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_CLDIRECTCONVOLUTIONLAYERKERNEL_H 25 #define ARM_COMPUTE_CLDIRECTCONVOLUTIONLAYERKERNEL_H 26 27 #include "arm_compute/core/Types.h" 28 #include "src/core/CL/ICLKernel.h" 29 30 namespace arm_compute 31 { 32 class ICLTensor; 33 34 /** Interface for the direct convolution kernel. 35 */ 36 class CLDirectConvolutionLayerKernel : public ICLKernel 37 { 38 public: 39 /** Default constructor */ 40 CLDirectConvolutionLayerKernel(); 41 /** Prevent instances of this class from being copied (As this class contains pointers) */ 42 CLDirectConvolutionLayerKernel(const CLDirectConvolutionLayerKernel &) = delete; 43 /** Prevent instances of this class from being copied (As this class contains pointers) */ 44 CLDirectConvolutionLayerKernel &operator=(const CLDirectConvolutionLayerKernel &) = delete; 45 /** Allow instances of this class to be moved */ 46 CLDirectConvolutionLayerKernel(CLDirectConvolutionLayerKernel &&) = default; 47 /** Allow instances of this class to be moved */ 48 CLDirectConvolutionLayerKernel &operator=(CLDirectConvolutionLayerKernel &&) = default; 49 /** Default destructor */ 50 ~CLDirectConvolutionLayerKernel() = default; 51 /** Set the input, weights, biases and output tensors. 52 * 53 * @note: DirectConvolution only works in the following configurations: 54 * 1x1 convolution with stride_x = 1/2/3, stride_y = 1/2/3 55 * 3x3 convolution with stride_x = 1/2, stride_y = 1/2 56 * 5x5 convolution with stride_x = 1/2, stride_y = 1/2 57 * 9x9 convolution with stride_x = 1/2, stride_y = 1/2 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: QASYMM8_SIGNED/QASYMM8/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[in] biases Biases tensor. Biases are 1D tensor with dimension [OFM]. 65 * Data type supported: Should match @p input data type, except for input of QASYMM8 and QASYMM8_SIGNED type where biases should be of S32 type 66 * @param[out] output Output tensor. 67 * The 3rd dimensions must be equal to the 4th dimension of the @p kernels tensor. Data types supported: Same as @p input. 68 * @param[in] conv_info Contains padding and stride information described in @ref PadStrideInfo. 69 */ 70 void configure(const ICLTensor *input, const ICLTensor *weights, const ICLTensor *biases, ICLTensor *output, const PadStrideInfo &conv_info); 71 /** Set the input, weights, biases and output tensors. 72 * 73 * @note: DirectConvolution only works in the following configurations: 74 * 1x1 convolution with stride_x = 1/2/3, stride_y = 1/2/3 75 * 3x3 convolution with stride_x = 1/2, stride_y = 1/2 76 * 5x5 convolution with stride_x = 1/2, stride_y = 1/2 77 * 9x9 convolution with stride_x = 1/2, stride_y = 1/2 78 * 79 * @param[in] compile_context The compile context to be used. 80 * @param[in] input The input tensor to convolve. 3 lower dimensions represent a single input [width, height, IFM], 81 * while every optional dimension from 4 and above represent a batch of inputs. Data types supported: QASYMM8_SIGNED/QASYMM8/F16/F32. 82 * @param[in] weights Weights tensor. Weights are 4D tensor with dimensions [kernel_x, kernel_y, IFM, OFM]. 83 * The 3rd dimension must be the same as the input's volume 3rd dimension. 84 * Data type supported:Same as @p input. 85 * @param[in] biases Biases tensor. Biases are 1D tensor with dimension [OFM]. 86 * Data type supported: Should match @p input data type, except for input of QASYMM8 and QASYMM8_SIGNED type where biases should be of S32 type 87 * @param[out] output Output tensor. 88 * The 3rd dimensions must be equal to the 4th dimension of the @p kernels tensor. Data types supported: Same as @p input. 89 * @param[in] conv_info Contains padding and stride information described in @ref PadStrideInfo. 90 */ 91 void configure(const CLCompileContext &compile_context, const ICLTensor *input, const ICLTensor *weights, const ICLTensor *biases, ICLTensor *output, const PadStrideInfo &conv_info); 92 /** Static function to check if given info will lead to a valid configuration of @ref CLDirectConvolutionLayerKernel 93 * 94 * @param[in] input The input tensor to convolve. 3 lower dimensions represent a single input [width, height, IFM], 95 * while every optional dimension from 4 and above represent a batch of inputs. Data types supported: QASYMM8_SIGNED/QASYMM8/F16/F32. 96 * @param[in] weights Weights tensor. Weights are 4D tensor with dimensions [kernel_x, kernel_y, IFM, OFM]. 97 * The 3rd dimension must be the same as the input's volume 3rd dimension. 98 * Data type supported:Same as @p input. 99 * @param[in] biases Biases tensor. Biases are 1D tensor with dimension [OFM]. 100 * Data type supported: Should match @p input data type, except for input of QASYMM8 and QASYMM8_SIGNED type where biases should be of S32 type. 101 * @param[in] output Output tensor. 102 * The 3rd dimensions must be equal to the 4th dimension of the @p kernels tensor. Data types supported: Same as @p input. 103 * @param[in] conv_info Contains padding and stride information described in @ref PadStrideInfo. 104 * @param[in] target Target GPU architecture. 105 * 106 * @return a status 107 */ 108 static Status validate(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *biases, const ITensorInfo *output, const PadStrideInfo &conv_info, const GPUTarget target); 109 110 // Inherited methods overridden: 111 void run(const Window &window, cl::CommandQueue &queue) override; 112 BorderSize border_size() const override; 113 114 public: 115 const ICLTensor *_input; 116 const ICLTensor *_biases; 117 const ICLTensor *_weights; 118 ICLTensor *_output; 119 DataLayout _data_layout; 120 BorderSize _border_size; 121 int _conv_stride_x; 122 int _conv_stride_y; 123 }; 124 } // namespace arm_compute 125 #endif /*ARM_COMPUTE_CLDIRECTCONVOLUTIONLAYERKERNEL_H */ 126