Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 1 | /* |
Michele Di Giorgio | d9eaf61 | 2020-07-08 11:12:57 +0100 | [diff] [blame] | 2 | * Copyright (c) 2018-2020 Arm Limited. |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +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 | */ |
Michalis Spyrou | f464337 | 2019-11-29 16:17:13 +0000 | [diff] [blame] | 24 | #ifndef ARM_COMPUTE_GRAPH_BACKENDS_DETAIL_FUNCTION_HELPERS_H |
| 25 | #define ARM_COMPUTE_GRAPH_BACKENDS_DETAIL_FUNCTION_HELPERS_H |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 26 | |
| 27 | #include "arm_compute/graph/Logger.h" |
| 28 | #include "arm_compute/graph/Tensor.h" |
| 29 | #include "arm_compute/graph/TypePrinter.h" |
| 30 | #include "arm_compute/graph/Types.h" |
Georgios Pinitas | 9e4824c | 2019-04-12 13:15:58 +0100 | [diff] [blame] | 31 | #include "arm_compute/graph/Utils.h" |
giuros01 | acce504 | 2019-02-21 17:32:34 +0000 | [diff] [blame] | 32 | #include "arm_compute/graph/backends/FusedConvolutionBatchNormalizationFunction.h" |
Manuel Bottini | bffb41e | 2019-06-20 16:00:27 +0100 | [diff] [blame] | 33 | #include "arm_compute/graph/backends/FusedDepthwiseConvolutionBatchNormalizationFunction.h" |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 34 | #include "arm_compute/graph/backends/Utils.h" |
| 35 | #include "arm_compute/graph/nodes/Nodes.h" |
| 36 | |
| 37 | #include "arm_compute/core/Error.h" |
| 38 | #include "arm_compute/core/Helpers.h" |
| 39 | #include "arm_compute/core/ITensorInfo.h" |
Sang-Hoon Park | 68dd25f | 2020-10-19 16:00:11 +0100 | [diff] [blame] | 40 | #include "support/Cast.h" |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 41 | |
| 42 | namespace arm_compute |
| 43 | { |
| 44 | namespace graph |
| 45 | { |
| 46 | namespace backends |
| 47 | { |
| 48 | namespace detail |
| 49 | { |
Georgios Pinitas | 0b192e8 | 2020-02-20 17:09:28 +0000 | [diff] [blame] | 50 | // Address rule DR-9R5 (1579. Return by converting move constructor) |
| 51 | #if defined(__clang__) || (defined(__GNUC__) && (__GNUC__ >= 5)) |
| 52 | #define RETURN_UNIQUE_PTR(x) (x) |
| 53 | #else /* defined(__clang__) || (defined(__GNUC__) && (__GNUC__ >= 5)) */ |
| 54 | #define RETURN_UNIQUE_PTR(x) (std::move(x)) |
| 55 | #endif /* defined(__clang__) || (defined(__GNUC__) && (__GNUC__ >= 5)) */ |
| 56 | |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 57 | /** Returns backing tensor of a given tensor |
| 58 | * |
| 59 | * @tparam TargetInfo Target information |
| 60 | * |
| 61 | * @param[in] tensor Tensor to extract the backing tensor from |
| 62 | * |
| 63 | * @return Backing tensor if present else nullptr |
| 64 | */ |
| 65 | template <typename TargetInfo> |
| 66 | typename TargetInfo::TensorType *get_backing_tensor(arm_compute::graph::Tensor *tensor) |
| 67 | { |
| 68 | typename TargetInfo::TensorType *backing_tensor = nullptr; |
| 69 | if(tensor != nullptr) |
| 70 | { |
| 71 | ARM_COMPUTE_ERROR_ON(tensor->desc().target != TargetInfo::TargetType); |
| 72 | // Get backing tensor handle |
| 73 | ITensorHandle *tensor_handle = tensor->handle(); |
| 74 | // Get backing tensor |
| 75 | backing_tensor = (tensor_handle != nullptr) ? arm_compute::utils::cast::polymorphic_cast<typename TargetInfo::TensorType *>(&tensor_handle->tensor()) : nullptr; |
| 76 | } |
| 77 | |
| 78 | return backing_tensor; |
| 79 | } |
| 80 | |
| 81 | template <typename TargetInfo> |
| 82 | void validate_node(const INode &node, size_t num_expected_inputs, size_t num_expected_outputs) |
| 83 | { |
| 84 | ARM_COMPUTE_LOG_GRAPH_VERBOSE("Creating " << node.type() |
Pablo Tello | 3252143 | 2018-11-15 14:43:10 +0000 | [diff] [blame] | 85 | << " Target: " << TargetInfo::TargetType |
| 86 | << " ID: " << node.id() |
| 87 | << node.name() |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 88 | << std::endl); |
| 89 | |
| 90 | ARM_COMPUTE_ERROR_ON(TargetInfo::TargetType != node.assigned_target()); |
| 91 | ARM_COMPUTE_ERROR_ON(node.num_inputs() != num_expected_inputs); |
| 92 | ARM_COMPUTE_ERROR_ON(node.num_outputs() != num_expected_outputs); |
Michalis Spyrou | 6bff195 | 2019-10-02 17:22:11 +0100 | [diff] [blame] | 93 | ARM_COMPUTE_UNUSED(node, num_expected_inputs, num_expected_outputs); |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 94 | } |
| 95 | |
| 96 | /** Creates a backend activation layer function |
| 97 | * |
| 98 | * @tparam ActivationLayerFunction Backend activation function |
| 99 | * @tparam TargetInfo Target-specific information |
| 100 | * |
| 101 | * @param[in] node Node to create the backend function for |
| 102 | * |
| 103 | * @return Backend activation layer function |
| 104 | */ |
| 105 | template <typename ActivationLayerFunction, typename TargetInfo> |
| 106 | std::unique_ptr<IFunction> create_activation_layer(ActivationLayerNode &node) |
| 107 | { |
| 108 | validate_node<TargetInfo>(node, 1 /* expected inputs */, 1 /* expected outputs */); |
| 109 | |
| 110 | // Extract IO and info |
| 111 | typename TargetInfo::TensorType *input = get_backing_tensor<TargetInfo>(node.input(0)); |
| 112 | typename TargetInfo::TensorType *output = get_backing_tensor<TargetInfo>(node.output(0)); |
| 113 | const ActivationLayerInfo act_info = node.activation_info(); |
| 114 | |
| 115 | // Create function |
| 116 | auto func = support::cpp14::make_unique<ActivationLayerFunction>(); |
| 117 | func->configure(input, output, act_info); |
| 118 | |
Pablo Tello | 3252143 | 2018-11-15 14:43:10 +0000 | [diff] [blame] | 119 | ARM_COMPUTE_LOG_GRAPH_INFO("Instantiated " |
| 120 | << node.name() |
| 121 | << " Type: " << node.type() |
Isabella Gottardi | 0ae5de9 | 2019-03-14 10:32:11 +0000 | [diff] [blame] | 122 | << " Target: " << TargetInfo::TargetType |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 123 | << " Data Type: " << input->info()->data_type() |
| 124 | << " Shape: " << input->info()->tensor_shape() |
| 125 | << " Activation function: " << act_info.activation() |
| 126 | << " a: " << act_info.a() |
| 127 | << " b: " << act_info.b() |
| 128 | << " InPlace : " << is_in_place_operation(input, output) |
| 129 | << std::endl); |
| 130 | |
Georgios Pinitas | 0b192e8 | 2020-02-20 17:09:28 +0000 | [diff] [blame] | 131 | return RETURN_UNIQUE_PTR(func); |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 132 | } |
| 133 | |
thecha01 | e8f05da | 2020-08-24 17:21:41 +0100 | [diff] [blame] | 134 | /** Creates a backend argminmax layer function |
| 135 | * |
| 136 | * @tparam ArgMinMaxLayerFunction Backend activation function |
| 137 | * @tparam TargetInfo Target-specific information |
| 138 | * |
| 139 | * @param[in] node Node to create the backend function for |
| 140 | * |
| 141 | * @return Backend argminmax layer function |
| 142 | */ |
| 143 | template <typename ArgMinMaxLayerFunction, typename TargetInfo> |
| 144 | std::unique_ptr<IFunction> create_arg_min_max_layer(ArgMinMaxLayerNode &node) |
| 145 | { |
| 146 | validate_node<TargetInfo>(node, 1 /* expected inputs */, 1 /* expected outputs */); |
| 147 | |
| 148 | // Extract IO and info |
| 149 | typename TargetInfo::TensorType *input = get_backing_tensor<TargetInfo>(node.input(0)); |
| 150 | typename TargetInfo::TensorType *output = get_backing_tensor<TargetInfo>(node.output(0)); |
| 151 | const ReductionOperation op = node.reduction_operation(); |
| 152 | unsigned int axis = node.axis(); |
| 153 | |
| 154 | // Create function |
| 155 | auto func = support::cpp14::make_unique<ArgMinMaxLayerFunction>(); |
| 156 | func->configure(input, axis, output, op); |
| 157 | |
| 158 | ARM_COMPUTE_LOG_GRAPH_INFO("Instantiated " |
| 159 | << node.name() |
| 160 | << " Type: " << node.type() |
| 161 | << " Target: " << TargetInfo::TargetType |
| 162 | << " Data Type: " << input->info()->data_type() |
| 163 | << " Shape: " << input->info()->tensor_shape() |
| 164 | << " Reduction Operation: " << op |
| 165 | << " axis: " << axis |
| 166 | << std::endl); |
| 167 | |
| 168 | return RETURN_UNIQUE_PTR(func); |
| 169 | } |
| 170 | |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 171 | /** Create a backend batch normalization layer function |
| 172 | * |
| 173 | * @tparam BatchNormalizationLayerFunction Backend batch normalization function |
| 174 | * @tparam TargetInfo Target-specific information |
| 175 | * |
| 176 | * @param[in] node Node to create the backend function for |
| 177 | * |
| 178 | * @return Backend batch normalization layer function |
| 179 | */ |
| 180 | template <typename BatchNormalizationLayerFunction, typename TargetInfo> |
| 181 | std::unique_ptr<IFunction> create_batch_normalization_layer(BatchNormalizationLayerNode &node) |
| 182 | { |
| 183 | validate_node<TargetInfo>(node, 5 /* expected inputs */, 1 /* expected outputs */); |
| 184 | |
| 185 | // Extract IO and info |
giuros01 | acce504 | 2019-02-21 17:32:34 +0000 | [diff] [blame] | 186 | typename TargetInfo::TensorType *input = get_backing_tensor<TargetInfo>(node.input(0)); |
| 187 | typename TargetInfo::TensorType *mean = get_backing_tensor<TargetInfo>(node.input(1)); |
| 188 | typename TargetInfo::TensorType *var = get_backing_tensor<TargetInfo>(node.input(2)); |
| 189 | typename TargetInfo::TensorType *beta = get_backing_tensor<TargetInfo>(node.input(3)); |
| 190 | typename TargetInfo::TensorType *gamma = get_backing_tensor<TargetInfo>(node.input(4)); |
| 191 | |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 192 | typename TargetInfo::TensorType *output = get_backing_tensor<TargetInfo>(node.output(0)); |
| 193 | const float epsilon = node.epsilon(); |
| 194 | const ActivationLayerInfo fused_act = node.fused_activation(); |
| 195 | |
| 196 | // Create and configure function |
| 197 | auto func = support::cpp14::make_unique<BatchNormalizationLayerFunction>(); |
| 198 | func->configure(input, output, mean, var, beta, gamma, epsilon, fused_act); |
| 199 | |
| 200 | // Log info |
Pablo Tello | 3252143 | 2018-11-15 14:43:10 +0000 | [diff] [blame] | 201 | ARM_COMPUTE_LOG_GRAPH_INFO("Instantiated " |
| 202 | << node.name() |
| 203 | << " Type: " << node.type() |
| 204 | << " Target: " << TargetInfo::TargetType |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 205 | << " Data Type: " << input->info()->data_type() |
| 206 | << " Shape: " << input->info()->tensor_shape() |
| 207 | << " Epsilon: " << epsilon << " " |
| 208 | << (fused_act.enabled() ? to_string(fused_act.activation()) : "") |
Pablo Tello | 3252143 | 2018-11-15 14:43:10 +0000 | [diff] [blame] | 209 | << " InPlace: " << is_in_place_operation(input, output) |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 210 | << std::endl); |
| 211 | |
Georgios Pinitas | 0b192e8 | 2020-02-20 17:09:28 +0000 | [diff] [blame] | 212 | return RETURN_UNIQUE_PTR(func); |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 213 | } |
| 214 | |
giuros01 | acce504 | 2019-02-21 17:32:34 +0000 | [diff] [blame] | 215 | /** Create a backend batch normalization layer function |
| 216 | * |
| 217 | * @tparam BatchNormalizationLayerFunction Backend batch normalization function |
| 218 | * @tparam TargetInfo Target-specific information |
| 219 | * |
| 220 | * @param[in] node Node to create the backend function for |
Gian Marco Iodice | 5dea19e | 2019-11-08 12:13:48 +0000 | [diff] [blame] | 221 | * @param[in] ctx Graph context |
giuros01 | acce504 | 2019-02-21 17:32:34 +0000 | [diff] [blame] | 222 | * |
| 223 | * @return Backend batch normalization layer function |
| 224 | */ |
| 225 | template <typename FusedLayerTypes, typename TargetInfo> |
Gian Marco Iodice | 5dea19e | 2019-11-08 12:13:48 +0000 | [diff] [blame] | 226 | std::unique_ptr<IFunction> create_fused_convolution_batch_normalization_layer(FusedConvolutionBatchNormalizationNode &node, GraphContext &ctx) |
giuros01 | acce504 | 2019-02-21 17:32:34 +0000 | [diff] [blame] | 227 | { |
| 228 | validate_node<TargetInfo>(node, 7 /* expected inputs */, 1 /* expected outputs */); |
| 229 | |
| 230 | // Extract IO and info |
| 231 | typename TargetInfo::TensorType *input = get_backing_tensor<TargetInfo>(node.input(0)); |
| 232 | typename TargetInfo::TensorType *weights = get_backing_tensor<TargetInfo>(node.input(1)); |
| 233 | typename TargetInfo::TensorType *biases = get_backing_tensor<TargetInfo>(node.input(2)); |
| 234 | typename TargetInfo::TensorType *mean = get_backing_tensor<TargetInfo>(node.input(3)); |
| 235 | typename TargetInfo::TensorType *var = get_backing_tensor<TargetInfo>(node.input(4)); |
| 236 | typename TargetInfo::TensorType *beta = get_backing_tensor<TargetInfo>(node.input(5)); |
| 237 | typename TargetInfo::TensorType *gamma = get_backing_tensor<TargetInfo>(node.input(6)); |
| 238 | |
| 239 | typename TargetInfo::TensorType *output = get_backing_tensor<TargetInfo>(node.output(0)); |
| 240 | |
| 241 | const PadStrideInfo conv_info = node.convolution_info(); |
| 242 | const unsigned int num_groups = node.num_groups(); |
| 243 | const bool fast_math = node.fast_math_hint() == FastMathHint::Enabled; |
| 244 | const ActivationLayerInfo fused_act = node.fused_activation(); |
| 245 | const float epsilon = node.epsilon(); |
| 246 | |
Gian Marco Iodice | 5dea19e | 2019-11-08 12:13:48 +0000 | [diff] [blame] | 247 | // Create and configure function (we assume that functions have been validated before creation) |
| 248 | std::shared_ptr<IMemoryManager> mm = get_memory_manager(ctx, TargetInfo::TargetType); |
| 249 | std::unique_ptr<IFunction> func; |
| 250 | std::string func_name; |
| 251 | |
| 252 | using FType = FusedConvolutionBatchNormalizationFunction<TargetInfo, FusedLayerTypes>; |
| 253 | |
giuros01 | acce504 | 2019-02-21 17:32:34 +0000 | [diff] [blame] | 254 | // Create and configure function |
Gian Marco Iodice | 5dea19e | 2019-11-08 12:13:48 +0000 | [diff] [blame] | 255 | std::tie(func, func_name) = create_named_memory_managed_function<FType>( |
| 256 | std::string("FusedConvolutionBatchNormalizationLayer"), mm, input, weights, biases, output, mean, var, beta, gamma, epsilon, conv_info, num_groups, fast_math, fused_act); |
giuros01 | acce504 | 2019-02-21 17:32:34 +0000 | [diff] [blame] | 257 | |
| 258 | // Log info |
| 259 | ARM_COMPUTE_LOG_GRAPH_INFO("Instantiated " |
| 260 | << node.name() |
Manuel Bottini | bffb41e | 2019-06-20 16:00:27 +0100 | [diff] [blame] | 261 | << " Type: " << node.type() |
| 262 | << " Target: " << TargetInfo::TargetType |
| 263 | << " Data Type: " << input->info()->data_type() |
| 264 | << " Input shape: " << input->info()->tensor_shape() |
| 265 | << " Weights shape: " << weights->info()->tensor_shape() |
| 266 | << " Output shape: " << output->info()->tensor_shape() |
| 267 | << (fused_act.enabled() ? " " + to_string(fused_act.activation()) : "") |
| 268 | << std::endl); |
Georgios Pinitas | 0b192e8 | 2020-02-20 17:09:28 +0000 | [diff] [blame] | 269 | return RETURN_UNIQUE_PTR(func); |
Manuel Bottini | bffb41e | 2019-06-20 16:00:27 +0100 | [diff] [blame] | 270 | } |
| 271 | |
| 272 | /** Create a backend fused depthwise convolution batch normalization layer function |
| 273 | * |
| 274 | * @tparam FusedLayerTypes Fused layer types |
| 275 | * @tparam TargetInfo Target-specific information |
| 276 | * |
| 277 | * @param[in] node Node to create the backend function for |
Gian Marco Iodice | 5dea19e | 2019-11-08 12:13:48 +0000 | [diff] [blame] | 278 | * @param[in] ctx Graph context |
Manuel Bottini | bffb41e | 2019-06-20 16:00:27 +0100 | [diff] [blame] | 279 | * |
| 280 | * @return Backend fused depthwise convolution batch normalization layer function |
| 281 | */ |
| 282 | template <typename FusedLayerTypes, typename TargetInfo> |
Gian Marco Iodice | 5dea19e | 2019-11-08 12:13:48 +0000 | [diff] [blame] | 283 | std::unique_ptr<IFunction> create_fused_depthwise_convolution_batch_normalization_layer(FusedDepthwiseConvolutionBatchNormalizationNode &node, GraphContext &ctx) |
Manuel Bottini | bffb41e | 2019-06-20 16:00:27 +0100 | [diff] [blame] | 284 | { |
| 285 | validate_node<TargetInfo>(node, 7 /* expected inputs */, 1 /* expected outputs */); |
| 286 | |
| 287 | // Extract IO and info |
| 288 | typename TargetInfo::TensorType *input = get_backing_tensor<TargetInfo>(node.input(0)); |
| 289 | typename TargetInfo::TensorType *weights = get_backing_tensor<TargetInfo>(node.input(1)); |
| 290 | typename TargetInfo::TensorType *biases = get_backing_tensor<TargetInfo>(node.input(2)); |
| 291 | typename TargetInfo::TensorType *mean = get_backing_tensor<TargetInfo>(node.input(3)); |
| 292 | typename TargetInfo::TensorType *var = get_backing_tensor<TargetInfo>(node.input(4)); |
| 293 | typename TargetInfo::TensorType *beta = get_backing_tensor<TargetInfo>(node.input(5)); |
| 294 | typename TargetInfo::TensorType *gamma = get_backing_tensor<TargetInfo>(node.input(6)); |
| 295 | |
| 296 | typename TargetInfo::TensorType *output = get_backing_tensor<TargetInfo>(node.output(0)); |
| 297 | |
| 298 | const PadStrideInfo conv_info = node.convolution_info(); |
| 299 | const unsigned int depth_multiplier = node.depth_multiplier(); |
| 300 | const ActivationLayerInfo fused_act = node.fused_activation(); |
| 301 | const float epsilon = node.epsilon(); |
| 302 | |
Gian Marco Iodice | 5dea19e | 2019-11-08 12:13:48 +0000 | [diff] [blame] | 303 | // Create and configure function (we assume that functions have been validated before creation) |
| 304 | std::shared_ptr<IMemoryManager> mm = get_memory_manager(ctx, TargetInfo::TargetType); |
| 305 | std::unique_ptr<IFunction> func; |
| 306 | std::string func_name; |
| 307 | |
| 308 | using FType = FusedDepthwiseConvolutionBatchNormalizationFunction<TargetInfo, FusedLayerTypes>; |
| 309 | |
Manuel Bottini | bffb41e | 2019-06-20 16:00:27 +0100 | [diff] [blame] | 310 | // Create and configure function |
Gian Marco Iodice | 5dea19e | 2019-11-08 12:13:48 +0000 | [diff] [blame] | 311 | std::tie(func, func_name) = create_named_memory_managed_function<FType>( |
| 312 | std::string("FusedDepthwiseConvolutionBatchNormalizationLayer"), mm, input, weights, biases, output, mean, var, beta, gamma, epsilon, conv_info, depth_multiplier, fused_act); |
Manuel Bottini | bffb41e | 2019-06-20 16:00:27 +0100 | [diff] [blame] | 313 | |
| 314 | // Log info |
| 315 | ARM_COMPUTE_LOG_GRAPH_INFO("Instantiated " |
| 316 | << node.name() |
| 317 | << " Type: " << node.type() |
giuros01 | acce504 | 2019-02-21 17:32:34 +0000 | [diff] [blame] | 318 | << " Target: " << TargetInfo::TargetType |
| 319 | << " Data Type: " << input->info()->data_type() |
| 320 | << " Input shape: " << input->info()->tensor_shape() |
| 321 | << " Weights shape: " << weights->info()->tensor_shape() |
| 322 | << " Output shape: " << output->info()->tensor_shape() |
| 323 | << (fused_act.enabled() ? " " + to_string(fused_act.activation()) : "") |
| 324 | << std::endl); |
Georgios Pinitas | 0b192e8 | 2020-02-20 17:09:28 +0000 | [diff] [blame] | 325 | return RETURN_UNIQUE_PTR(func); |
giuros01 | acce504 | 2019-02-21 17:32:34 +0000 | [diff] [blame] | 326 | } |
| 327 | |
Manuel Bottini | d2048ce | 2018-10-23 17:00:42 +0100 | [diff] [blame] | 328 | /** Create a backend bounding box transform layer function |
| 329 | * |
| 330 | * @tparam BoundingBoxTransformLayerFunction Backend bounding box transform function |
| 331 | * @tparam TargetInfo Target-specific information |
| 332 | * |
| 333 | * @param[in] node Node to create the backend function for |
| 334 | * |
| 335 | * @return Backend bounding box transform layer function |
| 336 | */ |
| 337 | template <typename BoundingBoxTransformLayerFunction, typename TargetInfo> |
| 338 | std::unique_ptr<IFunction> create_bounding_box_transform_layer(BoundingBoxTransformLayerNode &node) |
| 339 | { |
| 340 | validate_node<TargetInfo>(node, 2 /* expected inputs */, 1 /* expected outputs */); |
| 341 | |
| 342 | // Extract IO and info |
| 343 | typename TargetInfo::TensorType *input = get_backing_tensor<TargetInfo>(node.input(0)); |
| 344 | typename TargetInfo::TensorType *deltas = get_backing_tensor<TargetInfo>(node.input(1)); |
| 345 | typename TargetInfo::TensorType *output = get_backing_tensor<TargetInfo>(node.output(0)); |
| 346 | const BoundingBoxTransformInfo bbox_info = node.info(); |
| 347 | |
| 348 | // Create and configure function |
| 349 | auto func = support::cpp14::make_unique<BoundingBoxTransformLayerFunction>(); |
| 350 | func->configure(input, output, deltas, bbox_info); |
| 351 | |
| 352 | // Log info |
Isabella Gottardi | 0ae5de9 | 2019-03-14 10:32:11 +0000 | [diff] [blame] | 353 | ARM_COMPUTE_LOG_GRAPH_INFO("Instantiated " |
| 354 | << node.name() |
| 355 | << " Type: " << node.type() |
| 356 | << " Target: " << TargetInfo::TargetType |
Manuel Bottini | d2048ce | 2018-10-23 17:00:42 +0100 | [diff] [blame] | 357 | << " Data Type: " << input->info()->data_type() |
| 358 | << " Shape: " << input->info()->tensor_shape() |
| 359 | << " BoundingBox Info img W: " << bbox_info.img_width() << " " |
| 360 | << " BoundingBox Info img H: " << bbox_info.img_height() << " " |
| 361 | << std::endl); |
| 362 | |
Georgios Pinitas | 0b192e8 | 2020-02-20 17:09:28 +0000 | [diff] [blame] | 363 | return RETURN_UNIQUE_PTR(func); |
Manuel Bottini | d2048ce | 2018-10-23 17:00:42 +0100 | [diff] [blame] | 364 | } |
| 365 | |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 366 | /** Create a backend channel shuffle layer function |
| 367 | * |
| 368 | * @tparam ChannelShuffleLayerFunction Backend channel shuffle function |
| 369 | * @tparam TargetInfo Target-specific information |
| 370 | * |
| 371 | * @param[in] node Node to create the backend function for |
| 372 | * |
| 373 | * @return Backend channel shuffle layer function |
| 374 | */ |
| 375 | template <typename ChannelShuffleLayerFunction, typename TargetInfo> |
| 376 | std::unique_ptr<IFunction> create_channel_shuffle_layer(ChannelShuffleLayerNode &node) |
| 377 | { |
| 378 | validate_node<TargetInfo>(node, 1 /* expected inputs */, 1 /* expected outputs */); |
| 379 | |
| 380 | // Extract IO and info |
| 381 | typename TargetInfo::TensorType *input = get_backing_tensor<TargetInfo>(node.input(0)); |
| 382 | typename TargetInfo::TensorType *output = get_backing_tensor<TargetInfo>(node.output(0)); |
| 383 | const unsigned int num_groups = node.num_groups(); |
| 384 | |
| 385 | // Create function |
| 386 | auto func = support::cpp14::make_unique<ChannelShuffleLayerFunction>(); |
| 387 | func->configure(input, output, num_groups); |
| 388 | |
Pablo Tello | 3252143 | 2018-11-15 14:43:10 +0000 | [diff] [blame] | 389 | ARM_COMPUTE_LOG_GRAPH_INFO("Instantiated " |
| 390 | << node.name() |
| 391 | << " Type: " << node.type() |
| 392 | << " Target: " << TargetInfo::TargetType |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 393 | << " Data Type: " << input->info()->data_type() |
| 394 | << " Shape: " << input->info()->tensor_shape() |
| 395 | << " Num groups: " << num_groups |
| 396 | << std::endl); |
| 397 | |
Georgios Pinitas | 0b192e8 | 2020-02-20 17:09:28 +0000 | [diff] [blame] | 398 | return RETURN_UNIQUE_PTR(func); |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 399 | } |
| 400 | |
Georgios Pinitas | e222055 | 2018-07-20 13:23:44 +0100 | [diff] [blame] | 401 | /** Create a backend layer concatenate function |
| 402 | * |
| 403 | * @tparam ConcatenateLayerFunction Backend concatenate function |
| 404 | * @tparam TargetInfo Target-specific information |
| 405 | * |
| 406 | * @param[in] node Node to create the backend function for |
| 407 | * |
| 408 | * @return Backend concatenate layer function |
| 409 | */ |
| 410 | template <typename ConcatenateLayerFunction, typename TargetInfo> |
| 411 | std::unique_ptr<arm_compute::IFunction> create_concatenate_layer(ConcatenateLayerNode &node) |
| 412 | { |
| 413 | ARM_COMPUTE_LOG_GRAPH_VERBOSE("Creating Concatenate node with ID : " << node.id() << " and Name: " << node.name() << std::endl); |
| 414 | ARM_COMPUTE_ERROR_ON(node.num_outputs() != 1); |
| 415 | |
| 416 | // Return nullptr if depth concatenate is switched off |
| 417 | if(!node.is_enabled()) |
| 418 | { |
| 419 | return nullptr; |
| 420 | } |
| 421 | |
| 422 | // Extract IO and info |
Georgios Pinitas | 4667ddd | 2020-07-13 21:21:33 +0100 | [diff] [blame] | 423 | std::vector<typename TargetInfo::SrcTensorType *> inputs; |
Georgios Pinitas | e222055 | 2018-07-20 13:23:44 +0100 | [diff] [blame] | 424 | for(unsigned int i = 0; i < node.num_inputs(); ++i) |
| 425 | { |
| 426 | inputs.push_back(get_backing_tensor<TargetInfo>(node.input(i))); |
| 427 | } |
| 428 | typename TargetInfo::TensorType *output = get_backing_tensor<TargetInfo>(node.output(0)); |
Georgios Pinitas | 9e4824c | 2019-04-12 13:15:58 +0100 | [diff] [blame] | 429 | const DataLayout data_layout = node.output(0) != nullptr ? node.output(0)->desc().layout : DataLayout::UNKNOWN; |
| 430 | const size_t concat_axis = get_dimension_idx(data_layout, node.concatenation_axis()); |
Georgios Pinitas | e222055 | 2018-07-20 13:23:44 +0100 | [diff] [blame] | 431 | |
| 432 | // Create and configure function |
| 433 | auto func = support::cpp14::make_unique<ConcatenateLayerFunction>(); |
| 434 | func->configure(inputs, output, concat_axis); |
| 435 | |
| 436 | // Log info |
Isabella Gottardi | 0ae5de9 | 2019-03-14 10:32:11 +0000 | [diff] [blame] | 437 | const bool is_quantized = is_data_type_quantized_asymmetric(output->info()->data_type()); |
| 438 | std::ostringstream qss; |
| 439 | if(is_quantized) |
| 440 | { |
| 441 | qss << " Output QuantInfo: " << output->info()->quantization_info(); |
| 442 | } |
Pablo Tello | 3252143 | 2018-11-15 14:43:10 +0000 | [diff] [blame] | 443 | ARM_COMPUTE_LOG_GRAPH_INFO("Instantiated " |
| 444 | << node.name() |
| 445 | << " Type: " << node.type() |
| 446 | << " Target: " << TargetInfo::TargetType |
Georgios Pinitas | e222055 | 2018-07-20 13:23:44 +0100 | [diff] [blame] | 447 | << " Data Type: " << output->info()->data_type() |
| 448 | << " Shape: " << output->info()->tensor_shape() |
| 449 | << " Num Inputs: " << inputs.size() |
| 450 | << " Axis: " << concat_axis |
Isabella Gottardi | 0ae5de9 | 2019-03-14 10:32:11 +0000 | [diff] [blame] | 451 | << qss.str() |
Georgios Pinitas | e222055 | 2018-07-20 13:23:44 +0100 | [diff] [blame] | 452 | << std::endl); |
| 453 | |
Georgios Pinitas | 0b192e8 | 2020-02-20 17:09:28 +0000 | [diff] [blame] | 454 | return RETURN_UNIQUE_PTR(func); |
Georgios Pinitas | e222055 | 2018-07-20 13:23:44 +0100 | [diff] [blame] | 455 | } |
| 456 | |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 457 | /** Create a backend convolution layer function |
| 458 | * |
| 459 | * @tparam ConvolutionLayerFunctions Backend convolution functions |
| 460 | * @tparam TargetInfo Target-specific information |
| 461 | * |
| 462 | * @param[in] node Node to create the backend function for |
| 463 | * @param[in] ctx Graph context |
| 464 | * |
| 465 | * @return Backend convolution layer function |
| 466 | */ |
| 467 | template <typename ConvolutionLayerFunctions, typename TargetInfo> |
| 468 | std::unique_ptr<IFunction> create_convolution_layer(ConvolutionLayerNode &node, GraphContext &ctx) |
| 469 | { |
| 470 | validate_node<TargetInfo>(node, 3 /* expected inputs */, 1 /* expected outputs */); |
| 471 | |
| 472 | // Extract IO and info |
| 473 | typename TargetInfo::TensorType *input = get_backing_tensor<TargetInfo>(node.input(0)); |
| 474 | typename TargetInfo::TensorType *weights = get_backing_tensor<TargetInfo>(node.input(1)); |
| 475 | typename TargetInfo::TensorType *biases = get_backing_tensor<TargetInfo>(node.input(2)); |
| 476 | typename TargetInfo::TensorType *output = get_backing_tensor<TargetInfo>(node.output(0)); |
| 477 | |
Georgios Pinitas | fd7e853 | 2018-09-07 10:51:27 +0100 | [diff] [blame] | 478 | const bool is_quantized = is_data_type_quantized_asymmetric(input->info()->data_type()); |
| 479 | |
| 480 | if(is_quantized) |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 481 | { |
| 482 | biases->info()->set_data_type(DataType::S32); |
| 483 | } |
| 484 | |
Georgios Pinitas | 08346e9 | 2018-10-16 19:10:46 +0100 | [diff] [blame] | 485 | const PadStrideInfo conv_info = node.convolution_info(); |
| 486 | const unsigned int num_groups = node.num_groups(); |
| 487 | const ConvolutionMethod conv_algorithm = node.convolution_method(); |
| 488 | const bool fast_math = node.fast_math_hint() == FastMathHint::Enabled; |
| 489 | const ActivationLayerInfo fused_act = node.fused_activation(); |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 490 | |
| 491 | // Create and configure function (we assume that functions have been validated before creation) |
| 492 | std::shared_ptr<IMemoryManager> mm = get_memory_manager(ctx, TargetInfo::TargetType); |
| 493 | std::unique_ptr<IFunction> func; |
| 494 | std::string func_name; |
| 495 | |
Georgios Pinitas | e222055 | 2018-07-20 13:23:44 +0100 | [diff] [blame] | 496 | if(conv_algorithm == ConvolutionMethod::Winograd) |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 497 | { |
Georgios Pinitas | 2a2db59 | 2018-08-15 12:14:46 +0100 | [diff] [blame] | 498 | ARM_COMPUTE_ERROR_ON_MSG(num_groups != 1, "WinogradConvolutionLayer does not support grouping!"); |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 499 | std::tie(func, func_name) = create_named_memory_managed_function<typename ConvolutionLayerFunctions::WinogradConvolutionLayer>( |
| 500 | std::string("WinogradConvolutionLayer"), mm, |
Georgios Pinitas | 08346e9 | 2018-10-16 19:10:46 +0100 | [diff] [blame] | 501 | input, weights, biases, output, conv_info, fused_act, fast_math); |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 502 | } |
Georgios Pinitas | e222055 | 2018-07-20 13:23:44 +0100 | [diff] [blame] | 503 | else if(conv_algorithm == ConvolutionMethod::Direct) |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 504 | { |
Georgios Pinitas | 2a2db59 | 2018-08-15 12:14:46 +0100 | [diff] [blame] | 505 | ARM_COMPUTE_ERROR_ON_MSG(num_groups != 1, "DirectConvolutionLayer does not support grouping!"); |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 506 | std::tie(func, func_name) = create_named_function<typename ConvolutionLayerFunctions::DirectConvolutionLayer>( |
| 507 | std::string("DirectConvolutionLayer"), |
Georgios Pinitas | 08346e9 | 2018-10-16 19:10:46 +0100 | [diff] [blame] | 508 | input, weights, biases, output, conv_info, fused_act); |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 509 | } |
| 510 | else if(conv_algorithm == ConvolutionMethod::GEMM) |
| 511 | { |
| 512 | std::tie(func, func_name) = create_named_memory_managed_function<typename ConvolutionLayerFunctions::GEMMConvolutionLayer>( |
| 513 | std::string("GEMMConvolutionLayer"), mm, |
Georgios Pinitas | 2a2db59 | 2018-08-15 12:14:46 +0100 | [diff] [blame] | 514 | input, weights, biases, output, conv_info, |
Georgios Pinitas | 08346e9 | 2018-10-16 19:10:46 +0100 | [diff] [blame] | 515 | WeightsInfo(), Size2D(1U, 1U), fused_act, num_groups); |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 516 | } |
| 517 | else |
| 518 | { |
| 519 | std::tie(func, func_name) = create_named_memory_managed_function<typename ConvolutionLayerFunctions::GenericConvolutionLayer>( |
| 520 | std::string("GenericConvolutionLayer"), mm, |
Georgios Pinitas | 2a2db59 | 2018-08-15 12:14:46 +0100 | [diff] [blame] | 521 | input, weights, biases, output, conv_info, |
Georgios Pinitas | 08346e9 | 2018-10-16 19:10:46 +0100 | [diff] [blame] | 522 | WeightsInfo(), Size2D(1U, 1U), fused_act, fast_math, num_groups); |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 523 | } |
| 524 | |
| 525 | // Log info |
Georgios Pinitas | fd7e853 | 2018-09-07 10:51:27 +0100 | [diff] [blame] | 526 | std::ostringstream qss; |
| 527 | if(is_quantized) |
| 528 | { |
| 529 | qss << " Input QuantInfo: " << input->info()->quantization_info() |
| 530 | << " Weights QuantInfo: " << weights->info()->quantization_info() |
| 531 | << " Output QuantInfo: " << output->info()->quantization_info(); |
| 532 | } |
Pablo Tello | 3252143 | 2018-11-15 14:43:10 +0000 | [diff] [blame] | 533 | ARM_COMPUTE_LOG_GRAPH_INFO("Instantiated " |
| 534 | << node.name() |
| 535 | << " Type: " << func_name |
| 536 | << " Target: " << TargetInfo::TargetType |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 537 | << " Data Type: " << input->info()->data_type() |
Georgios Pinitas | 2a2db59 | 2018-08-15 12:14:46 +0100 | [diff] [blame] | 538 | << " Groups: " << num_groups |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 539 | << " Input shape: " << input->info()->tensor_shape() |
| 540 | << " Weights shape: " << weights->info()->tensor_shape() |
| 541 | << " Output shape: " << output->info()->tensor_shape() |
Isabella Gottardi | 0ae5de9 | 2019-03-14 10:32:11 +0000 | [diff] [blame] | 542 | << qss.str() |
Georgios Pinitas | 08346e9 | 2018-10-16 19:10:46 +0100 | [diff] [blame] | 543 | << (fused_act.enabled() ? " " + to_string(fused_act.activation()) : "") |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 544 | << std::endl); |
Georgios Pinitas | 0b192e8 | 2020-02-20 17:09:28 +0000 | [diff] [blame] | 545 | return RETURN_UNIQUE_PTR(func); |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 546 | } |
| 547 | |
| 548 | /** Create a backend deconvolution layer function |
| 549 | * |
| 550 | * @tparam DeconvolutionLayerFunction Backend deconvolution function |
| 551 | * @tparam TargetInfo Target-specific information |
| 552 | * |
| 553 | * @param[in] node Node to create the backend function for |
| 554 | * @param[in] ctx Graph context |
| 555 | * |
| 556 | * @return Backend deconvolution layer function |
| 557 | */ |
| 558 | template <typename DeconvolutionLayerFunction, typename TargetInfo> |
| 559 | std::unique_ptr<IFunction> create_deconvolution_layer(DeconvolutionLayerNode &node, GraphContext &ctx) |
| 560 | { |
| 561 | validate_node<TargetInfo>(node, 3 /* expected inputs */, 1 /* expected outputs */); |
| 562 | |
| 563 | // Extract IO and info |
| 564 | typename TargetInfo::TensorType *input = get_backing_tensor<TargetInfo>(node.input(0)); |
| 565 | typename TargetInfo::TensorType *weights = get_backing_tensor<TargetInfo>(node.input(1)); |
| 566 | typename TargetInfo::TensorType *biases = get_backing_tensor<TargetInfo>(node.input(2)); |
| 567 | typename TargetInfo::TensorType *output = get_backing_tensor<TargetInfo>(node.output(0)); |
| 568 | |
Manuel Bottini | c1b76fa | 2019-06-17 12:04:40 +0100 | [diff] [blame] | 569 | const PadStrideInfo deconv_info = node.deconvolution_info(); |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 570 | |
| 571 | // Create and configure function (we assume that functions have been validated before creation) |
| 572 | std::shared_ptr<IMemoryManager> mm = get_memory_manager(ctx, TargetInfo::TargetType); |
| 573 | std::unique_ptr<IFunction> func; |
| 574 | |
| 575 | std::tie(func, std::ignore) = create_named_memory_managed_function<DeconvolutionLayerFunction>( |
| 576 | std::string(), mm, |
Manuel Bottini | c1b76fa | 2019-06-17 12:04:40 +0100 | [diff] [blame] | 577 | input, weights, biases, output, deconv_info); |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 578 | |
| 579 | // Log info |
Pablo Tello | 3252143 | 2018-11-15 14:43:10 +0000 | [diff] [blame] | 580 | ARM_COMPUTE_LOG_GRAPH_INFO("Instantiated " |
| 581 | << node.name() |
| 582 | << " Type: " << node.type() |
| 583 | << " Target: " << TargetInfo::TargetType |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 584 | << " Data Type: " << input->info()->data_type() |
| 585 | << " Input shape: " << input->info()->tensor_shape() |
| 586 | << " Weights shape: " << weights->info()->tensor_shape() |
| 587 | << " Output shape: " << output->info()->tensor_shape() |
| 588 | << std::endl); |
| 589 | return func; |
| 590 | } |
| 591 | |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 592 | /** Create a backend layer depth-wise convolution function |
| 593 | * |
| 594 | * @tparam DepthwiseConvolutionLayerFunctions Backend depthwise convolution function |
| 595 | * @tparam TargetInfo Target-specific information |
| 596 | * |
| 597 | * @param[in] node Node to create the backend function for |
| 598 | * |
| 599 | * @return Backend depth-wise convolution layer function |
| 600 | */ |
Manuel Bottini | 05069f0 | 2019-09-26 17:18:26 +0100 | [diff] [blame] | 601 | template <typename DepthwiseConvolutionLayer, typename TargetInfo> |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 602 | std::unique_ptr<IFunction> create_depthwise_convolution_layer(DepthwiseConvolutionLayerNode &node) |
| 603 | { |
| 604 | validate_node<TargetInfo>(node, 3 /* expected inputs */, 1 /* expected outputs */); |
| 605 | |
| 606 | // Extract IO and info |
| 607 | typename TargetInfo::TensorType *input = get_backing_tensor<TargetInfo>(node.input(0)); |
| 608 | typename TargetInfo::TensorType *weights = get_backing_tensor<TargetInfo>(node.input(1)); |
| 609 | typename TargetInfo::TensorType *biases = get_backing_tensor<TargetInfo>(node.input(2)); |
| 610 | typename TargetInfo::TensorType *output = get_backing_tensor<TargetInfo>(node.output(0)); |
| 611 | |
Georgios Pinitas | fd7e853 | 2018-09-07 10:51:27 +0100 | [diff] [blame] | 612 | const bool is_quantized = is_data_type_quantized_asymmetric(input->info()->data_type()); |
| 613 | |
| 614 | if(is_quantized) |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 615 | { |
| 616 | biases->info()->set_data_type(DataType::S32); |
| 617 | } |
| 618 | |
Manuel Bottini | 05069f0 | 2019-09-26 17:18:26 +0100 | [diff] [blame] | 619 | const PadStrideInfo conv_info = node.convolution_info(); |
| 620 | const unsigned int depth_multiplier = node.depth_multiplier(); |
| 621 | const ActivationLayerInfo fused_act = node.fused_activation(); |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 622 | |
| 623 | // Create and configure function (we assume that functions have been validated before creation) |
| 624 | std::unique_ptr<IFunction> func; |
| 625 | std::string func_name; |
Manuel Bottini | 05069f0 | 2019-09-26 17:18:26 +0100 | [diff] [blame] | 626 | |
| 627 | std::tie(func, func_name) = create_named_function<DepthwiseConvolutionLayer>( |
| 628 | std::string("DepthwiseConvolutionLayer"), |
| 629 | input, weights, biases, output, conv_info, depth_multiplier, fused_act); |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 630 | |
| 631 | // Log info |
Georgios Pinitas | fd7e853 | 2018-09-07 10:51:27 +0100 | [diff] [blame] | 632 | std::ostringstream qss; |
| 633 | if(is_quantized) |
| 634 | { |
| 635 | qss << " Input QuantInfo: " << input->info()->quantization_info() |
| 636 | << " Weights QuantInfo: " << weights->info()->quantization_info() |
| 637 | << " Output QuantInfo: " << output->info()->quantization_info(); |
| 638 | } |
Pablo Tello | 3252143 | 2018-11-15 14:43:10 +0000 | [diff] [blame] | 639 | ARM_COMPUTE_LOG_GRAPH_INFO("Instantiated " |
| 640 | << node.name() |
| 641 | << " Type: " << func_name |
| 642 | << " Target: " << TargetInfo::TargetType |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 643 | << " Data Type: " << input->info()->data_type() |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 644 | << " Input shape: " << input->info()->tensor_shape() |
| 645 | << " Weights shape: " << weights->info()->tensor_shape() |
| 646 | << " Output shape: " << output->info()->tensor_shape() |
Georgios Pinitas | 05045c1 | 2018-12-07 18:31:47 +0000 | [diff] [blame] | 647 | << " Depth multiplier: " << depth_multiplier |
Isabella Gottardi | 0ae5de9 | 2019-03-14 10:32:11 +0000 | [diff] [blame] | 648 | << qss.str() |
Georgios Pinitas | 60e9825 | 2018-10-22 16:17:20 +0100 | [diff] [blame] | 649 | << (fused_act.enabled() ? " " + to_string(fused_act.activation()) : "") |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 650 | << std::endl); |
Georgios Pinitas | 0b192e8 | 2020-02-20 17:09:28 +0000 | [diff] [blame] | 651 | return RETURN_UNIQUE_PTR(func); |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 652 | } |
| 653 | |
thecha01 | 0a05e6a | 2020-08-28 18:40:38 +0100 | [diff] [blame] | 654 | /** Create a backend depth to space layer function |
| 655 | * |
| 656 | * @tparam DepthToSpaceLayerNode Function Backend depth to space function |
| 657 | * @tparam TargetInfo Target-specific information |
| 658 | * |
| 659 | * @param[in] node Node to create the backend function for |
| 660 | * |
| 661 | * @return Backend depth to space layer function |
| 662 | */ |
| 663 | template <typename DepthToSpaceLayerFunction, typename TargetInfo> |
| 664 | std::unique_ptr<IFunction> create_depth_to_space_layer(DepthToSpaceLayerNode &node) |
| 665 | { |
| 666 | validate_node<TargetInfo>(node, 1 /* expected inputs */, 1 /* expected outputs */); |
| 667 | |
| 668 | // Extract IO and info |
| 669 | typename TargetInfo::TensorType *input = get_backing_tensor<TargetInfo>(node.input(0)); |
| 670 | typename TargetInfo::TensorType *output = get_backing_tensor<TargetInfo>(node.output(0)); |
| 671 | |
| 672 | ARM_COMPUTE_ERROR_ON(input == nullptr); |
| 673 | ARM_COMPUTE_ERROR_ON(output == nullptr); |
| 674 | |
| 675 | // Create and configure function |
| 676 | auto func = support::cpp14::make_unique<DepthToSpaceLayerFunction>(); |
| 677 | func->configure(input, output, node.block_shape()); |
| 678 | |
| 679 | // Log info |
| 680 | ARM_COMPUTE_LOG_GRAPH_INFO("Instantiated " |
| 681 | << node.name() |
| 682 | << " Type: " << node.type() |
| 683 | << " Target: " << TargetInfo::TargetType |
| 684 | << " Data Type: " << input->info()->data_type() |
| 685 | << " Input shape: " << input->info()->tensor_shape() |
| 686 | << " Block Size: " << node.block_shape() |
| 687 | << " Output shape: " << output->info()->tensor_shape() |
| 688 | << std::endl); |
| 689 | |
| 690 | return RETURN_UNIQUE_PTR(func); |
| 691 | } |
| 692 | |
Isabella Gottardi | cd4e9ab | 2019-11-05 17:50:27 +0000 | [diff] [blame] | 693 | /** Create a backend dequantize layer function |
| 694 | * |
| 695 | * @tparam DequantizationLayer Function Backend dequantize function |
| 696 | * @tparam TargetInfo Target-specific information |
| 697 | * |
| 698 | * @param[in] node Node to create the backend function for |
| 699 | * |
| 700 | * @return Backend dequantize layer function |
| 701 | */ |
| 702 | template <typename DequantizationLayerFunction, typename TargetInfo> |
| 703 | std::unique_ptr<IFunction> create_dequantization_layer(DequantizationLayerNode &node) |
| 704 | { |
| 705 | validate_node<TargetInfo>(node, 1 /* expected inputs */, 1 /* expected outputs */); |
| 706 | |
| 707 | // Extract IO and info |
| 708 | typename TargetInfo::TensorType *input = get_backing_tensor<TargetInfo>(node.input(0)); |
| 709 | typename TargetInfo::TensorType *output = get_backing_tensor<TargetInfo>(node.output(0)); |
| 710 | |
| 711 | ARM_COMPUTE_ERROR_ON(input == nullptr); |
| 712 | ARM_COMPUTE_ERROR_ON(output == nullptr); |
| 713 | |
| 714 | // Create and configure function |
| 715 | auto func = support::cpp14::make_unique<DequantizationLayerFunction>(); |
| 716 | func->configure(input, output); |
| 717 | |
| 718 | // Log info |
| 719 | ARM_COMPUTE_LOG_GRAPH_INFO("Instantiated " |
| 720 | << node.name() |
| 721 | << " Type: " << node.type() |
| 722 | << " Target: " << TargetInfo::TargetType |
| 723 | << " Data Type: " << input->info()->data_type() |
| 724 | << " Input shape: " << input->info()->tensor_shape() |
| 725 | << " Input quantization info: " << output->info()->quantization_info() |
| 726 | << " Output shape: " << output->info()->tensor_shape() |
| 727 | << std::endl); |
| 728 | |
Georgios Pinitas | 0b192e8 | 2020-02-20 17:09:28 +0000 | [diff] [blame] | 729 | return RETURN_UNIQUE_PTR(func); |
Isabella Gottardi | cd4e9ab | 2019-11-05 17:50:27 +0000 | [diff] [blame] | 730 | } |
Isabella Gottardi | 7234ed8 | 2018-11-27 08:51:10 +0000 | [diff] [blame] | 731 | /** Create a backend detection output layer function |
| 732 | * |
| 733 | * @tparam DetectionOutputLayer Function Backend detection output function |
| 734 | * @tparam TargetInfo Target-specific information |
| 735 | * |
| 736 | * @param[in] node Node to create the backend function for |
| 737 | * |
| 738 | * @return Backend detection output layer function |
| 739 | */ |
| 740 | template <typename DetectionOutputLayerFunction, typename TargetInfo> |
| 741 | std::unique_ptr<IFunction> create_detection_output_layer(DetectionOutputLayerNode &node) |
| 742 | { |
| 743 | validate_node<TargetInfo>(node, 3 /* expected inputs */, 1 /* expected outputs */); |
| 744 | |
| 745 | // Extract IO and info |
| 746 | typename TargetInfo::TensorType *input0 = get_backing_tensor<TargetInfo>(node.input(0)); |
| 747 | typename TargetInfo::TensorType *input1 = get_backing_tensor<TargetInfo>(node.input(1)); |
| 748 | typename TargetInfo::TensorType *input2 = get_backing_tensor<TargetInfo>(node.input(2)); |
| 749 | typename TargetInfo::TensorType *output = get_backing_tensor<TargetInfo>(node.output(0)); |
| 750 | const DetectionOutputLayerInfo detect_info = node.detection_output_info(); |
| 751 | |
| 752 | ARM_COMPUTE_ERROR_ON(input0 == nullptr); |
| 753 | ARM_COMPUTE_ERROR_ON(input1 == nullptr); |
| 754 | ARM_COMPUTE_ERROR_ON(input2 == nullptr); |
| 755 | ARM_COMPUTE_ERROR_ON(output == nullptr); |
| 756 | |
| 757 | // Create and configure function |
| 758 | auto func = support::cpp14::make_unique<DetectionOutputLayerFunction>(); |
| 759 | func->configure(input0, input1, input2, output, detect_info); |
| 760 | |
| 761 | // Log info |
| 762 | ARM_COMPUTE_LOG_GRAPH_INFO("Instantiated " |
| 763 | << node.name() |
| 764 | << " Type: " << node.type() |
| 765 | << " Target: " << TargetInfo::TargetType |
| 766 | << " Data Type: " << input0->info()->data_type() |
| 767 | << " Input0 shape: " << input0->info()->tensor_shape() |
| 768 | << " Input1 shape: " << input1->info()->tensor_shape() |
| 769 | << " Input2 shape: " << input2->info()->tensor_shape() |
| 770 | << " Output shape: " << output->info()->tensor_shape() |
| 771 | << " DetectionOutputLayer info: " << detect_info |
| 772 | << std::endl); |
| 773 | |
Georgios Pinitas | 0b192e8 | 2020-02-20 17:09:28 +0000 | [diff] [blame] | 774 | return RETURN_UNIQUE_PTR(func); |
Isabella Gottardi | 7234ed8 | 2018-11-27 08:51:10 +0000 | [diff] [blame] | 775 | } |
Isabella Gottardi | a7acb3c | 2019-01-08 13:48:44 +0000 | [diff] [blame] | 776 | |
| 777 | /** Create a backend detection post process layer function |
| 778 | * |
| 779 | * @tparam DetectionPostProcessLayerFunction Backend detection output function |
| 780 | * @tparam TargetInfo Target-specific information |
| 781 | * |
| 782 | * @param[in] node Node to create the backend function for |
| 783 | * |
| 784 | * @return Backend detection post process layer function |
| 785 | */ |
| 786 | template <typename DetectionPostProcessLayerFunction, typename TargetInfo> |
| 787 | std::unique_ptr<IFunction> create_detection_post_process_layer(DetectionPostProcessLayerNode &node) |
| 788 | { |
| 789 | validate_node<TargetInfo>(node, 3 /* expected inputs */, 4 /* expected outputs */); |
| 790 | |
| 791 | // Extract IO and info |
| 792 | typename TargetInfo::TensorType *input0 = get_backing_tensor<TargetInfo>(node.input(0)); |
| 793 | typename TargetInfo::TensorType *input1 = get_backing_tensor<TargetInfo>(node.input(1)); |
| 794 | typename TargetInfo::TensorType *input2 = get_backing_tensor<TargetInfo>(node.input(2)); |
| 795 | typename TargetInfo::TensorType *output0 = get_backing_tensor<TargetInfo>(node.output(0)); |
| 796 | typename TargetInfo::TensorType *output1 = get_backing_tensor<TargetInfo>(node.output(1)); |
| 797 | typename TargetInfo::TensorType *output2 = get_backing_tensor<TargetInfo>(node.output(2)); |
| 798 | typename TargetInfo::TensorType *output3 = get_backing_tensor<TargetInfo>(node.output(3)); |
| 799 | const DetectionPostProcessLayerInfo detect_info = node.detection_post_process_info(); |
| 800 | |
| 801 | ARM_COMPUTE_ERROR_ON(input0 == nullptr); |
| 802 | ARM_COMPUTE_ERROR_ON(input1 == nullptr); |
| 803 | ARM_COMPUTE_ERROR_ON(input2 == nullptr); |
| 804 | ARM_COMPUTE_ERROR_ON(output0 == nullptr); |
| 805 | ARM_COMPUTE_ERROR_ON(output1 == nullptr); |
| 806 | ARM_COMPUTE_ERROR_ON(output2 == nullptr); |
| 807 | ARM_COMPUTE_ERROR_ON(output3 == nullptr); |
| 808 | |
| 809 | // Create and configure function |
| 810 | auto func = support::cpp14::make_unique<DetectionPostProcessLayerFunction>(); |
| 811 | func->configure(input0, input1, input2, output0, output1, output2, output3, detect_info); |
| 812 | |
| 813 | // Log info |
| 814 | ARM_COMPUTE_LOG_GRAPH_INFO("Instantiated " |
| 815 | << node.name() |
| 816 | << " Type: " << node.type() |
| 817 | << " Target: " << TargetInfo::TargetType |
| 818 | << " Data Type: " << input0->info()->data_type() |
| 819 | << " Input0 shape: " << input0->info()->tensor_shape() |
| 820 | << " Input1 shape: " << input1->info()->tensor_shape() |
| 821 | << " Input2 shape: " << input2->info()->tensor_shape() |
| 822 | << " Output0 shape: " << output0->info()->tensor_shape() |
| 823 | << " Output1 shape: " << output1->info()->tensor_shape() |
| 824 | << " Output2 shape: " << output2->info()->tensor_shape() |
| 825 | << " Output3 shape: " << output3->info()->tensor_shape() |
| 826 | << " DetectionPostProcessLayer info: " << detect_info |
| 827 | << std::endl); |
| 828 | |
Georgios Pinitas | 0b192e8 | 2020-02-20 17:09:28 +0000 | [diff] [blame] | 829 | return RETURN_UNIQUE_PTR(func); |
Isabella Gottardi | a7acb3c | 2019-01-08 13:48:44 +0000 | [diff] [blame] | 830 | } |
| 831 | |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 832 | /** Create a backend element-wise operation layer function |
| 833 | * |
| 834 | * @tparam EltwiseFunctions Backend element-wise function |
| 835 | * @tparam TargetInfo Target-specific information |
| 836 | * |
| 837 | * @param[in] node Node to create the backend function for |
| 838 | * |
| 839 | * @return Backend element-wise operation layer function |
| 840 | */ |
| 841 | template <typename EltwiseFunctions, typename TargetInfo> |
| 842 | std::unique_ptr<IFunction> create_eltwise_layer(EltwiseLayerNode &node) |
| 843 | { |
| 844 | validate_node<TargetInfo>(node, 2 /* expected inputs */, 1 /* expected outputs */); |
| 845 | |
| 846 | // Extract IO and info |
| 847 | typename TargetInfo::TensorType *input1 = get_backing_tensor<TargetInfo>(node.input(0)); |
| 848 | typename TargetInfo::TensorType *input2 = get_backing_tensor<TargetInfo>(node.input(1)); |
| 849 | typename TargetInfo::TensorType *output = get_backing_tensor<TargetInfo>(node.output(0)); |
| 850 | const EltwiseOperation eltwise_op = node.eltwise_operation(); |
| 851 | const ConvertPolicy convert_policy = node.convert_policy(); |
Giorgio Arena | 8b2a7d3 | 2020-02-11 17:21:31 +0000 | [diff] [blame] | 852 | const ActivationLayerInfo act_info = node.fused_activation(); |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 853 | ARM_COMPUTE_ERROR_ON(input1 == nullptr); |
| 854 | ARM_COMPUTE_ERROR_ON(input2 == nullptr); |
| 855 | ARM_COMPUTE_ERROR_ON(output == nullptr); |
| 856 | |
| 857 | std::unique_ptr<IFunction> func = nullptr; |
| 858 | std::string func_name; |
Georgios Pinitas | e222055 | 2018-07-20 13:23:44 +0100 | [diff] [blame] | 859 | if(eltwise_op == EltwiseOperation::Add) |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 860 | { |
| 861 | std::tie(func, func_name) = create_named_function<typename EltwiseFunctions::Addition>( |
| 862 | std::string("ArithmeticAddition"), |
Giorgio Arena | 8b2a7d3 | 2020-02-11 17:21:31 +0000 | [diff] [blame] | 863 | input1, input2, output, convert_policy, act_info); |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 864 | } |
Georgios Pinitas | e222055 | 2018-07-20 13:23:44 +0100 | [diff] [blame] | 865 | else if(eltwise_op == EltwiseOperation::Sub) |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 866 | { |
| 867 | std::tie(func, func_name) = create_named_function<typename EltwiseFunctions::Subtraction>( |
| 868 | std::string("ArithmeticSubtraction"), |
Giorgio Arena | 8b2a7d3 | 2020-02-11 17:21:31 +0000 | [diff] [blame] | 869 | input1, input2, output, convert_policy, act_info); |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 870 | } |
Georgios Pinitas | e222055 | 2018-07-20 13:23:44 +0100 | [diff] [blame] | 871 | else if(eltwise_op == EltwiseOperation::Mul) |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 872 | { |
| 873 | std::tie(func, func_name) = create_named_function<typename EltwiseFunctions::Multiplication>( |
| 874 | std::string("PixelWiseMultiplication"), |
Giorgio Arena | 8b2a7d3 | 2020-02-11 17:21:31 +0000 | [diff] [blame] | 875 | input1, input2, output, 1.f, convert_policy, node.rounding_policy(), act_info); |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 876 | } |
thecha01 | f8e3584 | 2020-07-28 17:28:17 +0100 | [diff] [blame] | 877 | else if(eltwise_op == EltwiseOperation::Max) |
| 878 | { |
| 879 | std::tie(func, func_name) = create_named_function<typename EltwiseFunctions::Maximum>( |
| 880 | std::string("ElementwiseMaximum"), |
| 881 | input1, input2, output, act_info); |
| 882 | } |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 883 | else |
| 884 | { |
| 885 | ARM_COMPUTE_ERROR("Unsupported element-wise operation!"); |
| 886 | } |
| 887 | |
| 888 | // Log info |
Pablo Tello | 3252143 | 2018-11-15 14:43:10 +0000 | [diff] [blame] | 889 | ARM_COMPUTE_LOG_GRAPH_INFO("Instantiated " |
| 890 | << node.name() |
| 891 | << " Type: " << node.type() |
| 892 | << " Target: " << TargetInfo::TargetType |
| 893 | << " Operation: " << func_name |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 894 | << " Data Type: " << input1->info()->data_type() |
Pablo Tello | 3252143 | 2018-11-15 14:43:10 +0000 | [diff] [blame] | 895 | << " Shape: " << input1->info()->tensor_shape() |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 896 | << std::endl); |
| 897 | |
Georgios Pinitas | 0b192e8 | 2020-02-20 17:09:28 +0000 | [diff] [blame] | 898 | return RETURN_UNIQUE_PTR(func); |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 899 | } |
| 900 | |
Sheri Zhang | 16dddd2 | 2020-05-27 15:03:48 +0100 | [diff] [blame] | 901 | /** Create a backend unary element-wise operation layer function |
| 902 | * |
| 903 | * @tparam UnaryEltwiseFunctions Backend unary element-wise function |
| 904 | * @tparam TargetInfo Target-specific information |
| 905 | * |
| 906 | * @param[in] node Node to create the backend function for |
| 907 | * |
| 908 | * @return Backend unary element-wise operation layer function |
| 909 | */ |
| 910 | template <typename UnaryEltwiseFunctions, typename TargetInfo> |
| 911 | std::unique_ptr<IFunction> create_unary_eltwise_layer(UnaryEltwiseLayerNode &node) |
| 912 | { |
| 913 | validate_node<TargetInfo>(node, 1 /* expected inputs */, 1 /* expected outputs */); |
| 914 | |
| 915 | // Extract IO and info |
| 916 | typename TargetInfo::TensorType *input = get_backing_tensor<TargetInfo>(node.input(0)); |
| 917 | typename TargetInfo::TensorType *output = get_backing_tensor<TargetInfo>(node.output(0)); |
| 918 | const UnaryEltwiseOperation eltwise_op = node.eltwise_descriptor().op; |
| 919 | |
| 920 | ARM_COMPUTE_ERROR_ON(input == nullptr); |
| 921 | ARM_COMPUTE_ERROR_ON(output == nullptr); |
| 922 | |
| 923 | std::unique_ptr<IFunction> func = nullptr; |
| 924 | std::string func_name; |
| 925 | if(eltwise_op == UnaryEltwiseOperation::Exp) |
| 926 | { |
| 927 | std::tie(func, func_name) = create_named_function<typename UnaryEltwiseFunctions::Exp>( |
| 928 | std::string("Exp"), |
| 929 | input, output); |
| 930 | } |
| 931 | else |
| 932 | { |
| 933 | ARM_COMPUTE_ERROR("Unsupported unary element-wise operation!"); |
| 934 | } |
| 935 | |
| 936 | // Log info |
| 937 | ARM_COMPUTE_LOG_GRAPH_INFO("Instantiated " |
| 938 | << node.name() |
| 939 | << " Type: " << node.type() |
| 940 | << " Target: " << TargetInfo::TargetType |
| 941 | << " Operation: " << func_name |
| 942 | << " Data Type: " << input->info()->data_type() |
| 943 | << " Shape: " << input->info()->tensor_shape() |
| 944 | << std::endl); |
| 945 | |
| 946 | return RETURN_UNIQUE_PTR(func); |
| 947 | } |
| 948 | |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 949 | /** Create a backend flatten layer function |
| 950 | * |
| 951 | * @tparam FlattenLayerFunction Backend flatten function |
| 952 | * @tparam TargetInfo Target-specific information |
| 953 | * |
| 954 | * @param[in] node Node to create the backend function for |
| 955 | * |
| 956 | * @return Backend flatten layer function |
| 957 | */ |
| 958 | template <typename FlattenLayerFunction, typename TargetInfo> |
| 959 | std::unique_ptr<IFunction> create_flatten_layer(FlattenLayerNode &node) |
| 960 | { |
| 961 | validate_node<TargetInfo>(node, 1 /* expected inputs */, 1 /* expected outputs */); |
| 962 | |
| 963 | // Extract IO and info |
| 964 | typename TargetInfo::TensorType *input = get_backing_tensor<TargetInfo>(node.input(0)); |
| 965 | typename TargetInfo::TensorType *output = get_backing_tensor<TargetInfo>(node.output(0)); |
| 966 | |
Georgios Pinitas | e222055 | 2018-07-20 13:23:44 +0100 | [diff] [blame] | 967 | ARM_COMPUTE_ERROR_ON(input == nullptr); |
| 968 | ARM_COMPUTE_ERROR_ON(output == nullptr); |
| 969 | |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 970 | // Create and configure function |
| 971 | auto func = support::cpp14::make_unique<FlattenLayerFunction>(); |
| 972 | func->configure(input, output); |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 973 | |
| 974 | // Log info |
Pablo Tello | 3252143 | 2018-11-15 14:43:10 +0000 | [diff] [blame] | 975 | ARM_COMPUTE_LOG_GRAPH_INFO("Instantiated " |
| 976 | << node.name() |
| 977 | << " Type: " << node.type() |
| 978 | << " Target: " << TargetInfo::TargetType |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 979 | << " Data Type: " << input->info()->data_type() |
| 980 | << " Input shape: " << input->info()->tensor_shape() |
| 981 | << " Output shape: " << output->info()->tensor_shape() |
| 982 | << std::endl); |
| 983 | |
Georgios Pinitas | 0b192e8 | 2020-02-20 17:09:28 +0000 | [diff] [blame] | 984 | return RETURN_UNIQUE_PTR(func); |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 985 | } |
| 986 | |
| 987 | /** Create a backend fully connected layer function |
| 988 | * |
| 989 | * @tparam FullyConnectedLayerFunction Backend fully-connected function |
| 990 | * @tparam TargetInfo Target-specific information |
| 991 | * |
| 992 | * @param[in] node Node to create the backend function for |
| 993 | * @param[in] ctx Graph context |
| 994 | * |
| 995 | * @return Backend fully connected layer function |
| 996 | */ |
| 997 | template <typename FullyConnectedLayerFunction, typename TargetInfo> |
| 998 | std::unique_ptr<IFunction> create_fully_connected_layer(FullyConnectedLayerNode &node, GraphContext &ctx) |
| 999 | { |
| 1000 | validate_node<TargetInfo>(node, 3 /* expected inputs */, 1 /* expected outputs */); |
| 1001 | |
| 1002 | // Extract IO and info |
| 1003 | typename TargetInfo::TensorType *input = get_backing_tensor<TargetInfo>(node.input(0)); |
| 1004 | typename TargetInfo::TensorType *weights = get_backing_tensor<TargetInfo>(node.input(1)); |
| 1005 | typename TargetInfo::TensorType *biases = get_backing_tensor<TargetInfo>(node.input(2)); |
| 1006 | typename TargetInfo::TensorType *output = get_backing_tensor<TargetInfo>(node.output(0)); |
Georgios Pinitas | 7d66a8e | 2018-07-17 12:28:42 +0100 | [diff] [blame] | 1007 | const FullyConnectedLayerInfo fc_info = node.info(); |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 1008 | |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 1009 | ARM_COMPUTE_ERROR_ON(input == nullptr); |
| 1010 | ARM_COMPUTE_ERROR_ON(weights == nullptr); |
| 1011 | ARM_COMPUTE_ERROR_ON(output == nullptr); |
| 1012 | |
Georgios Pinitas | e222055 | 2018-07-20 13:23:44 +0100 | [diff] [blame] | 1013 | // Create and configure function |
Michalis Spyrou | 1a569a3 | 2019-09-10 17:20:34 +0100 | [diff] [blame] | 1014 | auto wm = get_weights_manager(ctx, TargetInfo::TargetType); |
| 1015 | auto mm = get_memory_manager(ctx, TargetInfo::TargetType); |
| 1016 | auto func = support::cpp14::make_unique<FullyConnectedLayerFunction>(mm, wm.get()); |
Georgios Pinitas | e222055 | 2018-07-20 13:23:44 +0100 | [diff] [blame] | 1017 | func->configure(input, weights, biases, output, fc_info); |
| 1018 | |
Georgios Pinitas | fd7e853 | 2018-09-07 10:51:27 +0100 | [diff] [blame] | 1019 | const bool is_quantized = is_data_type_quantized_asymmetric(input->info()->data_type()); |
| 1020 | |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 1021 | // Log info |
Georgios Pinitas | fd7e853 | 2018-09-07 10:51:27 +0100 | [diff] [blame] | 1022 | std::ostringstream qss; |
| 1023 | if(is_quantized) |
| 1024 | { |
| 1025 | qss << " Input QuantInfo: " << input->info()->quantization_info() |
| 1026 | << " Weights QuantInfo: " << weights->info()->quantization_info() |
| 1027 | << " Output QuantInfo: " << output->info()->quantization_info(); |
| 1028 | } |
Pablo Tello | 3252143 | 2018-11-15 14:43:10 +0000 | [diff] [blame] | 1029 | ARM_COMPUTE_LOG_GRAPH_INFO("Instantiated " |
| 1030 | << node.name() |
| 1031 | << " Type: " << node.type() |
| 1032 | << " Target: " << TargetInfo::TargetType |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 1033 | << " Data Type: " << input->info()->data_type() |
Georgios Pinitas | fd7e853 | 2018-09-07 10:51:27 +0100 | [diff] [blame] | 1034 | << qss.str() |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 1035 | << " Input shape: " << input->info()->tensor_shape() |
| 1036 | << " Weights shape: " << weights->info()->tensor_shape() |
| 1037 | << " Output shape: " << output->info()->tensor_shape() |
| 1038 | << std::endl); |
| 1039 | |
Georgios Pinitas | 0b192e8 | 2020-02-20 17:09:28 +0000 | [diff] [blame] | 1040 | return RETURN_UNIQUE_PTR(func); |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 1041 | } |
| 1042 | |
Manuel Bottini | 5209be5 | 2019-02-13 16:34:56 +0000 | [diff] [blame] | 1043 | /** Create a backend generate proposals layer function |
| 1044 | * |
| 1045 | * @tparam GenerateProposalsLayerFunction Backend generate proposals function |
| 1046 | * @tparam TargetInfo Target-specific information |
| 1047 | * |
| 1048 | * @param[in] node Node to create the backend function for |
| 1049 | * @param[in] ctx Graph context |
| 1050 | * |
| 1051 | * @return Backend generate proposals layer function |
| 1052 | */ |
| 1053 | template <typename GenerateProposalsLayerFunction, typename TargetInfo> |
| 1054 | std::unique_ptr<IFunction> create_generate_proposals_layer(GenerateProposalsLayerNode &node, GraphContext &ctx) |
| 1055 | { |
| 1056 | validate_node<TargetInfo>(node, 3 /* expected inputs */, 3 /* expected outputs */); |
| 1057 | |
| 1058 | // Extract IO and info |
| 1059 | typename TargetInfo::TensorType *scores = get_backing_tensor<TargetInfo>(node.input(0)); |
| 1060 | typename TargetInfo::TensorType *deltas = get_backing_tensor<TargetInfo>(node.input(1)); |
| 1061 | typename TargetInfo::TensorType *anchors = get_backing_tensor<TargetInfo>(node.input(2)); |
| 1062 | typename TargetInfo::TensorType *proposals = get_backing_tensor<TargetInfo>(node.output(0)); |
| 1063 | typename TargetInfo::TensorType *scores_out = get_backing_tensor<TargetInfo>(node.output(1)); |
| 1064 | typename TargetInfo::TensorType *num_valid_proposals = get_backing_tensor<TargetInfo>(node.output(2)); |
| 1065 | const GenerateProposalsInfo info = node.info(); |
| 1066 | |
| 1067 | ARM_COMPUTE_ERROR_ON(scores == nullptr); |
| 1068 | ARM_COMPUTE_ERROR_ON(deltas == nullptr); |
| 1069 | ARM_COMPUTE_ERROR_ON(anchors == nullptr); |
| 1070 | ARM_COMPUTE_ERROR_ON(proposals == nullptr); |
| 1071 | ARM_COMPUTE_ERROR_ON(scores_out == nullptr); |
| 1072 | |
| 1073 | // Create and configure function |
| 1074 | auto func = support::cpp14::make_unique<GenerateProposalsLayerFunction>(get_memory_manager(ctx, TargetInfo::TargetType)); |
| 1075 | func->configure(scores, deltas, anchors, proposals, scores_out, num_valid_proposals, info); |
| 1076 | |
| 1077 | // Log info |
| 1078 | ARM_COMPUTE_LOG_GRAPH_INFO("Instantiated " << node.type() |
| 1079 | << " Target " << TargetInfo::TargetType |
| 1080 | << " Data Type: " << scores->info()->data_type() |
| 1081 | << " Scores shape: " << scores->info()->tensor_shape() |
| 1082 | << " Deltas shape: " << deltas->info()->tensor_shape() |
| 1083 | << " Anchors shape: " << anchors->info()->tensor_shape() |
| 1084 | << " Proposals shape: " << proposals->info()->tensor_shape() |
| 1085 | << " Num valid proposals shape: " << num_valid_proposals->info()->tensor_shape() |
| 1086 | << " Scores Out shape: " << scores_out->info()->tensor_shape() |
| 1087 | << std::endl); |
| 1088 | |
Georgios Pinitas | 0b192e8 | 2020-02-20 17:09:28 +0000 | [diff] [blame] | 1089 | return RETURN_UNIQUE_PTR(func); |
Manuel Bottini | 5209be5 | 2019-02-13 16:34:56 +0000 | [diff] [blame] | 1090 | } |
| 1091 | |
thecha01 | 3603aff | 2020-09-01 14:52:38 +0100 | [diff] [blame] | 1092 | /** Create a backend l2 normalization layer function |
| 1093 | * |
| 1094 | * @tparam NormalizationLayerFunction Backend normalization function |
| 1095 | * @tparam TargetInfo Target-specific information |
| 1096 | * |
| 1097 | * @param[in] node Node to create the backend function for |
| 1098 | * @param[in] ctx Graph context |
| 1099 | * |
| 1100 | * @return Backend normalization layer function |
| 1101 | */ |
| 1102 | template <typename L2NormalizeLayerFunction, typename TargetInfo> |
| 1103 | std::unique_ptr<IFunction> create_l2_normalize_layer(L2NormalizeLayerNode &node, GraphContext &ctx) |
| 1104 | { |
| 1105 | validate_node<TargetInfo>(node, 1 /* expected inputs */, 1 /* expected outputs */); |
| 1106 | |
| 1107 | // Extract IO and info |
| 1108 | typename TargetInfo::TensorType *input = get_backing_tensor<TargetInfo>(node.input(0)); |
| 1109 | typename TargetInfo::TensorType *output = get_backing_tensor<TargetInfo>(node.output(0)); |
| 1110 | int axis = node.axis(); |
| 1111 | float epsilon = node.epsilon(); |
| 1112 | |
| 1113 | ARM_COMPUTE_ERROR_ON(input == nullptr); |
| 1114 | ARM_COMPUTE_ERROR_ON(output == nullptr); |
| 1115 | |
| 1116 | // Create and configure function |
| 1117 | auto mm = get_memory_manager(ctx, TargetInfo::TargetType); |
| 1118 | auto func = support::cpp14::make_unique<L2NormalizeLayerFunction>(mm); |
| 1119 | func->configure(input, output, axis, epsilon); |
| 1120 | |
| 1121 | // Log info |
| 1122 | ARM_COMPUTE_LOG_GRAPH_INFO("Instantiated " |
| 1123 | << node.name() |
| 1124 | << " Type: " << node.type() |
| 1125 | << " Target: " << TargetInfo::TargetType |
| 1126 | << " Data Type: " << input->info()->data_type() |
| 1127 | << " Input shape: " << input->info()->tensor_shape() |
| 1128 | << " Output shape: " << output->info()->tensor_shape() |
| 1129 | << " Axis: " << axis |
| 1130 | << " Epsilon: " << epsilon |
| 1131 | << std::endl); |
| 1132 | |
| 1133 | return RETURN_UNIQUE_PTR(func); |
| 1134 | } |
| 1135 | |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 1136 | /** Create a backend normalization layer function |
| 1137 | * |
| 1138 | * @tparam NormalizationLayerFunction Backend normalization function |
| 1139 | * @tparam TargetInfo Target-specific information |
| 1140 | * |
| 1141 | * @param[in] node Node to create the backend function for |
| 1142 | * @param[in] ctx Graph context |
| 1143 | * |
| 1144 | * @return Backend normalization layer function |
| 1145 | */ |
| 1146 | template <typename NormalizationLayerFunction, typename TargetInfo> |
| 1147 | std::unique_ptr<IFunction> create_normalization_layer(NormalizationLayerNode &node, GraphContext &ctx) |
| 1148 | { |
| 1149 | ARM_COMPUTE_UNUSED(ctx); |
| 1150 | |
| 1151 | validate_node<TargetInfo>(node, 1 /* expected inputs */, 1 /* expected outputs */); |
| 1152 | |
| 1153 | // Extract IO and info |
| 1154 | typename TargetInfo::TensorType *input = get_backing_tensor<TargetInfo>(node.input(0)); |
| 1155 | typename TargetInfo::TensorType *output = get_backing_tensor<TargetInfo>(node.output(0)); |
| 1156 | const NormalizationLayerInfo norm_info = node.normalization_info(); |
| 1157 | ARM_COMPUTE_ERROR_ON(input == nullptr); |
| 1158 | ARM_COMPUTE_ERROR_ON(output == nullptr); |
| 1159 | |
| 1160 | // Create and configure function |
| 1161 | auto func = support::cpp14::make_unique<NormalizationLayerFunction>(); |
| 1162 | func->configure(input, output, norm_info); |
| 1163 | |
| 1164 | // Log info |
Pablo Tello | 3252143 | 2018-11-15 14:43:10 +0000 | [diff] [blame] | 1165 | ARM_COMPUTE_LOG_GRAPH_INFO("Instantiated " |
| 1166 | << node.name() |
| 1167 | << " Type: " << node.type() |
| 1168 | << " Target: " << TargetInfo::TargetType |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 1169 | << " Data Type: " << input->info()->data_type() |
| 1170 | << " Input shape: " << input->info()->tensor_shape() |
| 1171 | << " Output shape: " << output->info()->tensor_shape() |
| 1172 | << " Normalization info: " << norm_info.type() |
| 1173 | << std::endl); |
| 1174 | |
Georgios Pinitas | 0b192e8 | 2020-02-20 17:09:28 +0000 | [diff] [blame] | 1175 | return RETURN_UNIQUE_PTR(func); |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 1176 | } |
| 1177 | |
Michele Di Giorgio | 555d110 | 2018-09-12 13:51:59 +0100 | [diff] [blame] | 1178 | /** Create a backend normalize planar YUV layer function |
| 1179 | * |
| 1180 | * @tparam NormalizePlanarYUVLayerFunction Backend normalize planar YUV function |
| 1181 | * @tparam TargetInfo Target-specific information |
| 1182 | * |
| 1183 | * @param[in] node Node to create the backend function for |
| 1184 | * |
| 1185 | * @return Backend normalize plnar YUV layer function |
| 1186 | */ |
| 1187 | template <typename NormalizePlanarYUVLayerFunction, typename TargetInfo> |
| 1188 | std::unique_ptr<IFunction> create_normalize_planar_yuv_layer(NormalizePlanarYUVLayerNode &node) |
| 1189 | { |
| 1190 | validate_node<TargetInfo>(node, 3 /* expected inputs */, 1 /* expected outputs */); |
| 1191 | |
| 1192 | // Extract IO and info |
| 1193 | typename TargetInfo::TensorType *input = get_backing_tensor<TargetInfo>(node.input(0)); |
| 1194 | typename TargetInfo::TensorType *mean = get_backing_tensor<TargetInfo>(node.input(1)); |
| 1195 | typename TargetInfo::TensorType *std = get_backing_tensor<TargetInfo>(node.input(2)); |
| 1196 | typename TargetInfo::TensorType *output = get_backing_tensor<TargetInfo>(node.output(0)); |
| 1197 | ARM_COMPUTE_ERROR_ON(input == nullptr); |
| 1198 | ARM_COMPUTE_ERROR_ON(mean == nullptr); |
| 1199 | ARM_COMPUTE_ERROR_ON(std == nullptr); |
| 1200 | ARM_COMPUTE_ERROR_ON(output == nullptr); |
| 1201 | |
| 1202 | // Create and configure function |
| 1203 | auto func = support::cpp14::make_unique<NormalizePlanarYUVLayerFunction>(); |
| 1204 | func->configure(input, output, mean, std); |
| 1205 | |
| 1206 | // Log info |
Pablo Tello | 3252143 | 2018-11-15 14:43:10 +0000 | [diff] [blame] | 1207 | ARM_COMPUTE_LOG_GRAPH_INFO("Instantiated " |
| 1208 | << node.name() |
| 1209 | << " Type: " << node.type() |
| 1210 | << " Target: " << TargetInfo::TargetType |
Michele Di Giorgio | 555d110 | 2018-09-12 13:51:59 +0100 | [diff] [blame] | 1211 | << " Data Type: " << input->info()->data_type() |
| 1212 | << " Shape: " << input->info()->tensor_shape() |
| 1213 | << std::endl); |
| 1214 | |
Georgios Pinitas | 0b192e8 | 2020-02-20 17:09:28 +0000 | [diff] [blame] | 1215 | return RETURN_UNIQUE_PTR(func); |
Michele Di Giorgio | 555d110 | 2018-09-12 13:51:59 +0100 | [diff] [blame] | 1216 | } |
| 1217 | |
Michele Di Giorgio | 4bb1733 | 2018-09-26 13:56:51 +0100 | [diff] [blame] | 1218 | /** Create a backend pad layer function |
| 1219 | * |
| 1220 | * @tparam PadLayerFunction Backend pad function |
| 1221 | * @tparam TargetInfo Target-specific information |
| 1222 | * |
| 1223 | * @param[in] node Node to create the backend function for |
| 1224 | * |
| 1225 | * @return Backend pad layer function |
| 1226 | */ |
| 1227 | template <typename PadLayerFunction, typename TargetInfo> |
| 1228 | std::unique_ptr<IFunction> create_pad_layer(PadLayerNode &node) |
| 1229 | { |
| 1230 | validate_node<TargetInfo>(node, 1 /* expected inputs */, 1 /* expected outputs */); |
| 1231 | |
| 1232 | // Extract IO and info |
Georgios Pinitas | 102b0ce | 2020-02-13 17:59:09 +0000 | [diff] [blame] | 1233 | typename TargetInfo::TensorType *input = get_backing_tensor<TargetInfo>(node.input(0)); |
| 1234 | typename TargetInfo::TensorType *output = get_backing_tensor<TargetInfo>(node.output(0)); |
| 1235 | const PaddingList &padding = node.padding(); |
| 1236 | const PixelValue pad_value = node.pad_value(); |
Michele Di Giorgio | 4bb1733 | 2018-09-26 13:56:51 +0100 | [diff] [blame] | 1237 | ARM_COMPUTE_ERROR_ON(input == nullptr); |
| 1238 | ARM_COMPUTE_ERROR_ON(output == nullptr); |
| 1239 | |
| 1240 | // Create and configure function |
| 1241 | auto func = support::cpp14::make_unique<PadLayerFunction>(); |
Georgios Pinitas | 102b0ce | 2020-02-13 17:59:09 +0000 | [diff] [blame] | 1242 | func->configure(input, output, padding, pad_value); |
Michele Di Giorgio | 4bb1733 | 2018-09-26 13:56:51 +0100 | [diff] [blame] | 1243 | |
| 1244 | // Log info |
Pablo Tello | 3252143 | 2018-11-15 14:43:10 +0000 | [diff] [blame] | 1245 | ARM_COMPUTE_LOG_GRAPH_INFO("Instantiated " |
| 1246 | << node.name() |
| 1247 | << " Type: " << node.type() |
| 1248 | << " Target: " << TargetInfo::TargetType |
Michele Di Giorgio | 4bb1733 | 2018-09-26 13:56:51 +0100 | [diff] [blame] | 1249 | << " Data Type: " << input->info()->data_type() |
| 1250 | << " Input shape: " << input->info()->tensor_shape() |
| 1251 | << " Output shape: " << output->info()->tensor_shape() |
| 1252 | << std::endl); |
| 1253 | |
Georgios Pinitas | 0b192e8 | 2020-02-20 17:09:28 +0000 | [diff] [blame] | 1254 | return RETURN_UNIQUE_PTR(func); |
Michele Di Giorgio | 4bb1733 | 2018-09-26 13:56:51 +0100 | [diff] [blame] | 1255 | } |
| 1256 | |
Georgios Pinitas | 57c4824 | 2018-08-02 13:41:49 +0100 | [diff] [blame] | 1257 | /** Create a backend permute layer function |
| 1258 | * |
| 1259 | * @tparam PermuteLayerFunction Backend permute function |
| 1260 | * @tparam TargetInfo Target-specific information |
| 1261 | * |
| 1262 | * @param[in] node Node to create the backend function for |
| 1263 | * |
| 1264 | * @return Backend permute layer function |
| 1265 | */ |
| 1266 | template <typename PermuteLayerFunction, typename TargetInfo> |
| 1267 | std::unique_ptr<IFunction> create_permute_layer(PermuteLayerNode &node) |
| 1268 | { |
| 1269 | validate_node<TargetInfo>(node, 1 /* expected inputs */, 1 /* expected outputs */); |
| 1270 | |
| 1271 | // Extract IO and info |
| 1272 | typename TargetInfo::TensorType *input = get_backing_tensor<TargetInfo>(node.input(0)); |
| 1273 | typename TargetInfo::TensorType *output = get_backing_tensor<TargetInfo>(node.output(0)); |
| 1274 | const PermutationVector &perm = node.permutation_vector(); |
| 1275 | ARM_COMPUTE_ERROR_ON(input == nullptr); |
| 1276 | ARM_COMPUTE_ERROR_ON(output == nullptr); |
| 1277 | |
| 1278 | // Create and configure function |
| 1279 | auto func = support::cpp14::make_unique<PermuteLayerFunction>(); |
| 1280 | func->configure(input, output, perm); |
| 1281 | |
| 1282 | // Log info |
Pablo Tello | 3252143 | 2018-11-15 14:43:10 +0000 | [diff] [blame] | 1283 | ARM_COMPUTE_LOG_GRAPH_INFO("Instantiated " |
| 1284 | << node.name() |
| 1285 | << " Type: " << node.type() |
| 1286 | << " Target: " << TargetInfo::TargetType |
Georgios Pinitas | 57c4824 | 2018-08-02 13:41:49 +0100 | [diff] [blame] | 1287 | << " Data Type: " << input->info()->data_type() |
| 1288 | << " Input shape: " << input->info()->tensor_shape() |
| 1289 | << " Output shape: " << output->info()->tensor_shape() |
| 1290 | << " Permutation vector: " << perm |
| 1291 | << std::endl); |
| 1292 | |
Georgios Pinitas | 0b192e8 | 2020-02-20 17:09:28 +0000 | [diff] [blame] | 1293 | return RETURN_UNIQUE_PTR(func); |
Georgios Pinitas | 57c4824 | 2018-08-02 13:41:49 +0100 | [diff] [blame] | 1294 | } |
| 1295 | |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 1296 | /** Create a backend pooling layer function |
| 1297 | * |
| 1298 | * @tparam PoolingLayerFunction Backend pooling function |
| 1299 | * @tparam TargetInfo Target-specific information |
| 1300 | * |
| 1301 | * @param[in] node Node to create the backend function for |
| 1302 | * |
| 1303 | * @return Backend pooling layer function |
| 1304 | */ |
| 1305 | template <typename PoolingLayerFunction, typename TargetInfo> |
| 1306 | std::unique_ptr<IFunction> create_pooling_layer(PoolingLayerNode &node) |
| 1307 | { |
| 1308 | validate_node<TargetInfo>(node, 1 /* expected inputs */, 1 /* expected outputs */); |
| 1309 | |
| 1310 | // Extract IO and info |
| 1311 | typename TargetInfo::TensorType *input = get_backing_tensor<TargetInfo>(node.input(0)); |
| 1312 | typename TargetInfo::TensorType *output = get_backing_tensor<TargetInfo>(node.output(0)); |
| 1313 | const PoolingLayerInfo pool_info = node.pooling_info(); |
| 1314 | ARM_COMPUTE_ERROR_ON(input == nullptr); |
| 1315 | ARM_COMPUTE_ERROR_ON(output == nullptr); |
| 1316 | |
| 1317 | // Create and configure function |
| 1318 | auto func = support::cpp14::make_unique<PoolingLayerFunction>(); |
| 1319 | func->configure(input, output, pool_info); |
| 1320 | |
| 1321 | // Log info |
Pablo Tello | 3252143 | 2018-11-15 14:43:10 +0000 | [diff] [blame] | 1322 | ARM_COMPUTE_LOG_GRAPH_INFO("Instantiated " |
| 1323 | << node.name() |
| 1324 | << " Type: " << node.type() |
| 1325 | << " Target: " << TargetInfo::TargetType |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 1326 | << " Data Type: " << input->info()->data_type() |
| 1327 | << " Input shape: " << input->info()->tensor_shape() |
| 1328 | << " Output shape: " << output->info()->tensor_shape() |
Sang-Hoon Park | 0cb3da6 | 2020-01-15 12:39:56 +0000 | [diff] [blame] | 1329 | << " Pooling info: " << pool_info.pool_type |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 1330 | << std::endl); |
| 1331 | |
Georgios Pinitas | 0b192e8 | 2020-02-20 17:09:28 +0000 | [diff] [blame] | 1332 | return RETURN_UNIQUE_PTR(func); |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 1333 | } |
| 1334 | |
Georgios Pinitas | f8c4749 | 2020-02-04 17:39:59 +0000 | [diff] [blame] | 1335 | /** Create a backend PRelu layer function |
| 1336 | * |
| 1337 | * @tparam PReluFunction Backend PRelu function |
| 1338 | * @tparam TargetInfo Target-specific information |
| 1339 | * |
| 1340 | * @param[in] node Node to create the backend function for |
| 1341 | * |
| 1342 | * @return Backend PRelu layer function |
| 1343 | */ |
| 1344 | template <typename PReluFunction, typename TargetInfo> |
| 1345 | std::unique_ptr<IFunction> create_prelu_layer(PReluLayerNode &node) |
| 1346 | { |
| 1347 | validate_node<TargetInfo>(node, 2 /* expected inputs */, 1 /* expected outputs */); |
| 1348 | |
| 1349 | // Extract IO and info |
| 1350 | typename TargetInfo::TensorType *input = get_backing_tensor<TargetInfo>(node.input(0)); |
| 1351 | typename TargetInfo::TensorType *alpha = get_backing_tensor<TargetInfo>(node.input(1)); |
| 1352 | typename TargetInfo::TensorType *output = get_backing_tensor<TargetInfo>(node.output(0)); |
| 1353 | ARM_COMPUTE_ERROR_ON(input == nullptr || alpha == nullptr); |
| 1354 | ARM_COMPUTE_ERROR_ON(output == nullptr); |
| 1355 | |
| 1356 | // Create and configure function |
| 1357 | auto func = support::cpp14::make_unique<PReluFunction>(); |
| 1358 | func->configure(input, alpha, output); |
| 1359 | |
| 1360 | // Log info |
| 1361 | ARM_COMPUTE_LOG_GRAPH_INFO("Instantiated " |
| 1362 | << node.name() |
| 1363 | << " Type: " << node.type() |
| 1364 | << " Target: " << TargetInfo::TargetType |
| 1365 | << " Data Type: " << input->info()->data_type() |
| 1366 | << " Input shape: " << input->info()->tensor_shape() |
| 1367 | << " Output shape: " << output->info()->tensor_shape() |
| 1368 | << std::endl); |
| 1369 | |
Georgios Pinitas | 0b192e8 | 2020-02-20 17:09:28 +0000 | [diff] [blame] | 1370 | return RETURN_UNIQUE_PTR(func); |
Georgios Pinitas | f8c4749 | 2020-02-04 17:39:59 +0000 | [diff] [blame] | 1371 | } |
| 1372 | |
Giorgio Arena | 6e9d0e0 | 2020-01-03 15:02:04 +0000 | [diff] [blame] | 1373 | /** Create a backend print layer function |
| 1374 | * |
| 1375 | * @tparam TargetInfo Target-specific information |
| 1376 | * |
| 1377 | * @param[in] node Node to create the backend function for |
| 1378 | * |
| 1379 | * @return Backend print layer function |
| 1380 | */ |
| 1381 | template <typename TargetInfo> |
| 1382 | std::unique_ptr<IFunction> create_print_layer(PrintLayerNode &node) |
| 1383 | { |
| 1384 | validate_node<TargetInfo>(node, 1 /* expected inputs */, 1 /* expected outputs */); |
| 1385 | |
| 1386 | typename TargetInfo::TensorType *input = get_backing_tensor<TargetInfo>(node.input(0)); |
| 1387 | ARM_COMPUTE_ERROR_ON(input == nullptr); |
| 1388 | ARM_COMPUTE_UNUSED(input); |
| 1389 | |
| 1390 | // Log info |
| 1391 | ARM_COMPUTE_LOG_GRAPH_INFO("Instantiated " |
| 1392 | << node.name() |
| 1393 | << " Type: " << node.type() |
| 1394 | << " Target: " << TargetInfo::TargetType |
| 1395 | << " Data Type: " << input->info()->data_type() |
| 1396 | << " Input shape: " << input->info()->tensor_shape() |
| 1397 | << std::endl); |
| 1398 | |
| 1399 | return nullptr; |
| 1400 | } |
| 1401 | |
Pablo Tello | 3252143 | 2018-11-15 14:43:10 +0000 | [diff] [blame] | 1402 | /** Create a backend priorbox layer function |
| 1403 | * |
| 1404 | * @tparam PriorBoxLayerFunction Backend priorbox function |
| 1405 | * @tparam TargetInfo Target-specific information |
| 1406 | * |
| 1407 | * @param[in] node Node to create the backend function for |
| 1408 | * |
| 1409 | * @return Backend priorbox layer function |
| 1410 | */ |
| 1411 | template <typename PriorBoxLayerFunction, typename TargetInfo> |
| 1412 | std::unique_ptr<IFunction> create_priorbox_layer(PriorBoxLayerNode &node) |
| 1413 | { |
| 1414 | validate_node<TargetInfo>(node, 2 /* expected inputs */, 1 /* expected outputs */); |
| 1415 | |
| 1416 | // Extract IO and info |
| 1417 | typename TargetInfo::TensorType *input0 = get_backing_tensor<TargetInfo>(node.input(0)); |
| 1418 | typename TargetInfo::TensorType *input1 = get_backing_tensor<TargetInfo>(node.input(1)); |
| 1419 | typename TargetInfo::TensorType *output = get_backing_tensor<TargetInfo>(node.output(0)); |
| 1420 | const PriorBoxLayerInfo prior_info = node.priorbox_info(); |
| 1421 | ARM_COMPUTE_ERROR_ON(input0 == nullptr); |
| 1422 | ARM_COMPUTE_ERROR_ON(input1 == nullptr); |
| 1423 | ARM_COMPUTE_ERROR_ON(output == nullptr); |
| 1424 | |
| 1425 | // Create and configure function |
| 1426 | auto func = support::cpp14::make_unique<PriorBoxLayerFunction>(); |
| 1427 | func->configure(input0, input1, output, prior_info); |
| 1428 | |
| 1429 | // Log info |
| 1430 | ARM_COMPUTE_LOG_GRAPH_INFO("Instantiated " |
| 1431 | << node.name() |
| 1432 | << " Type: " << node.type() |
| 1433 | << " Target: " << TargetInfo::TargetType |
| 1434 | << " Data Type: " << input0->info()->data_type() |
| 1435 | << " Input0 shape: " << input0->info()->tensor_shape() |
| 1436 | << " Input1 shape: " << input1->info()->tensor_shape() |
| 1437 | << " Output shape: " << output->info()->tensor_shape() |
| 1438 | << " PriorBoxLayer info: " << prior_info |
| 1439 | << std::endl); |
| 1440 | |
Georgios Pinitas | 0b192e8 | 2020-02-20 17:09:28 +0000 | [diff] [blame] | 1441 | return RETURN_UNIQUE_PTR(func); |
Pablo Tello | 3252143 | 2018-11-15 14:43:10 +0000 | [diff] [blame] | 1442 | } |
| 1443 | |
Isabella Gottardi | 3db1ba9 | 2019-05-17 12:35:20 +0100 | [diff] [blame] | 1444 | /** Create a backend quantization layer function |
| 1445 | * |
| 1446 | * @tparam QuantizationLayerFunction Backend quantization function |
| 1447 | * @tparam TargetInfo Target-specific information |
| 1448 | * |
| 1449 | * @param[in] node Node to create the backend function for |
| 1450 | * |
| 1451 | * @return Backend quantization layer function |
| 1452 | */ |
| 1453 | template <typename QuantizationLayerFunction, typename TargetInfo> |
| 1454 | std::unique_ptr<IFunction> create_quantization_layer(QuantizationLayerNode &node) |
| 1455 | { |
| 1456 | validate_node<TargetInfo>(node, 1 /* expected inputs */, 1 /* expected outputs */); |
| 1457 | |
| 1458 | // Extract IO and info |
| 1459 | typename TargetInfo::TensorType *input = get_backing_tensor<TargetInfo>(node.input(0)); |
| 1460 | typename TargetInfo::TensorType *output = get_backing_tensor<TargetInfo>(node.output(0)); |
| 1461 | ARM_COMPUTE_ERROR_ON(input == nullptr); |
| 1462 | ARM_COMPUTE_ERROR_ON(output == nullptr); |
| 1463 | |
| 1464 | // Create and configure function |
| 1465 | auto func = support::cpp14::make_unique<QuantizationLayerFunction>(); |
| 1466 | func->configure(input, output); |
| 1467 | |
| 1468 | // Log info |
| 1469 | ARM_COMPUTE_LOG_GRAPH_INFO("Instantiated " |
| 1470 | << node.name() |
| 1471 | << " Type: " << node.type() |
| 1472 | << " Target: " << TargetInfo::TargetType |
| 1473 | << " Data Type: " << input->info()->data_type() |
| 1474 | << " Input shape: " << input->info()->tensor_shape() |
| 1475 | << " Output shape: " << output->info()->tensor_shape() |
| 1476 | << std::endl); |
| 1477 | |
Georgios Pinitas | 0b192e8 | 2020-02-20 17:09:28 +0000 | [diff] [blame] | 1478 | return RETURN_UNIQUE_PTR(func); |
Isabella Gottardi | 3db1ba9 | 2019-05-17 12:35:20 +0100 | [diff] [blame] | 1479 | } |
| 1480 | |
thecha01 | d64444b | 2020-09-07 14:50:21 +0100 | [diff] [blame] | 1481 | /** Create a backend reduction operation layer function |
| 1482 | * |
| 1483 | * @tparam ReductionOperationFunction Backend reduction operation function |
| 1484 | * @tparam TargetInfo Target-specific information |
| 1485 | * |
| 1486 | * @param[in] node Node to create the backend function for |
| 1487 | * @param[in] ctx Graph context |
| 1488 | * |
| 1489 | * @return Backend reduction sum layer function |
| 1490 | */ |
| 1491 | template <typename ReductionOperationFunction, typename TargetInfo> |
| 1492 | std::unique_ptr<IFunction> create_reduction_operation_layer(ReductionLayerNode &node, GraphContext &ctx) |
| 1493 | { |
| 1494 | validate_node<TargetInfo>(node, 1 /* expected inputs */, 1 /* expected outputs */); |
| 1495 | |
| 1496 | // Extract IO and info |
| 1497 | typename TargetInfo::TensorType *input = get_backing_tensor<TargetInfo>(node.input(0)); |
| 1498 | typename TargetInfo::TensorType *output = get_backing_tensor<TargetInfo>(node.output(0)); |
| 1499 | ReductionOperation op = node.op(); |
| 1500 | int axis = node.axis(); |
| 1501 | bool keep_dims = node.keep_dims(); |
| 1502 | ARM_COMPUTE_ERROR_ON(input == nullptr); |
| 1503 | ARM_COMPUTE_ERROR_ON(output == nullptr); |
| 1504 | |
| 1505 | // Create and configure function |
| 1506 | auto func = support::cpp14::make_unique<ReductionOperationFunction>(get_memory_manager(ctx, TargetInfo::TargetType)); |
| 1507 | func->configure(input, output, axis, op, keep_dims); |
| 1508 | |
| 1509 | // Log info |
| 1510 | ARM_COMPUTE_LOG_GRAPH_INFO("Instantiated " |
| 1511 | << node.name() |
| 1512 | << " Type: " << node.type() |
| 1513 | << " Target: " << TargetInfo::TargetType |
| 1514 | << " Data Type: " << input->info()->data_type() |
| 1515 | << " Input shape: " << input->info()->tensor_shape() |
| 1516 | << " Output shape: " << output->info()->tensor_shape() |
| 1517 | << " Operation: " << op |
| 1518 | << " Axis: " << axis |
| 1519 | << " Keep dimensions:" << keep_dims |
| 1520 | << std::endl); |
| 1521 | |
| 1522 | return RETURN_UNIQUE_PTR(func); |
| 1523 | } |
| 1524 | |
Gian Marco Iodice | 23e2479 | 2018-09-07 15:32:14 +0100 | [diff] [blame] | 1525 | /** Create a backend reorg layer function |
| 1526 | * |
Michele Di Giorgio | c30b668 | 2018-09-12 17:44:08 +0100 | [diff] [blame] | 1527 | * @tparam ReorgLayerFunction Backend reorg function |
Gian Marco Iodice | 23e2479 | 2018-09-07 15:32:14 +0100 | [diff] [blame] | 1528 | * @tparam TargetInfo Target-specific information |
| 1529 | * |
| 1530 | * @param[in] node Node to create the backend function for |
| 1531 | * |
| 1532 | * @return Backend reshape layer function |
| 1533 | */ |
| 1534 | template <typename ReorgLayerFunction, typename TargetInfo> |
| 1535 | std::unique_ptr<IFunction> create_reorg_layer(ReorgLayerNode &node) |
| 1536 | { |
| 1537 | validate_node<TargetInfo>(node, 1 /* expected inputs */, 1 /* expected outputs */); |
| 1538 | |
| 1539 | // Extract IO and info |
| 1540 | typename TargetInfo::TensorType *input = get_backing_tensor<TargetInfo>(node.input(0)); |
| 1541 | typename TargetInfo::TensorType *output = get_backing_tensor<TargetInfo>(node.output(0)); |
| 1542 | ARM_COMPUTE_ERROR_ON(input == nullptr); |
| 1543 | ARM_COMPUTE_ERROR_ON(output == nullptr); |
| 1544 | |
| 1545 | // Create and configure function |
| 1546 | auto func = support::cpp14::make_unique<ReorgLayerFunction>(); |
| 1547 | func->configure(input, output, node.stride()); |
| 1548 | |
| 1549 | // Log info |
Pablo Tello | 3252143 | 2018-11-15 14:43:10 +0000 | [diff] [blame] | 1550 | ARM_COMPUTE_LOG_GRAPH_INFO("Instantiated " |
| 1551 | << node.name() |
| 1552 | << " Type: " << node.type() |
| 1553 | << " Target: " << TargetInfo::TargetType |
Gian Marco Iodice | 23e2479 | 2018-09-07 15:32:14 +0100 | [diff] [blame] | 1554 | << " Data Type: " << input->info()->data_type() |
| 1555 | << " Input shape: " << input->info()->tensor_shape() |
| 1556 | << " Output shape: " << output->info()->tensor_shape() |
| 1557 | << std::endl); |
| 1558 | |
Georgios Pinitas | 0b192e8 | 2020-02-20 17:09:28 +0000 | [diff] [blame] | 1559 | return RETURN_UNIQUE_PTR(func); |
Gian Marco Iodice | 23e2479 | 2018-09-07 15:32:14 +0100 | [diff] [blame] | 1560 | } |
| 1561 | |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 1562 | /** Create a backend reshape layer function |
| 1563 | * |
| 1564 | * @tparam ReshapeLayerFunction Backend reshape function |
| 1565 | * @tparam TargetInfo Target-specific information |
| 1566 | * |
| 1567 | * @param[in] node Node to create the backend function for |
| 1568 | * |
| 1569 | * @return Backend reshape layer function |
| 1570 | */ |
| 1571 | template <typename ReshapeLayerFunction, typename TargetInfo> |
| 1572 | std::unique_ptr<IFunction> create_reshape_layer(ReshapeLayerNode &node) |
| 1573 | { |
| 1574 | validate_node<TargetInfo>(node, 1 /* expected inputs */, 1 /* expected outputs */); |
| 1575 | |
| 1576 | // Extract IO and info |
| 1577 | typename TargetInfo::TensorType *input = get_backing_tensor<TargetInfo>(node.input(0)); |
| 1578 | typename TargetInfo::TensorType *output = get_backing_tensor<TargetInfo>(node.output(0)); |
| 1579 | ARM_COMPUTE_ERROR_ON(input == nullptr); |
| 1580 | ARM_COMPUTE_ERROR_ON(output == nullptr); |
| 1581 | |
| 1582 | // Create and configure function |
| 1583 | auto func = support::cpp14::make_unique<ReshapeLayerFunction>(); |
| 1584 | func->configure(input, output); |
| 1585 | |
| 1586 | // Log info |
Pablo Tello | 3252143 | 2018-11-15 14:43:10 +0000 | [diff] [blame] | 1587 | ARM_COMPUTE_LOG_GRAPH_INFO("Instantiated " |
| 1588 | << node.name() |
| 1589 | << " Type: " << node.type() |
| 1590 | << " Target: " << TargetInfo::TargetType |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 1591 | << " Data Type: " << input->info()->data_type() |
| 1592 | << " Input shape: " << input->info()->tensor_shape() |
| 1593 | << " Output shape: " << output->info()->tensor_shape() |
| 1594 | << std::endl); |
| 1595 | |
Georgios Pinitas | 0b192e8 | 2020-02-20 17:09:28 +0000 | [diff] [blame] | 1596 | return RETURN_UNIQUE_PTR(func); |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 1597 | } |
| 1598 | |
| 1599 | /** Create a backend resize layer function |
| 1600 | * |
| 1601 | * @tparam ResizeLayerFunction Backend resize function |
| 1602 | * @tparam TargetInfo Target-specific information |
| 1603 | * |
| 1604 | * @param[in] node Node to create the backend function for |
| 1605 | * |
| 1606 | * @return Backend resize layer function |
| 1607 | */ |
| 1608 | template <typename ResizeLayerFunction, typename TargetInfo> |
| 1609 | std::unique_ptr<IFunction> create_resize_layer(ResizeLayerNode &node) |
| 1610 | { |
| 1611 | validate_node<TargetInfo>(node, 1 /* expected inputs */, 1 /* expected outputs */); |
| 1612 | |
| 1613 | // Extract IO and info |
| 1614 | typename TargetInfo::TensorType *input = get_backing_tensor<TargetInfo>(node.input(0)); |
| 1615 | typename TargetInfo::TensorType *output = get_backing_tensor<TargetInfo>(node.output(0)); |
| 1616 | ARM_COMPUTE_ERROR_ON(input == nullptr); |
| 1617 | ARM_COMPUTE_ERROR_ON(output == nullptr); |
| 1618 | const InterpolationPolicy policy = node.policy(); |
| 1619 | |
| 1620 | // Create and configure function |
| 1621 | auto func = support::cpp14::make_unique<ResizeLayerFunction>(); |
Sang-Hoon Park | ccd9496 | 2020-06-09 12:09:24 +0100 | [diff] [blame] | 1622 | func->configure(input, output, ScaleKernelInfo{ policy, BorderMode::CONSTANT }); |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 1623 | |
| 1624 | // Log info |
Pablo Tello | 3252143 | 2018-11-15 14:43:10 +0000 | [diff] [blame] | 1625 | ARM_COMPUTE_LOG_GRAPH_INFO("Instantiated " |
| 1626 | << node.name() |
| 1627 | << " Type: " << node.type() |
| 1628 | << " Target: " << TargetInfo::TargetType |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 1629 | << " Data Type: " << input->info()->data_type() |
| 1630 | << " Input shape: " << input->info()->tensor_shape() |
| 1631 | << " Output shape: " << output->info()->tensor_shape() |
| 1632 | << " Interpolation: " << policy |
| 1633 | << std::endl); |
| 1634 | |
Georgios Pinitas | 0b192e8 | 2020-02-20 17:09:28 +0000 | [diff] [blame] | 1635 | return RETURN_UNIQUE_PTR(func); |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 1636 | } |
| 1637 | |
Manuel Bottini | 3f9d4d7 | 2018-10-19 14:04:42 +0100 | [diff] [blame] | 1638 | /** Create a backend ROI align layer function |
| 1639 | * |
| 1640 | * @tparam ROIAlignLayerFunction ROI Align function |
| 1641 | * @tparam TargetInfo Target-specific information |
| 1642 | * |
| 1643 | * @param[in] node Node to create the backend function for |
| 1644 | * |
| 1645 | * @return ROI Align layer function |
| 1646 | */ |
| 1647 | template <typename ROIAlignLayerFunction, typename TargetInfo> |
| 1648 | std::unique_ptr<IFunction> create_roi_align_layer(ROIAlignLayerNode &node) |
| 1649 | { |
| 1650 | validate_node<TargetInfo>(node, 2 /* expected inputs */, 1 /* expected outputs */); |
| 1651 | |
| 1652 | // Extract IO and info |
| 1653 | typename TargetInfo::TensorType *input = get_backing_tensor<TargetInfo>(node.input(0)); |
| 1654 | typename TargetInfo::TensorType *rois = get_backing_tensor<TargetInfo>(node.input(1)); |
| 1655 | typename TargetInfo::TensorType *output = get_backing_tensor<TargetInfo>(node.output(0)); |
| 1656 | ARM_COMPUTE_ERROR_ON(input == nullptr); |
| 1657 | ARM_COMPUTE_ERROR_ON(output == nullptr); |
| 1658 | ARM_COMPUTE_ERROR_ON(rois == nullptr); |
| 1659 | |
| 1660 | const ROIPoolingLayerInfo pool_info = node.pooling_info(); |
| 1661 | |
| 1662 | // Create and configure function |
| 1663 | auto func = support::cpp14::make_unique<ROIAlignLayerFunction>(); |
| 1664 | |
| 1665 | func->configure(input, rois, output, pool_info); |
| 1666 | |
| 1667 | // Log info |
Isabella Gottardi | 0ae5de9 | 2019-03-14 10:32:11 +0000 | [diff] [blame] | 1668 | ARM_COMPUTE_LOG_GRAPH_INFO("Instantiated " |
| 1669 | << node.name() |
| 1670 | << " Type: " << node.type() |
| 1671 | << " Target: " << TargetInfo::TargetType |
Manuel Bottini | 3f9d4d7 | 2018-10-19 14:04:42 +0100 | [diff] [blame] | 1672 | << " Data Type: " << input->info()->data_type() |
| 1673 | << " Input shape: " << input->info()->tensor_shape() |
| 1674 | << " Output shape: " << output->info()->tensor_shape() |
| 1675 | << " ROIs shape: " << rois->info()->tensor_shape() |
| 1676 | << " ROIPooling width: " << pool_info.pooled_width() |
| 1677 | << " ROIPooling height: " << pool_info.pooled_height() |
| 1678 | << std::endl); |
| 1679 | |
Georgios Pinitas | 0b192e8 | 2020-02-20 17:09:28 +0000 | [diff] [blame] | 1680 | return RETURN_UNIQUE_PTR(func); |
Manuel Bottini | 3f9d4d7 | 2018-10-19 14:04:42 +0100 | [diff] [blame] | 1681 | } |
| 1682 | |
Michele Di Giorgio | c30b668 | 2018-09-12 17:44:08 +0100 | [diff] [blame] | 1683 | /** Create a backend slice layer function |
| 1684 | * |
| 1685 | * @tparam SliceLayerFunction Backend slice function |
| 1686 | * @tparam TargetInfo Target-specific information |
| 1687 | * |
| 1688 | * @param[in] node Node to create the backend function for |
| 1689 | * |
| 1690 | * @return Backend slice layer function |
| 1691 | */ |
| 1692 | template <typename SliceLayerFunction, typename TargetInfo> |
| 1693 | std::unique_ptr<IFunction> create_slice_layer(SliceLayerNode &node) |
| 1694 | { |
| 1695 | validate_node<TargetInfo>(node, 1 /* expected inputs */, 1 /* expected outputs */); |
| 1696 | |
| 1697 | // Extract IO and info |
| 1698 | typename TargetInfo::TensorType *input = get_backing_tensor<TargetInfo>(node.input(0)); |
| 1699 | typename TargetInfo::TensorType *output = get_backing_tensor<TargetInfo>(node.output(0)); |
| 1700 | ARM_COMPUTE_ERROR_ON(input == nullptr); |
| 1701 | ARM_COMPUTE_ERROR_ON(output == nullptr); |
| 1702 | |
| 1703 | // Create and configure function |
| 1704 | auto func = support::cpp14::make_unique<SliceLayerFunction>(); |
| 1705 | func->configure(input, output, node.starts(), node.ends()); |
| 1706 | |
| 1707 | // Log info |
Pablo Tello | 3252143 | 2018-11-15 14:43:10 +0000 | [diff] [blame] | 1708 | ARM_COMPUTE_LOG_GRAPH_INFO("Instantiated " |
| 1709 | << node.name() |
| 1710 | << " Type: " << node.type() |
| 1711 | << " Target: " << TargetInfo::TargetType |
Michele Di Giorgio | c30b668 | 2018-09-12 17:44:08 +0100 | [diff] [blame] | 1712 | << " Data Type: " << input->info()->data_type() |
| 1713 | << " Input shape: " << input->info()->tensor_shape() |
| 1714 | << " Output shape: " << output->info()->tensor_shape() |
| 1715 | << std::endl); |
| 1716 | |
Georgios Pinitas | 0b192e8 | 2020-02-20 17:09:28 +0000 | [diff] [blame] | 1717 | return RETURN_UNIQUE_PTR(func); |
Michele Di Giorgio | c30b668 | 2018-09-12 17:44:08 +0100 | [diff] [blame] | 1718 | } |
| 1719 | |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 1720 | /** Create a backend softmax layer function |
| 1721 | * |
| 1722 | * @tparam SoftmaxLayerFunction Backend softmax function |
| 1723 | * @tparam TargetInfo Target-specific information |
| 1724 | * |
| 1725 | * @param[in] node Node to create the backend function for |
| 1726 | * @param[in] ctx Graph context |
| 1727 | * |
| 1728 | * @return Backend softmax layer function |
| 1729 | */ |
| 1730 | template <typename SoftmaxLayerFunction, typename TargetInfo> |
| 1731 | std::unique_ptr<IFunction> create_softmax_layer(SoftmaxLayerNode &node, GraphContext &ctx) |
| 1732 | { |
| 1733 | validate_node<TargetInfo>(node, 1 /* expected inputs */, 1 /* expected outputs */); |
| 1734 | |
| 1735 | // Extract IO and info |
| 1736 | typename TargetInfo::TensorType *input = get_backing_tensor<TargetInfo>(node.input(0)); |
| 1737 | typename TargetInfo::TensorType *output = get_backing_tensor<TargetInfo>(node.output(0)); |
| 1738 | const float beta = node.beta(); |
| 1739 | ARM_COMPUTE_ERROR_ON(input == nullptr); |
| 1740 | ARM_COMPUTE_ERROR_ON(output == nullptr); |
| 1741 | |
| 1742 | // Create and configure function |
| 1743 | auto func = support::cpp14::make_unique<SoftmaxLayerFunction>(get_memory_manager(ctx, TargetInfo::TargetType)); |
| 1744 | func->configure(input, output, beta); |
| 1745 | |
| 1746 | // Log info |
Pablo Tello | 3252143 | 2018-11-15 14:43:10 +0000 | [diff] [blame] | 1747 | ARM_COMPUTE_LOG_GRAPH_INFO("Instantiated " |
| 1748 | << node.name() |
| 1749 | << " Type: " << node.type() |
| 1750 | << " Target: " << TargetInfo::TargetType |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 1751 | << " Data Type: " << input->info()->data_type() |
| 1752 | << " Input shape: " << input->info()->tensor_shape() |
| 1753 | << " Output shape: " << output->info()->tensor_shape() |
| 1754 | << std::endl); |
| 1755 | |
Georgios Pinitas | 0b192e8 | 2020-02-20 17:09:28 +0000 | [diff] [blame] | 1756 | return RETURN_UNIQUE_PTR(func); |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 1757 | } |
Michele Di Giorgio | ec69975 | 2019-03-22 15:25:32 +0000 | [diff] [blame] | 1758 | |
| 1759 | /** Create a backend layer stack function |
| 1760 | * |
| 1761 | * @tparam StackLayerFunction Backend stack function |
| 1762 | * @tparam TargetInfo Target-specific information |
| 1763 | * |
| 1764 | * @param[in] node Node to create the backend function for |
| 1765 | * |
| 1766 | * @return Backend stack layer function |
| 1767 | */ |
| 1768 | template <typename StackLayerFunction, typename TargetInfo> |
| 1769 | std::unique_ptr<arm_compute::IFunction> create_stack_layer(StackLayerNode &node) |
| 1770 | { |
| 1771 | ARM_COMPUTE_LOG_GRAPH_VERBOSE("Creating Stack node with ID : " << node.id() << " and Name: " << node.name() << std::endl); |
| 1772 | ARM_COMPUTE_ERROR_ON(node.num_outputs() != 1); |
| 1773 | |
| 1774 | // Extract IO and info |
| 1775 | std::vector<typename TargetInfo::TensorType *> inputs; |
| 1776 | for(unsigned int i = 0; i < node.num_inputs(); ++i) |
| 1777 | { |
| 1778 | inputs.push_back(get_backing_tensor<TargetInfo>(node.input(i))); |
| 1779 | } |
| 1780 | typename TargetInfo::TensorType *output = get_backing_tensor<TargetInfo>(node.output(0)); |
| 1781 | const int axis = node.axis(); |
| 1782 | |
| 1783 | // Create and configure function |
| 1784 | auto func = support::cpp14::make_unique<StackLayerFunction>(); |
| 1785 | func->configure(inputs, axis, output); |
| 1786 | |
| 1787 | // Log info |
| 1788 | ARM_COMPUTE_LOG_GRAPH_INFO("Instantiated " |
| 1789 | << node.name() |
| 1790 | << " Type: " << node.type() |
| 1791 | << " Target: " << TargetInfo::TargetType |
| 1792 | << " Data Type: " << output->info()->data_type() |
| 1793 | << " Inputs shape: " << inputs[0]->info()->tensor_shape() |
| 1794 | << " Output shape: " << output->info()->tensor_shape() |
| 1795 | << " Num Inputs: " << inputs.size() |
| 1796 | << " Axis: " << axis |
| 1797 | << std::endl); |
| 1798 | |
Georgios Pinitas | 0b192e8 | 2020-02-20 17:09:28 +0000 | [diff] [blame] | 1799 | return RETURN_UNIQUE_PTR(func); |
Michele Di Giorgio | ec69975 | 2019-03-22 15:25:32 +0000 | [diff] [blame] | 1800 | } |
thecha01 | 2bfadd9 | 2020-08-12 17:25:51 +0100 | [diff] [blame] | 1801 | |
| 1802 | /** Create a backend slice layer function |
| 1803 | * |
| 1804 | * @tparam StridedSliceLayerFunction Backend strided slice function |
| 1805 | * @tparam TargetInfo Target-specific information |
| 1806 | * |
| 1807 | * @param[in] node Node to create the backend function for |
| 1808 | * |
| 1809 | * @return Backend strided slice layer function |
| 1810 | */ |
| 1811 | template <typename StridedSliceLayerFunction, typename TargetInfo> |
| 1812 | std::unique_ptr<IFunction> create_strided_slice_layer(StridedSliceLayerNode &node) |
| 1813 | { |
| 1814 | validate_node<TargetInfo>(node, 1 /* expected inputs */, 1 /* expected outputs */); |
| 1815 | |
| 1816 | // Extract IO and info |
| 1817 | typename TargetInfo::TensorType *input = get_backing_tensor<TargetInfo>(node.input(0)); |
| 1818 | typename TargetInfo::TensorType *output = get_backing_tensor<TargetInfo>(node.output(0)); |
| 1819 | Coordinates starts = node.starts(); |
| 1820 | Coordinates ends = node.ends(); |
| 1821 | BiStrides strides = node.strides(); |
| 1822 | StridedSliceLayerInfo info = node.strided_slice_info(); |
| 1823 | |
| 1824 | ARM_COMPUTE_ERROR_ON(input == nullptr); |
| 1825 | ARM_COMPUTE_ERROR_ON(output == nullptr); |
| 1826 | |
| 1827 | // Create and configure function |
| 1828 | auto func = support::cpp14::make_unique<StridedSliceLayerFunction>(); |
| 1829 | func->configure(input, output, starts, ends, strides, info.begin_mask(), info.end_mask(), info.shrink_axis_mask()); |
| 1830 | |
| 1831 | // Log info |
| 1832 | ARM_COMPUTE_LOG_GRAPH_INFO("Instantiated " |
| 1833 | << node.name() |
| 1834 | << " Type: " << node.type() |
| 1835 | << " Target: " << TargetInfo::TargetType |
| 1836 | << " Data Type: " << input->info()->data_type() |
| 1837 | << " Input shape: " << input->info()->tensor_shape() |
| 1838 | << " Output shape: " << output->info()->tensor_shape() |
| 1839 | << std::endl); |
| 1840 | |
| 1841 | return RETURN_UNIQUE_PTR(func); |
| 1842 | } |
| 1843 | |
Michalis Spyrou | 4e1c3f3 | 2018-09-20 17:14:03 +0100 | [diff] [blame] | 1844 | /** Create a backend Upsample layer function |
| 1845 | * |
| 1846 | * @tparam UpsampleLayerFunction Backend Upsample function |
| 1847 | * @tparam TargetInfo Target-specific information |
| 1848 | * |
| 1849 | * @param[in] node Node to create the backend function for |
| 1850 | * @param[in] ctx Graph context |
| 1851 | * |
| 1852 | * @return Backend Upsample layer function |
| 1853 | */ |
| 1854 | template <typename UpsampleLayerFunction, typename TargetInfo> |
| 1855 | std::unique_ptr<IFunction> create_upsample_layer(UpsampleLayerNode &node, GraphContext &ctx) |
| 1856 | { |
Michalis Spyrou | 6bff195 | 2019-10-02 17:22:11 +0100 | [diff] [blame] | 1857 | ARM_COMPUTE_UNUSED(ctx); |
Michalis Spyrou | 4e1c3f3 | 2018-09-20 17:14:03 +0100 | [diff] [blame] | 1858 | validate_node<TargetInfo>(node, 1 /* expected inputs */, 1 /* expected outputs */); |
| 1859 | |
| 1860 | // Extract IO and info |
| 1861 | typename TargetInfo::TensorType *input = get_backing_tensor<TargetInfo>(node.input(0)); |
| 1862 | typename TargetInfo::TensorType *output = get_backing_tensor<TargetInfo>(node.output(0)); |
| 1863 | const Size2D info = node.info(); |
| 1864 | const InterpolationPolicy upsampling_policy = node.upsampling_policy(); |
| 1865 | ARM_COMPUTE_ERROR_ON(upsampling_policy != InterpolationPolicy::NEAREST_NEIGHBOR); |
| 1866 | ARM_COMPUTE_ERROR_ON(info.x() != 2 || info.y() != 2); |
| 1867 | ARM_COMPUTE_ERROR_ON(input == nullptr); |
| 1868 | ARM_COMPUTE_ERROR_ON(output == nullptr); |
| 1869 | |
| 1870 | // Create and configure function |
| 1871 | auto func = support::cpp14::make_unique<UpsampleLayerFunction>(); |
| 1872 | func->configure(input, output, info, upsampling_policy); |
| 1873 | |
| 1874 | // Log info |
Pablo Tello | 3252143 | 2018-11-15 14:43:10 +0000 | [diff] [blame] | 1875 | ARM_COMPUTE_LOG_GRAPH_INFO("Instantiated " |
| 1876 | << node.name() |
| 1877 | << " Type: " << node.type() |
| 1878 | << " Target: " << TargetInfo::TargetType |
Michalis Spyrou | 4e1c3f3 | 2018-09-20 17:14:03 +0100 | [diff] [blame] | 1879 | << " Data Type: " << input->info()->data_type() |
| 1880 | << " Input shape: " << input->info()->tensor_shape() |
| 1881 | << " Output shape: " << output->info()->tensor_shape() |
| 1882 | << " Strides: " << info |
| 1883 | << " Upsampling policy: " << upsampling_policy |
| 1884 | << std::endl); |
| 1885 | |
Georgios Pinitas | 0b192e8 | 2020-02-20 17:09:28 +0000 | [diff] [blame] | 1886 | return RETURN_UNIQUE_PTR(func); |
Michalis Spyrou | 4e1c3f3 | 2018-09-20 17:14:03 +0100 | [diff] [blame] | 1887 | } |
Michalis Spyrou | 96f6769 | 2018-09-13 11:39:28 +0100 | [diff] [blame] | 1888 | /** Create a backend YOLO layer function |
| 1889 | * |
| 1890 | * @tparam YoloLayerFunction Backend YOLO function |
| 1891 | * @tparam TargetInfo Target-specific information |
| 1892 | * |
| 1893 | * @param[in] node Node to create the backend function for |
| 1894 | * @param[in] ctx Graph context |
| 1895 | * |
| 1896 | * @return Backend YOLO layer function |
| 1897 | */ |
| 1898 | template <typename YOLOlayerFunction, typename TargetInfo> |
| 1899 | std::unique_ptr<IFunction> create_yolo_layer(YOLOLayerNode &node, GraphContext &ctx) |
| 1900 | { |
Michalis Spyrou | 6bff195 | 2019-10-02 17:22:11 +0100 | [diff] [blame] | 1901 | ARM_COMPUTE_UNUSED(ctx); |
Michalis Spyrou | 96f6769 | 2018-09-13 11:39:28 +0100 | [diff] [blame] | 1902 | validate_node<TargetInfo>(node, 1 /* expected inputs */, 1 /* expected outputs */); |
| 1903 | |
| 1904 | // Extract IO and info |
| 1905 | typename TargetInfo::TensorType *input = get_backing_tensor<TargetInfo>(node.input(0)); |
| 1906 | typename TargetInfo::TensorType *output = get_backing_tensor<TargetInfo>(node.output(0)); |
| 1907 | const ActivationLayerInfo act_info = node.activation_info(); |
| 1908 | const int32_t num_classes = node.num_classes(); |
| 1909 | ARM_COMPUTE_ERROR_ON(num_classes <= 0); |
| 1910 | ARM_COMPUTE_ERROR_ON(input == nullptr); |
| 1911 | ARM_COMPUTE_ERROR_ON(output == nullptr); |
| 1912 | |
| 1913 | // Create and configure function |
| 1914 | auto func = support::cpp14::make_unique<YOLOlayerFunction>(); |
| 1915 | func->configure(input, output, act_info, num_classes); |
| 1916 | |
| 1917 | // Log info |
Pablo Tello | 3252143 | 2018-11-15 14:43:10 +0000 | [diff] [blame] | 1918 | ARM_COMPUTE_LOG_GRAPH_INFO("Instantiated " |
| 1919 | << node.name() |
| 1920 | << " Type: " << node.type() |
| 1921 | << " Target: " << TargetInfo::TargetType |
Michalis Spyrou | 96f6769 | 2018-09-13 11:39:28 +0100 | [diff] [blame] | 1922 | << " Data Type: " << input->info()->data_type() |
| 1923 | << " Input shape: " << input->info()->tensor_shape() |
| 1924 | << " Output shape: " << output->info()->tensor_shape() |
| 1925 | << " Activation function: " << act_info.activation() |
| 1926 | << " Num classes: " << num_classes |
| 1927 | << std::endl); |
| 1928 | |
Georgios Pinitas | 0b192e8 | 2020-02-20 17:09:28 +0000 | [diff] [blame] | 1929 | return RETURN_UNIQUE_PTR(func); |
Michalis Spyrou | 96f6769 | 2018-09-13 11:39:28 +0100 | [diff] [blame] | 1930 | } |
Georgios Pinitas | da2491f | 2018-06-01 17:49:09 +0100 | [diff] [blame] | 1931 | } // namespace detail |
| 1932 | } // namespace backends |
| 1933 | } // namespace graph |
| 1934 | } // namespace arm_compute |
| 1935 | |
Michalis Spyrou | f464337 | 2019-11-29 16:17:13 +0000 | [diff] [blame] | 1936 | #endif /* ARM_COMPUTE_GRAPH_BACKENDS_DETAIL_FUNCTION_HELPERS_H */ |