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_NEDIRECTCONVOLUTIONLAYER_H 25 #define ARM_COMPUTE_NEDIRECTCONVOLUTIONLAYER_H 26 27 #include "arm_compute/core/Types.h" 28 #include "arm_compute/runtime/IFunction.h" 29 #include "arm_compute/runtime/IMemoryManager.h" 30 #include "arm_compute/runtime/MemoryGroup.h" 31 #include "arm_compute/runtime/NEON/functions/NEActivationLayer.h" 32 #include "arm_compute/runtime/Tensor.h" 33 34 #include <memory> 35 36 namespace arm_compute 37 { 38 class NEDirectConvolutionLayerOutputStageKernel; 39 class NEDirectConvolutionLayerKernel; 40 class NEFillBorderKernel; 41 42 /** Function to run the direct convolution. 43 * 44 * This function calls the following NEON kernels: 45 * 46 * -# @ref NEFillBorderKernel for the input 47 * -# @ref NEDirectConvolutionLayerOutputStageKernel 48 * -# @ref NEDirectConvolutionLayerKernel 49 */ 50 class NEDirectConvolutionLayer : public IFunction 51 { 52 public: 53 /** Constructor */ 54 NEDirectConvolutionLayer(std::shared_ptr<IMemoryManager> memory_manager = nullptr); 55 /** Prevent instances of this class from being copied (As this class contains pointers) */ 56 NEDirectConvolutionLayer(const NEDirectConvolutionLayer &) = delete; 57 /** Prevent instances of this class from being copied (As this class contains pointers) */ 58 NEDirectConvolutionLayer &operator=(const NEDirectConvolutionLayer &) = delete; 59 /** Prevent instances of this class from being moved (As this class contains non movable objects) */ 60 NEDirectConvolutionLayer(NEDirectConvolutionLayer &&) = delete; 61 /** Prevent instances of this class from being moved (As this class contains non movable objects) */ 62 NEDirectConvolutionLayer &operator=(NEDirectConvolutionLayer &&) = delete; 63 /** Default destructor */ 64 ~NEDirectConvolutionLayer(); 65 /** Set the input, weights, biases and output tensors. 66 * 67 * @note: DirectConvolution only works in the following configurations: 68 * 1x1 convolution with stride_x = 1/2/3, stride_y = 1/2/3 data type = F16/F32 69 * 3x3 convolution with stride_x = 1/2/3, stride_y = 1/2/3 data type = F16/F32 70 * 5x5 convolution with stride_x = 1/2/3, stride_y = 1/2/3 data type = F32 71 * 72 * @param[in, out] input Input tensor. Data types supported: F16/F32. 73 * @param[in] weights Set of kernels to convolve the input volume. 74 * Supported sizes: 1x1, 3x3 and 5x5. 75 * The 3rd dimension must be the same as the input's volume 3rd dimension. 76 * Data type supported: Same as @p input. 77 * @param[in] bias Set of biases. Can be nullptr. Data type supported: Same as @p input. 78 * @param[out] output Output tensor. 79 * The 3rd dimensions must be equal to the 4th dimension of the @p kernels tensor. Data types supported: Same as @p input. 80 * @param[in] conv_info Contains padding and stride information described in @ref PadStrideInfo. 81 * @param[in] act_info (Optional) Activation layer information in case of a fused activation. 82 */ 83 void configure(ITensor *input, const ITensor *weights, const ITensor *bias, ITensor *output, const PadStrideInfo &conv_info, const ActivationLayerInfo &act_info = ActivationLayerInfo()); 84 /** Static function to check if given info will lead to a valid configuration of @ref NEDirectConvolutionLayer 85 * 86 * @note: DirectConvolution only works in the following configurations: 87 * 1x1 convolution with stride_x = 1/2/3, stride_y = 1/2/3 data type = F16/F32 88 * 3x3 convolution with stride_x = 1/2/3, stride_y = 1/2/3 data type = F16/F32 89 * 5x5 convolution with stride_x = 1/2/3, stride_y = 1/2/3 data type = F32 90 * 91 * @param[in] input Input tensor. Data types supported: F16/F32. 92 * @param[in] weights Set of kernels to convolve the input volume. 93 * Supported sizes: 1x1, 3x3 and 5x5. 94 * The 3rd dimension must be the same as the input's volume 3rd dimension. 95 * Data type supported: Same as @p input. 96 * @param[in] bias Set of biases. Can be nullptr. Data type supported: Same as @p input. 97 * @param[in] output Output tensor. 98 * The 3rd dimensions must be equal to the 4th dimension of the @p kernels tensor. Data types supported: Same as @p input. 99 * @param[in] conv_info Contains padding and stride information described in @ref PadStrideInfo. 100 * @param[in] act_info (Optional) Activation layer information in case of a fused activation. 101 * 102 * @return a status 103 */ 104 static Status validate(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *bias, const ITensorInfo *output, const PadStrideInfo &conv_info, 105 const ActivationLayerInfo &act_info = ActivationLayerInfo()); 106 107 // Inherited methods overridden: 108 void run() override; 109 110 private: 111 MemoryGroup _memory_group; 112 std::unique_ptr<NEDirectConvolutionLayerOutputStageKernel> _output_stage_kernel; 113 std::unique_ptr<NEDirectConvolutionLayerKernel> _conv_kernel; 114 std::unique_ptr<NEFillBorderKernel> _input_border_handler; 115 NEActivationLayer _activationlayer_function; 116 Tensor _accumulator; 117 bool _has_bias; 118 bool _is_activationlayer_enabled; 119 unsigned int _dim_split; 120 bool _is_padding_required; 121 }; 122 } 123 #endif /* ARM_COMPUTE_NEDIRECTCONVOLUTIONLAYER_H */ 124