Giorgio Arena | 9fe4144 | 2017-08-23 16:36:24 +0100 | [diff] [blame] | 1 | /* |
Georgios Pinitas | de5a1cc | 2018-02-02 12:52:07 +0000 | [diff] [blame] | 2 | * Copyright (c) 2017-2018 ARM Limited. |
Giorgio Arena | 9fe4144 | 2017-08-23 16:36:24 +0100 | [diff] [blame] | 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_CLDEPTHWISEIM2COLKERNEL_H__ |
| 25 | #define __ARM_COMPUTE_CLDEPTHWISEIM2COLKERNEL_H__ |
| 26 | |
| 27 | #include "arm_compute/core/CL/ICLKernel.h" |
| 28 | #include "arm_compute/core/Size2D.h" |
| 29 | |
| 30 | namespace arm_compute |
| 31 | { |
| 32 | class ICLTensor; |
| 33 | |
| 34 | /** Interface for the depthwise im2col reshape kernel. |
| 35 | * This kernel reshape the input low 3 dimensions to a new 3D shape where the output's first dimension is |
Giorgio Arena | 7657224 | 2018-04-04 17:44:26 +0100 | [diff] [blame] | 36 | * the linear patch size (FILTER_WIDTH * FILTER_HEIGHT) and second dimension is number of patches per image and third dimension unchanged . |
Giorgio Arena | 9fe4144 | 2017-08-23 16:36:24 +0100 | [diff] [blame] | 37 | **/ |
| 38 | class CLDepthwiseIm2ColKernel : public ICLKernel |
| 39 | { |
| 40 | public: |
| 41 | /** Default constructor */ |
| 42 | CLDepthwiseIm2ColKernel(); |
| 43 | /** Prevent instances of this class from being copied (As this class contains pointers) */ |
| 44 | CLDepthwiseIm2ColKernel(const CLDepthwiseIm2ColKernel &) = delete; |
| 45 | /** Prevent instances of this class from being copied (As this class contains pointers) */ |
| 46 | CLDepthwiseIm2ColKernel &operator=(const CLDepthwiseIm2ColKernel &) = delete; |
| 47 | /** Allow instances of this class to be moved */ |
| 48 | CLDepthwiseIm2ColKernel(CLDepthwiseIm2ColKernel &&) = default; |
| 49 | /** Allow instances of this class to be moved */ |
| 50 | CLDepthwiseIm2ColKernel &operator=(CLDepthwiseIm2ColKernel &&) = default; |
| 51 | /** Set the input and output of the kernel. |
| 52 | * |
Giorgio Arena | 7657224 | 2018-04-04 17:44:26 +0100 | [diff] [blame] | 53 | * @param[in] input The input tensor to convert. 3 lower dimensions represent a single input [width, height, IFM], |
Michele Di Giorgio | d24af8a | 2018-05-08 17:23:52 +0100 | [diff] [blame] | 54 | * while every optional dimension from 4 and above represent a batch of inputs. Data types supported: QASYMM8/F16/F32 |
Giorgio Arena | 7657224 | 2018-04-04 17:44:26 +0100 | [diff] [blame] | 55 | * @param[out] output The output tensor. First 3 lower dimensions represent a transform of each 3D input, |
| 56 | * while every dimension above 3 represents a batch. Data types supported: Same as @p input |
| 57 | * @param[in] kernel_dims The kernel dimensions (width and height). |
| 58 | * @param[in] conv_info Contains padding and stride information described in @ref PadStrideInfo. |
| 59 | * @param[in] has_bias Boolean that specifies if the depthwise convolution has bias. |
| 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. |
Giorgio Arena | 9fe4144 | 2017-08-23 16:36:24 +0100 | [diff] [blame] | 61 | */ |
Giorgio Arena | 7657224 | 2018-04-04 17:44:26 +0100 | [diff] [blame] | 62 | void configure(const ICLTensor *input, ICLTensor *output, const Size2D &kernel_dims, const PadStrideInfo &conv_info, bool has_bias = false, unsigned int depth_multiplier = 1); |
Giorgio Arena | ad0c738 | 2018-04-23 16:16:21 +0100 | [diff] [blame] | 63 | /** Static function to check if given info will lead to a valid configuration of @ref CLDepthwiseIm2ColKernel |
| 64 | * |
| 65 | * @param[in] input The input tensor to convert. 3 lower dimensions represent a single input [width, height, IFM], |
| 66 | * while every optional dimension from 4 and above represent a batch of inputs. Data types supported: QASYMM8/F32 |
| 67 | * @param[in] output The output tensor. First 3 lower dimensions represent a transform of each 3D input, |
| 68 | * while every dimension above 3 represents a batch. Data types supported: Same as @p input |
| 69 | * @param[in] kernel_dims The kernel dimensions (width and height). |
| 70 | * @param[in] conv_info Contains padding and stride information described in @ref PadStrideInfo. |
| 71 | * @param[in] has_bias Boolean that specifies if the depthwise convolution has bias. |
| 72 | * @param[in] depth_multiplier (Optional) Multiplier to apply to the input's depth in order to retrieve the output's depth. Defaults to 1. |
| 73 | * |
| 74 | * @return a status |
| 75 | */ |
| 76 | static Status validate(const ITensorInfo *input, const ITensorInfo *output, const Size2D &kernel_dims, const PadStrideInfo &conv_info, bool has_bias, unsigned int depth_multiplier); |
Giorgio Arena | 9fe4144 | 2017-08-23 16:36:24 +0100 | [diff] [blame] | 77 | |
| 78 | // Inherited methods overridden: |
| 79 | void run(const Window &window, cl::CommandQueue &queue) override; |
| 80 | |
| 81 | private: |
| 82 | const ICLTensor *_input; |
| 83 | ICLTensor *_output; |
| 84 | }; |
| 85 | } // arm_compute |
| 86 | #endif /*__ARM_COMPUTE_CLDEPTHWISEIM2COLKERNEL_H__ */ |