Sang-Hoon Park | 63001ac | 2021-01-18 14:20:27 +0000 | [diff] [blame] | 1 | /* |
| 2 | * Copyright (c) 2018-2021 Arm Limited. |
| 3 | * |
| 4 | * SPDX-License-Identifier: MIT |
| 5 | * |
| 6 | * Permission is hereby granted, free of charge, to any person obtaining a copy |
| 7 | * of this software and associated documentation files (the "Software"), to |
| 8 | * deal in the Software without restriction, including without limitation the |
| 9 | * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or |
| 10 | * sell copies of the Software, and to permit persons to whom the Software is |
| 11 | * furnished to do so, subject to the following conditions: |
| 12 | * |
| 13 | * The above copyright notice and this permission notice shall be included in all |
| 14 | * copies or substantial portions of the Software. |
| 15 | * |
| 16 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR |
| 17 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, |
| 18 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE |
| 19 | * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER |
| 20 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, |
| 21 | * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE |
| 22 | * SOFTWARE. |
| 23 | */ |
| 24 | #include "arm_compute/runtime/NEON/functions/NEElementwiseOperations.h" |
| 25 | #include "arm_compute/core/Validate.h" |
| 26 | #include "src/runtime/cpu/operators/CpuElementwise.h" |
| 27 | |
| 28 | #include "arm_compute/core/ITensor.h" |
| 29 | |
| 30 | #include <utility> |
| 31 | |
| 32 | namespace arm_compute |
| 33 | { |
| 34 | struct NEElementwiseMax::Impl |
| 35 | { |
| 36 | const ITensor *src_0{ nullptr }; |
| 37 | const ITensor *src_1{ nullptr }; |
| 38 | ITensor *dst{ nullptr }; |
| 39 | std::unique_ptr<cpu::CpuElementwiseMax> op{ nullptr }; |
| 40 | }; |
| 41 | |
| 42 | NEElementwiseMax::NEElementwiseMax() |
| 43 | : _impl(std::make_unique<Impl>()) |
| 44 | { |
| 45 | } |
| 46 | NEElementwiseMax::NEElementwiseMax(NEElementwiseMax &&) = default; |
| 47 | NEElementwiseMax &NEElementwiseMax::operator=(NEElementwiseMax &&) = default; |
| 48 | NEElementwiseMax::~NEElementwiseMax() = default; |
| 49 | |
| 50 | void NEElementwiseMax::configure(ITensor *input1, ITensor *input2, ITensor *output, const ActivationLayerInfo &act_info) |
| 51 | { |
| 52 | ARM_COMPUTE_UNUSED(act_info); |
| 53 | _impl->src_0 = input1; |
| 54 | _impl->src_1 = input2; |
| 55 | _impl->dst = output; |
| 56 | _impl->op = std::make_unique<cpu::CpuElementwiseMax>(); |
| 57 | _impl->op->configure(input1->info(), input2->info(), output->info()); |
| 58 | } |
| 59 | |
| 60 | Status NEElementwiseMax::validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, const ActivationLayerInfo &act_info) |
| 61 | { |
| 62 | ARM_COMPUTE_RETURN_ERROR_ON(act_info.enabled()); |
| 63 | return cpu::CpuElementwiseMax::validate(input1, input2, output); |
| 64 | } |
| 65 | |
| 66 | void NEElementwiseMax::run() |
| 67 | { |
| 68 | ITensorPack pack; |
| 69 | pack.add_tensor(TensorType::ACL_SRC_0, _impl->src_0); |
| 70 | pack.add_tensor(TensorType::ACL_SRC_1, _impl->src_1); |
| 71 | pack.add_tensor(TensorType::ACL_DST, _impl->dst); |
| 72 | _impl->op->run(pack); |
| 73 | } |
| 74 | |
| 75 | struct NEElementwiseMin::Impl |
| 76 | { |
| 77 | const ITensor *src_0{ nullptr }; |
| 78 | const ITensor *src_1{ nullptr }; |
| 79 | ITensor *dst{ nullptr }; |
| 80 | std::unique_ptr<cpu::CpuElementwiseMin> op{ nullptr }; |
| 81 | }; |
| 82 | |
| 83 | NEElementwiseMin::NEElementwiseMin() |
| 84 | : _impl(std::make_unique<Impl>()) |
| 85 | { |
| 86 | } |
| 87 | NEElementwiseMin::NEElementwiseMin(NEElementwiseMin &&) = default; |
| 88 | NEElementwiseMin &NEElementwiseMin::operator=(NEElementwiseMin &&) = default; |
| 89 | NEElementwiseMin::~NEElementwiseMin() = default; |
| 90 | |
| 91 | void NEElementwiseMin::configure(ITensor *input1, ITensor *input2, ITensor *output, const ActivationLayerInfo &act_info) |
| 92 | { |
| 93 | ARM_COMPUTE_UNUSED(act_info); |
| 94 | _impl->src_0 = input1; |
| 95 | _impl->src_1 = input2; |
| 96 | _impl->dst = output; |
| 97 | _impl->op = std::make_unique<cpu::CpuElementwiseMin>(); |
| 98 | _impl->op->configure(input1->info(), input2->info(), output->info()); |
| 99 | } |
| 100 | |
| 101 | Status NEElementwiseMin::validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, const ActivationLayerInfo &act_info) |
| 102 | { |
| 103 | ARM_COMPUTE_RETURN_ERROR_ON(act_info.enabled()); |
| 104 | return cpu::CpuElementwiseMin::validate(input1, input2, output); |
| 105 | } |
| 106 | |
| 107 | void NEElementwiseMin::run() |
| 108 | { |
| 109 | ITensorPack pack; |
| 110 | pack.add_tensor(TensorType::ACL_SRC_0, _impl->src_0); |
| 111 | pack.add_tensor(TensorType::ACL_SRC_1, _impl->src_1); |
| 112 | pack.add_tensor(TensorType::ACL_DST, _impl->dst); |
| 113 | _impl->op->run(pack); |
| 114 | } |
| 115 | |
| 116 | struct NEElementwiseSquaredDiff::Impl |
| 117 | { |
| 118 | const ITensor *src_0{ nullptr }; |
| 119 | const ITensor *src_1{ nullptr }; |
| 120 | ITensor *dst{ nullptr }; |
| 121 | std::unique_ptr<cpu::CpuElementwiseSquaredDiff> op{ nullptr }; |
| 122 | }; |
| 123 | |
| 124 | NEElementwiseSquaredDiff::NEElementwiseSquaredDiff() |
| 125 | : _impl(std::make_unique<Impl>()) |
| 126 | { |
| 127 | } |
| 128 | NEElementwiseSquaredDiff::NEElementwiseSquaredDiff(NEElementwiseSquaredDiff &&) = default; |
| 129 | NEElementwiseSquaredDiff &NEElementwiseSquaredDiff::operator=(NEElementwiseSquaredDiff &&) = default; |
| 130 | NEElementwiseSquaredDiff::~NEElementwiseSquaredDiff() = default; |
| 131 | |
| 132 | void NEElementwiseSquaredDiff::configure(ITensor *input1, ITensor *input2, ITensor *output, const ActivationLayerInfo &act_info) |
| 133 | { |
| 134 | ARM_COMPUTE_UNUSED(act_info); |
| 135 | _impl->src_0 = input1; |
| 136 | _impl->src_1 = input2; |
| 137 | _impl->dst = output; |
| 138 | _impl->op = std::make_unique<cpu::CpuElementwiseSquaredDiff>(); |
| 139 | _impl->op->configure(input1->info(), input2->info(), output->info()); |
| 140 | } |
| 141 | |
| 142 | Status NEElementwiseSquaredDiff::validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, const ActivationLayerInfo &act_info) |
| 143 | { |
| 144 | ARM_COMPUTE_RETURN_ERROR_ON(act_info.enabled()); |
| 145 | return cpu::CpuElementwiseSquaredDiff::validate(input1, input2, output); |
| 146 | } |
| 147 | |
| 148 | void NEElementwiseSquaredDiff::run() |
| 149 | { |
| 150 | ITensorPack pack; |
| 151 | pack.add_tensor(TensorType::ACL_SRC_0, _impl->src_0); |
| 152 | pack.add_tensor(TensorType::ACL_SRC_1, _impl->src_1); |
| 153 | pack.add_tensor(TensorType::ACL_DST, _impl->dst); |
| 154 | _impl->op->run(pack); |
| 155 | } |
| 156 | |
| 157 | struct NEElementwiseDivision::Impl |
| 158 | { |
| 159 | const ITensor *src_0{ nullptr }; |
| 160 | const ITensor *src_1{ nullptr }; |
| 161 | ITensor *dst{ nullptr }; |
| 162 | std::unique_ptr<cpu::CpuElementwiseDivision> op{ nullptr }; |
| 163 | }; |
| 164 | |
| 165 | NEElementwiseDivision::NEElementwiseDivision() |
| 166 | : _impl(std::make_unique<Impl>()) |
| 167 | { |
| 168 | } |
| 169 | NEElementwiseDivision::NEElementwiseDivision(NEElementwiseDivision &&) = default; |
| 170 | NEElementwiseDivision &NEElementwiseDivision::operator=(NEElementwiseDivision &&) = default; |
| 171 | NEElementwiseDivision::~NEElementwiseDivision() = default; |
| 172 | |
| 173 | void NEElementwiseDivision::configure(ITensor *input1, ITensor *input2, ITensor *output, const ActivationLayerInfo &act_info) |
| 174 | { |
| 175 | ARM_COMPUTE_UNUSED(act_info); |
| 176 | _impl->src_0 = input1; |
| 177 | _impl->src_1 = input2; |
| 178 | _impl->dst = output; |
| 179 | _impl->op = std::make_unique<cpu::CpuElementwiseDivision>(); |
| 180 | _impl->op->configure(input1->info(), input2->info(), output->info()); |
| 181 | } |
| 182 | |
| 183 | Status NEElementwiseDivision::validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, const ActivationLayerInfo &act_info) |
| 184 | { |
| 185 | ARM_COMPUTE_RETURN_ERROR_ON(act_info.enabled()); |
| 186 | return cpu::CpuElementwiseDivision::validate(input1, input2, output); |
| 187 | } |
| 188 | |
| 189 | void NEElementwiseDivision::run() |
| 190 | { |
| 191 | ITensorPack pack; |
| 192 | pack.add_tensor(TensorType::ACL_SRC_0, _impl->src_0); |
| 193 | pack.add_tensor(TensorType::ACL_SRC_1, _impl->src_1); |
| 194 | pack.add_tensor(TensorType::ACL_DST, _impl->dst); |
| 195 | _impl->op->run(pack); |
| 196 | } |
| 197 | |
| 198 | struct NEElementwisePower::Impl |
| 199 | { |
| 200 | const ITensor *src_0{ nullptr }; |
| 201 | const ITensor *src_1{ nullptr }; |
| 202 | ITensor *dst{ nullptr }; |
| 203 | std::unique_ptr<cpu::CpuElementwisePower> op{ nullptr }; |
| 204 | }; |
| 205 | |
| 206 | NEElementwisePower::NEElementwisePower() |
| 207 | : _impl(std::make_unique<Impl>()) |
| 208 | { |
| 209 | } |
| 210 | NEElementwisePower::NEElementwisePower(NEElementwisePower &&) = default; |
| 211 | NEElementwisePower &NEElementwisePower::operator=(NEElementwisePower &&) = default; |
| 212 | NEElementwisePower::~NEElementwisePower() = default; |
| 213 | |
| 214 | void NEElementwisePower::configure(ITensor *input1, ITensor *input2, ITensor *output, const ActivationLayerInfo &act_info) |
| 215 | { |
| 216 | ARM_COMPUTE_UNUSED(act_info); |
| 217 | _impl->src_0 = input1; |
| 218 | _impl->src_1 = input2; |
| 219 | _impl->dst = output; |
| 220 | _impl->op = std::make_unique<cpu::CpuElementwisePower>(); |
| 221 | _impl->op->configure(input1->info(), input2->info(), output->info()); |
| 222 | } |
| 223 | |
| 224 | Status NEElementwisePower::validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, const ActivationLayerInfo &act_info) |
| 225 | { |
| 226 | ARM_COMPUTE_RETURN_ERROR_ON(act_info.enabled()); |
| 227 | return cpu::CpuElementwisePower::validate(input1, input2, output); |
| 228 | } |
| 229 | |
| 230 | void NEElementwisePower::run() |
| 231 | { |
| 232 | ITensorPack pack; |
| 233 | pack.add_tensor(TensorType::ACL_SRC_0, _impl->src_0); |
| 234 | pack.add_tensor(TensorType::ACL_SRC_1, _impl->src_1); |
| 235 | pack.add_tensor(TensorType::ACL_DST, _impl->dst); |
| 236 | _impl->op->run(pack); |
| 237 | } |
| 238 | |
| 239 | template <ComparisonOperation COP> |
| 240 | struct NEElementwiseComparisonStatic<COP>::Impl |
| 241 | { |
| 242 | const ITensor *src_0{ nullptr }; |
| 243 | const ITensor *src_1{ nullptr }; |
| 244 | ITensor *dst{ nullptr }; |
| 245 | std::unique_ptr<cpu::CpuElementwiseComparisonStatic<COP>> op{ nullptr }; |
| 246 | }; |
| 247 | |
| 248 | template <ComparisonOperation COP> |
| 249 | NEElementwiseComparisonStatic<COP>::NEElementwiseComparisonStatic() |
| 250 | : _impl(std::make_unique<Impl>()) |
| 251 | { |
| 252 | } |
| 253 | template <ComparisonOperation COP> |
| 254 | NEElementwiseComparisonStatic<COP>::NEElementwiseComparisonStatic(NEElementwiseComparisonStatic &&) = default; |
| 255 | template <ComparisonOperation COP> |
| 256 | NEElementwiseComparisonStatic<COP> &NEElementwiseComparisonStatic<COP>::operator=(NEElementwiseComparisonStatic &&) = default; |
| 257 | template <ComparisonOperation COP> |
| 258 | NEElementwiseComparisonStatic<COP>::~NEElementwiseComparisonStatic() = default; |
| 259 | |
| 260 | template <ComparisonOperation COP> |
| 261 | void NEElementwiseComparisonStatic<COP>::configure(ITensor *input1, ITensor *input2, ITensor *output) |
| 262 | { |
| 263 | _impl->src_0 = input1; |
| 264 | _impl->src_1 = input2; |
| 265 | _impl->dst = output; |
| 266 | _impl->op = std::make_unique<cpu::CpuElementwiseComparisonStatic<COP>>(); |
| 267 | _impl->op->configure(input1->info(), input2->info(), output->info()); |
| 268 | } |
| 269 | |
| 270 | template <ComparisonOperation COP> |
| 271 | Status NEElementwiseComparisonStatic<COP>::validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output) |
| 272 | { |
| 273 | return cpu::CpuElementwiseComparisonStatic<COP>::validate(input1, input2, output); |
| 274 | } |
| 275 | |
| 276 | template <ComparisonOperation COP> |
| 277 | void NEElementwiseComparisonStatic<COP>::run() |
| 278 | { |
| 279 | ITensorPack pack; |
| 280 | pack.add_tensor(TensorType::ACL_SRC_0, _impl->src_0); |
| 281 | pack.add_tensor(TensorType::ACL_SRC_1, _impl->src_1); |
| 282 | pack.add_tensor(TensorType::ACL_DST, _impl->dst); |
| 283 | _impl->op->run(pack); |
| 284 | } |
| 285 | |
| 286 | struct NEElementwiseComparison::Impl |
| 287 | { |
| 288 | const ITensor *src_0{ nullptr }; |
| 289 | const ITensor *src_1{ nullptr }; |
| 290 | ITensor *dst{ nullptr }; |
| 291 | std::unique_ptr<cpu::CpuElementwiseComparison> op{ nullptr }; |
| 292 | }; |
| 293 | |
| 294 | NEElementwiseComparison::NEElementwiseComparison() |
| 295 | : _impl(std::make_unique<Impl>()) |
| 296 | { |
| 297 | } |
| 298 | NEElementwiseComparison::NEElementwiseComparison(NEElementwiseComparison &&) = default; |
| 299 | NEElementwiseComparison &NEElementwiseComparison::operator=(NEElementwiseComparison &&) = default; |
| 300 | NEElementwiseComparison::~NEElementwiseComparison() = default; |
| 301 | |
| 302 | void NEElementwiseComparison::configure(ITensor *input1, ITensor *input2, ITensor *output, ComparisonOperation op) |
| 303 | { |
| 304 | _impl->src_0 = input1; |
| 305 | _impl->src_1 = input2; |
| 306 | _impl->dst = output; |
| 307 | _impl->op = std::make_unique<cpu::CpuElementwiseComparison>(); |
| 308 | _impl->op->configure(input1->info(), input2->info(), output->info(), op); |
| 309 | } |
| 310 | |
| 311 | Status NEElementwiseComparison::validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, ComparisonOperation op) |
| 312 | { |
| 313 | return cpu::CpuElementwiseComparison::validate(input1, input2, output, op); |
| 314 | } |
| 315 | |
| 316 | void NEElementwiseComparison::run() |
| 317 | { |
| 318 | ITensorPack pack; |
| 319 | pack.add_tensor(TensorType::ACL_SRC_0, _impl->src_0); |
| 320 | pack.add_tensor(TensorType::ACL_SRC_1, _impl->src_1); |
| 321 | pack.add_tensor(TensorType::ACL_DST, _impl->dst); |
| 322 | _impl->op->run(pack); |
| 323 | } |
| 324 | |
| 325 | // Supported Specializations |
| 326 | template class NEElementwiseComparisonStatic<ComparisonOperation::Equal>; |
| 327 | template class NEElementwiseComparisonStatic<ComparisonOperation::NotEqual>; |
| 328 | template class NEElementwiseComparisonStatic<ComparisonOperation::Greater>; |
| 329 | template class NEElementwiseComparisonStatic<ComparisonOperation::GreaterEqual>; |
| 330 | template class NEElementwiseComparisonStatic<ComparisonOperation::Less>; |
| 331 | template class NEElementwiseComparisonStatic<ComparisonOperation::LessEqual>; |
| 332 | } // namespace arm_compute |