Gunes Bayir | 7dc0234 | 2022-11-21 21:46:50 +0000 | [diff] [blame^] | 1 | /* |
| 2 | * Copyright (c) 2022 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 SRC_DYNAMIC_FUSION_SKETCH_GPU_COMPONENTS_CL_CLCOMPONENTDEPTHWISECONV2D |
| 25 | #define SRC_DYNAMIC_FUSION_SKETCH_GPU_COMPONENTS_CL_CLCOMPONENTDEPTHWISECONV2D |
| 26 | |
| 27 | #include "arm_compute/core/Error.h" |
| 28 | #include "src/dynamic_fusion/sketch/gpu/components/IGpuKernelComponent.h" |
| 29 | #include <memory> |
| 30 | |
| 31 | namespace arm_compute |
| 32 | { |
| 33 | /** Forward declaration */ |
| 34 | class ITensorInfo; |
| 35 | namespace experimental |
| 36 | { |
| 37 | namespace dynamic_fusion |
| 38 | { |
| 39 | /** Forward declaration */ |
| 40 | template <typename T> |
| 41 | class ArgumentPack; |
| 42 | class DepthwiseConv2dAttributes; |
| 43 | |
| 44 | /** Component specific settings |
| 45 | */ |
| 46 | class ClComponentDepthwiseConv2dSettings |
| 47 | { |
| 48 | public: |
| 49 | /** Set export_input_to_cl_image flag */ |
| 50 | ClComponentDepthwiseConv2dSettings &export_input_to_cl_image(bool cl_image); |
| 51 | /** Get export_input_to_cl_image flag */ |
| 52 | bool export_input_to_cl_image() const; |
| 53 | |
| 54 | /** Set export_weights_to_cl_image flag */ |
| 55 | ClComponentDepthwiseConv2dSettings &export_weights_to_cl_image(bool cl_image); |
| 56 | /** Get export_weights_to_cl_image flag */ |
| 57 | bool export_weights_to_cl_image() const; |
| 58 | |
| 59 | /** Set fast_relaxed_math flag */ |
| 60 | ClComponentDepthwiseConv2dSettings &fast_relaxed_math(bool fast_relaxed_math); |
| 61 | /** Get fast_relaxed_math flag */ |
| 62 | bool fast_relaxed_math() const; |
| 63 | |
| 64 | /** Set is_fma_available flag */ |
| 65 | ClComponentDepthwiseConv2dSettings &is_fma_available(bool is_fma_available); |
| 66 | /** Get is_fma_available flag */ |
| 67 | bool is_fma_available() const; |
| 68 | |
| 69 | /** Set N0: number of columns processed by each thread */ |
| 70 | ClComponentDepthwiseConv2dSettings &n0(unsigned int n0); |
| 71 | /** Get N0: number of columns processed by each thread */ |
| 72 | unsigned int n0() const; |
| 73 | |
| 74 | /** Set M0: number of rows processed by each thread */ |
| 75 | ClComponentDepthwiseConv2dSettings &m0(unsigned int m0); |
| 76 | /** Set M0: number of rows processed by each thread */ |
| 77 | unsigned int m0() const; |
| 78 | |
| 79 | private: |
| 80 | bool _export_input_to_cl_image{ false }; /**< Export input to cl_image */ |
| 81 | bool _export_weights_to_cl_image{ false }; /**< Export the weights to cl_image */ |
| 82 | bool _fast_relaxed_math{ true }; /**< Enable/disable -cl-fast-relaxed-math flag */ |
| 83 | bool _is_fma_available{ false }; /**< Is fma instruction available */ |
| 84 | unsigned int _n0{ 0 }; /**< Number of columns processed by each thread */ |
| 85 | unsigned int _m0{ 0 }; /**< Number of rows processed by each thread */ |
| 86 | }; |
| 87 | |
| 88 | /** Forward declaration */ |
| 89 | class ClTemplateDepthwiseConv2d; |
| 90 | |
| 91 | class ClComponentDepthwiseConv2d final : public IGpuKernelComponent |
| 92 | { |
| 93 | public: |
| 94 | /** Attributes are a set of backend-agnostic parameters that define what a component does */ |
| 95 | using Attributes = DepthwiseConv2dAttributes; |
| 96 | /** Settings are a set of backend-specific parameters that influence the implementation of a component */ |
| 97 | using Settings = ClComponentDepthwiseConv2dSettings; |
| 98 | |
| 99 | public: |
| 100 | /** Validate the component |
| 101 | * |
| 102 | * @param[in] properties Component properties @ref Properties |
| 103 | * @param[in,out] tensors Tensor arguments to the component |
| 104 | * @param[in] attributes Component attributes @ref Attributes |
| 105 | * @param[in] settings Component settings @ref Settings |
| 106 | * |
| 107 | * @return Status Validation results |
| 108 | * |
| 109 | * Tensor argument names: |
| 110 | * - ACL_SRC_0: Input |
| 111 | * - ACL_SRC_1: Weight |
| 112 | * - ACL_SRC_2: Bias (Optional) |
| 113 | * - ACL_DST_0: Output |
| 114 | * |
| 115 | * Tensor argument constness: |
| 116 | * - ACL_SRC_0: Const |
| 117 | * - ACL_SRC_1: Const |
| 118 | * - ACL_SRC_2: Const |
| 119 | * - ACL_DST_0: Const |
| 120 | * |
| 121 | * Valid data layouts: |
| 122 | * - NHWC |
| 123 | * |
| 124 | * Valid data type configurations: |
| 125 | * |ACL_SRC_0 |ACL_SRC_1 |ACL_SRC_2 |ACL_DST_0 | |
| 126 | * |:--------------|:--------------|:--------------|:--------------| |
| 127 | * |F16 |F16 |F16 |F16 | |
| 128 | * |F32 |F32 |F32 |F32 | |
| 129 | */ |
| 130 | static Status validate( |
| 131 | const Properties &properties, |
| 132 | const ArgumentPack<ITensorInfo> &tensors, |
| 133 | const Attributes &attributes, |
| 134 | const Settings &settings); |
| 135 | |
| 136 | /** Constructor |
| 137 | * |
| 138 | * Similar to @ref ClComponentDepthwiseConv2d::validate() |
| 139 | */ |
| 140 | ClComponentDepthwiseConv2d( |
| 141 | ComponentId id, |
| 142 | const Properties &properties, |
| 143 | const ArgumentPack<ITensorInfo> &tensors, |
| 144 | const Attributes &attributes, |
| 145 | const Settings &settings); |
| 146 | |
| 147 | /** Destructor */ |
| 148 | ~ClComponentDepthwiseConv2d() override; |
| 149 | /** Prevent instances of this class from being copy constructed */ |
| 150 | ClComponentDepthwiseConv2d(const ClComponentDepthwiseConv2d &component) = delete; |
| 151 | /** Prevent instances of this class from being copied */ |
| 152 | ClComponentDepthwiseConv2d &operator=(const ClComponentDepthwiseConv2d &component) = delete; |
| 153 | /** Allow instances of this class to be move constructed */ |
| 154 | ClComponentDepthwiseConv2d(ClComponentDepthwiseConv2d &&component) = default; |
| 155 | /** Allow instances of this class to be moved */ |
| 156 | ClComponentDepthwiseConv2d &operator=(ClComponentDepthwiseConv2d &&component) = default; |
| 157 | /** Get template writer for the component */ |
| 158 | const IGpuTemplateComponentWriter *template_writer() const override; |
| 159 | /** Get component type */ |
| 160 | GpuComponentType type() const override |
| 161 | { |
| 162 | return GpuComponentType::Complex; |
| 163 | } |
| 164 | |
| 165 | private: |
| 166 | std::unique_ptr<ClTemplateDepthwiseConv2d> _component_writer; |
| 167 | }; |
| 168 | } // namespace dynamic_fusion |
| 169 | } // namespace experimental |
| 170 | } // namespace arm_compute |
| 171 | #endif /* SRC_DYNAMIC_FUSION_SKETCH_GPU_COMPONENTS_CL_CLCOMPONENTDEPTHWISECONV2D */ |