Michele Di Giorgio | d556d7b | 2020-10-27 10:56:31 +0000 | [diff] [blame] | 1 | /* |
ramelg01 | c827e99 | 2022-04-08 03:52:28 +0100 | [diff] [blame] | 2 | * Copyright (c) 2021-2022 Arm Limited. |
Michele Di Giorgio | d556d7b | 2020-10-27 10:56:31 +0000 | [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 | |
| 25 | #pragma once |
| 26 | |
| 27 | #include "arm_gemm_local.hpp" |
| 28 | #include "pool_common.hpp" |
| 29 | |
Michele Di Giorgio | d556d7b | 2020-10-27 10:56:31 +0000 | [diff] [blame] | 30 | namespace arm_conv |
| 31 | { |
| 32 | namespace pooling |
| 33 | { |
| 34 | struct PoolingConfig |
| 35 | { |
| 36 | PoolingMethod method = PoolingMethod::DEFAULT; |
| 37 | std::string filter = ""; |
| 38 | |
| 39 | PoolingConfig(PoolingMethod method) |
| 40 | : method(method) {}; |
| 41 | PoolingConfig() {}; |
| 42 | }; |
| 43 | |
| 44 | struct PoolingArgs |
| 45 | { |
| 46 | const CPUInfo *cpu_info; |
| 47 | |
| 48 | PoolingType pool_type; |
| 49 | PoolingWindow pool_window; |
| 50 | PoolingStride pool_stride; |
| 51 | bool exclude_padding; |
| 52 | |
| 53 | unsigned int n_batches, input_rows, input_cols, n_channels; |
| 54 | unsigned int output_rows, output_cols; |
| 55 | |
| 56 | PaddingValues padding; |
| 57 | |
| 58 | const PoolingConfig *config; |
| 59 | |
| 60 | PoolingArgs( |
| 61 | const CPUInfo *cpu_info, |
| 62 | PoolingType pool_type, |
| 63 | const PoolingWindow &window, |
| 64 | const PoolingStride &stride, |
| 65 | bool exclude_padding, |
| 66 | unsigned int n_batches, |
| 67 | unsigned int input_rows, |
| 68 | unsigned int input_cols, |
| 69 | unsigned int n_channels, |
| 70 | unsigned int output_rows, |
| 71 | unsigned int output_cols, |
| 72 | const PaddingValues &padding, |
| 73 | const PoolingConfig *cfg) |
| 74 | : cpu_info(cpu_info), pool_type(pool_type), pool_window(window), pool_stride(stride), exclude_padding(exclude_padding), n_batches(n_batches), input_rows(input_rows), input_cols(input_cols), |
| 75 | n_channels(n_channels), output_rows(output_rows), output_cols(output_cols), padding(padding), config(cfg) |
| 76 | { |
| 77 | // If either of the pooling window dimensions are set to zero, meaning |
| 78 | // "pool everything", then replace with the corresponding input dimension. |
| 79 | if(pool_window.rows == 0) |
| 80 | { |
| 81 | pool_window.rows = input_rows; |
| 82 | } |
| 83 | if(pool_window.cols == 0) |
| 84 | { |
| 85 | pool_window.cols = input_cols; |
| 86 | } |
| 87 | } |
| 88 | }; |
| 89 | |
ramelg01 | c827e99 | 2022-04-08 03:52:28 +0100 | [diff] [blame] | 90 | struct Nothing |
| 91 | { |
| 92 | }; |
| 93 | |
Michele Di Giorgio | d556d7b | 2020-10-27 10:56:31 +0000 | [diff] [blame] | 94 | struct Requantize32 |
| 95 | { |
| 96 | int32_t input_offset = 0; |
| 97 | int32_t output_offset = 0; |
| 98 | |
| 99 | int32_t per_layer_left_shift = 0; |
| 100 | int32_t per_layer_right_shift = 0; |
| 101 | int32_t per_layer_mul = 0; |
| 102 | |
| 103 | Requantize32(int32_t input_offset, int32_t output_offset, |
| 104 | int32_t per_layer_left_shift, int32_t per_layer_right_shift, |
| 105 | int32_t per_layer_mul) |
| 106 | : input_offset(input_offset), output_offset(output_offset), per_layer_left_shift(per_layer_left_shift), per_layer_right_shift(per_layer_right_shift), per_layer_mul(per_layer_mul) |
| 107 | { |
| 108 | } |
| 109 | }; |
| 110 | |
ramelg01 | c827e99 | 2022-04-08 03:52:28 +0100 | [diff] [blame] | 111 | template <typename TInput, typename TOutput> |
| 112 | class PoolingCommon : public IPoolingCommon |
| 113 | { |
| 114 | protected: |
| 115 | const PoolingArgs m_args; |
| 116 | |
| 117 | public: |
| 118 | PoolingCommon(const PoolingArgs &args) |
| 119 | : m_args(args) |
| 120 | { |
| 121 | } |
| 122 | PoolingCommon(PoolingCommon &) = delete; |
| 123 | PoolingCommon &operator=(PoolingCommon &) = delete; |
| 124 | |
| 125 | size_t get_working_size(unsigned int, unsigned int) const override = 0; |
| 126 | size_t get_working_size(unsigned int n_threads) const override |
| 127 | { |
| 128 | return this->get_working_size(n_threads, m_args.n_channels); |
| 129 | } |
| 130 | |
| 131 | // Execute pooling over the specified area of memory. |
| 132 | void execute( |
| 133 | const void *const input, |
| 134 | void *const output, |
| 135 | void *working_space, |
| 136 | unsigned int thread_id, |
| 137 | unsigned int num_threads) const override |
| 138 | { |
| 139 | this->execute( |
| 140 | input, |
| 141 | m_args.n_channels, |
| 142 | m_args.n_channels * m_args.input_cols, |
| 143 | m_args.n_channels * m_args.input_cols * m_args.input_rows, |
| 144 | output, |
| 145 | m_args.n_channels, |
| 146 | m_args.n_channels * m_args.output_cols, |
| 147 | m_args.n_channels * m_args.output_cols * m_args.output_rows, |
| 148 | working_space, |
| 149 | thread_id, num_threads); |
| 150 | } |
| 151 | |
| 152 | void execute( |
| 153 | const void *const input, |
| 154 | size_t ld_input_col, |
| 155 | size_t ld_input_row, |
| 156 | size_t ld_input_batch, |
| 157 | void *const output, |
| 158 | size_t ld_output_col, |
| 159 | size_t ld_output_row, |
| 160 | size_t ld_output_batch, |
| 161 | void *working_space, |
| 162 | unsigned int thread_id, |
| 163 | unsigned int num_threads) const override |
| 164 | { |
| 165 | this->execute( |
| 166 | m_args.n_batches, m_args.input_rows, m_args.input_cols, m_args.n_channels, |
| 167 | input, ld_input_col, ld_input_row, ld_input_batch, |
| 168 | m_args.padding, m_args.output_rows, m_args.output_cols, |
| 169 | output, ld_output_col, ld_output_row, ld_output_batch, |
| 170 | working_space, thread_id, num_threads); |
| 171 | } |
| 172 | |
| 173 | void execute( |
| 174 | unsigned int batches, |
| 175 | unsigned int height, |
| 176 | unsigned int width, |
| 177 | unsigned int channels, |
| 178 | const void *const input, |
| 179 | size_t ld_input_col, |
| 180 | size_t ld_input_row, |
| 181 | size_t ld_input_batch, |
| 182 | const PaddingValues &padding, |
| 183 | unsigned int output_height, |
| 184 | unsigned int output_width, |
| 185 | void *const output, |
| 186 | size_t ld_output_col, |
| 187 | size_t ld_output_row, |
| 188 | size_t ld_output_batch, |
| 189 | void *working_space, |
| 190 | unsigned int thread_id, |
| 191 | unsigned int num_threads) const override |
| 192 | { |
| 193 | this->execute_internal( |
| 194 | batches, height, width, channels, padding, |
| 195 | input, ld_input_col, ld_input_row, ld_input_batch, |
| 196 | output_height, output_width, |
| 197 | output, ld_output_col, ld_output_row, ld_output_batch, |
| 198 | working_space, thread_id, num_threads); |
| 199 | } |
| 200 | |
| 201 | protected: |
| 202 | virtual void execute_internal( |
| 203 | unsigned int batches, |
| 204 | unsigned int height, |
| 205 | unsigned int width, |
| 206 | unsigned int channels, |
| 207 | const PaddingValues &, |
| 208 | const void *const input, |
| 209 | size_t ld_input_col, |
| 210 | size_t ld_input_row, |
| 211 | size_t ld_input_batch, |
| 212 | unsigned int output_height, |
| 213 | unsigned int output_width, |
| 214 | void *const output, |
| 215 | size_t ld_output_col, |
| 216 | size_t ld_output_row, |
| 217 | size_t ld_output_batch, |
| 218 | void *working_space, |
| 219 | unsigned int thread_id, |
| 220 | unsigned int num_threads) const = 0; |
| 221 | }; |
| 222 | |
| 223 | template <typename TInput, typename TOutput> |
| 224 | using UniquePoolingCommon = std::unique_ptr<PoolingCommon<TInput, TOutput>>; |
Michele Di Giorgio | d556d7b | 2020-10-27 10:56:31 +0000 | [diff] [blame] | 225 | |
| 226 | // Get a pooling engine |
| 227 | template <typename TInput, typename TOutput = TInput, class OutputStage = Nothing> |
ramelg01 | c827e99 | 2022-04-08 03:52:28 +0100 | [diff] [blame] | 228 | UniquePoolingCommon<TInput, TOutput> pooling(const PoolingArgs &, const OutputStage & = {}); |
Michele Di Giorgio | d556d7b | 2020-10-27 10:56:31 +0000 | [diff] [blame] | 229 | |
| 230 | } // namespace pooling |
| 231 | } // namespace arm_conv |