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_ICLDEPTHWISECONVOLUTIONKERNEL3x3_H 25 #define ARM_COMPUTE_ICLDEPTHWISECONVOLUTIONKERNEL3x3_H 26 27 #include "src/core/CL/ICLKernel.h" 28 29 namespace arm_compute 30 { 31 class ICLTensor; 32 33 /** Interface for the kernel to run a 3x3 depthwise convolution on a tensor. 34 */ 35 class ICLDepthwiseConvolutionLayer3x3Kernel : public ICLKernel 36 { 37 public: 38 /** Default constructor */ ICLDepthwiseConvolutionLayer3x3Kernel()39 ICLDepthwiseConvolutionLayer3x3Kernel() 40 : _border_size(0), _input(), _output(), _weights(), _biases(), _conv_stride_y(1), _output_multipliers(), _output_shifts(), _is_quantized(false) 41 { 42 } 43 /** Prevent instances of this class from being copied (As this class contains pointers) */ 44 ICLDepthwiseConvolutionLayer3x3Kernel(const ICLDepthwiseConvolutionLayer3x3Kernel &) = delete; 45 /** Prevent instances of this class from being copied (As this class contains pointers) */ 46 ICLDepthwiseConvolutionLayer3x3Kernel &operator=(const ICLDepthwiseConvolutionLayer3x3Kernel &) = delete; 47 /** Default Move Constructor. */ 48 ICLDepthwiseConvolutionLayer3x3Kernel(ICLDepthwiseConvolutionLayer3x3Kernel &&) = default; 49 /** Default move assignment operator */ 50 ICLDepthwiseConvolutionLayer3x3Kernel &operator=(ICLDepthwiseConvolutionLayer3x3Kernel &&) = default; 51 /** Initialize the function's source, destination, conv and border_size. 52 * 53 * @param[in] input Source tensor. DataType supported: QASYMM8/F16/F32. 54 * @param[in] weights Weights tensor. A 3D tensor with dimensions [3, 3, IFM]. 55 * Data type supported: Same as @p input, QASYMM8/QSYMM8_PER_CHANNEL when input is QASYMM8. 56 * @param[in] biases Biases tensor. A 1D tensor with dimensions [IFM]. Must be nullptr if not needed. 57 * Data type supported: Same as @p input, S32 when input is QASYMM8. 58 * @param[out] output Destination tensor. Data type supported: Same as @p input. 59 * @param[in] conv_info Padding and stride information to use for the convolution. 60 * @param[in] depth_multiplier (Optional) Multiplier to apply to the input's depth in order to retrieve the output's depth. Defaults to 1. 61 * @param[in] act_info (Optional) Activation layer information in case of a fused activation. Only RELU, BOUNDED_RELU and LU_BOUNDED_RELU are supported for QASYMM8. 62 * @param[in] dilation (Optional) Dilation, in elements, across x and y. Defaults to (1, 1). 63 * @param[in] output_multipliers (Optional) Output multipliers tensor for quantized computations. In case of per-channel quantization, 64 * the number of multipliers must be equal to the number of filters (IFM). Supported data types: S32 65 * @param[in] output_shifts (Optional) Output shifts tensor for quantized computations. In case of per-channel quantization, 66 * the number of multipliers must be equal to the number of filters (IFM). Supported data types: S32 67 */ 68 virtual void configure(const ICLTensor *input, const ICLTensor *weights, const ICLTensor *biases, ICLTensor *output, const PadStrideInfo &conv_info, 69 unsigned int depth_multiplier = 1, ActivationLayerInfo act_info = ActivationLayerInfo(), const Size2D &dilation = Size2D(1U, 1U), 70 const ICLTensor *output_multipliers = nullptr, const ICLTensor *output_shifts = nullptr) = 0; 71 /** Initialize the function's source, destination, conv and border_size. 72 * 73 * @param[in] compile_context The compile context to be used. 74 * @param[in] input Source tensor. DataType supported: QASYMM8/F16/F32. 75 * @param[in] weights Weights tensor. A 3D tensor with dimensions [3, 3, IFM]. 76 * Data type supported: Same as @p input, QASYMM8/QSYMM8_PER_CHANNEL when input is QASYMM8. 77 * @param[in] biases Biases tensor. A 1D tensor with dimensions [IFM]. Must be nullptr if not needed. 78 * Data type supported: Same as @p input, S32 when input is QASYMM8. 79 * @param[out] output Destination tensor. Data type supported: Same as @p input. 80 * @param[in] conv_info Padding and stride information to use for the convolution. 81 * @param[in] depth_multiplier (Optional) Multiplier to apply to the input's depth in order to retrieve the output's depth. Defaults to 1. 82 * @param[in] act_info (Optional) Activation layer information in case of a fused activation. Only RELU, BOUNDED_RELU and LU_BOUNDED_RELU are supported for QASYMM8. 83 * @param[in] dilation (Optional) Dilation, in elements, across x and y. Defaults to (1, 1). 84 * @param[in] output_multipliers (Optional) Output multipliers tensor for quantized computations. In case of per-channel quantization, 85 * the number of multipliers must be equal to the number of filters (IFM). Supported data types: S32 86 * @param[in] output_shifts (Optional) Output shifts tensor for quantized computations. In case of per-channel quantization, 87 * the number of multipliers must be equal to the number of filters (IFM). Supported data types: S32 88 */ 89 virtual void configure(const CLCompileContext &compile_context, const ICLTensor *input, const ICLTensor *weights, const ICLTensor *biases, ICLTensor *output, const PadStrideInfo &conv_info, 90 unsigned int depth_multiplier = 1, ActivationLayerInfo act_info = ActivationLayerInfo(), const Size2D &dilation = Size2D(1U, 1U), 91 const ICLTensor *output_multipliers = nullptr, const ICLTensor *output_shifts = nullptr) = 0; 92 93 protected: 94 BorderSize _border_size; 95 const ICLTensor *_input; 96 ICLTensor *_output; 97 const ICLTensor *_weights; 98 const ICLTensor *_biases; 99 unsigned int _conv_stride_y; 100 const ICLTensor *_output_multipliers; 101 const ICLTensor *_output_shifts; 102 bool _is_quantized; 103 }; 104 } // namespace arm_compute 105 #endif /*ARM_COMPUTE_ICLDEPTHWISECONVOLUTIONKERNEL3x3_H */ 106