) 78 void configure(const ITensor *input, const ITensor *weights, const ITensor *biases, ITensor *output, const PadStrideInfo &conv_info); 79 /** Static function to check if given info will lead to a valid configuration of @ref NELocallyConnectedLayer 80 * 81 * @param[in] input Input tensor info. 3 lower dimensions represent a single input [width, height, IFM], 82 * while every optional dimension from 4 and above represent a batch of inputs. 83 * Data types supported: F16, F32. 84 * @param[in] weights Weights tensor info. Weights are 5D tensor with dimensions [kernel_x, kernel_y, IFM, OFM, num_patches]. Data type supported:Same as @p input. 85 * @param[in] biases Biases tensor info. Shared biases supported. Biases are 2D tensor with dimensions [OFM, num_patches]. Data type supported:Same as @p input. 86 * @param[in] output Output tensor info. 3 lower dimensions represent a single output [width, height, OFM], while the rest represent batch of outputs. 87 * Data types supported: Same as @p input. 88 * @param[in] conv_info Contains padding and stride information described in @ref PadStrideInfo. 89 * 90 * @return a status 91 */ 92 static Status validate(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *biases, const ITensorInfo *output, const PadStrideInfo &conv_info); 93 94 // Inherited methods overridden: 95 void run() override; 96 void prepare() override; 97 98 private: 99 MemoryGroup _memory_group; 100 NEIm2Col _input_im2col; 101 std::unique_ptr<NEWeightsReshapeKernel> _weights_reshape_kernel; 102 std::unique_ptr<NELocallyConnectedMatrixMultiplyKernel> _mm_kernel; 103 NECol2Im _output_col2im; 104 Tensor _input_im2col_reshaped; 105 Tensor _weights_reshaped; 106 Tensor _gemm_output; 107 bool _is_prepared; 108 const ITensor *_original_weights; 109 }; 110 } 111 #endif /* ARM_COMPUTE_NELOCALLYCONNECTEDLAYER_H */ 112