Aron Virginas-Tar | 7010400 | 2018-10-24 15:33:28 +0100 | [diff] [blame] | 1 | // |
| 2 | // Copyright © 2017 Arm Ltd. All rights reserved. |
| 3 | // SPDX-License-Identifier: MIT |
| 4 | // |
| 5 | #pragma once |
| 6 | |
Matthew Bentham | 246bd46 | 2020-01-20 16:16:06 +0000 | [diff] [blame] | 7 | #include <armnn/Descriptors.hpp> |
narpra01 | b9546cf | 2018-11-20 15:21:28 +0000 | [diff] [blame] | 8 | #include <armnn/INetwork.hpp> |
Matthew Bentham | 246bd46 | 2020-01-20 16:16:06 +0000 | [diff] [blame] | 9 | #include <armnn/IRuntime.hpp> |
Aron Virginas-Tar | 7010400 | 2018-10-24 15:33:28 +0100 | [diff] [blame] | 10 | |
Aron Virginas-Tar | 48623a0 | 2019-10-22 10:00:28 +0100 | [diff] [blame] | 11 | #include <Profiling.hpp> |
| 12 | #include <QuantizeHelper.hpp> |
| 13 | #include <ResolveType.hpp> |
Aron Virginas-Tar | 7010400 | 2018-10-24 15:33:28 +0100 | [diff] [blame] | 14 | |
narpra01 | b9546cf | 2018-11-20 15:21:28 +0000 | [diff] [blame] | 15 | #include <boost/test/unit_test.hpp> |
| 16 | |
Aron Virginas-Tar | 7010400 | 2018-10-24 15:33:28 +0100 | [diff] [blame] | 17 | #include <vector> |
| 18 | |
| 19 | namespace |
| 20 | { |
| 21 | |
| 22 | using namespace armnn; |
| 23 | |
| 24 | template<typename T> |
| 25 | bool ConstantUsageTest(const std::vector<BackendId>& computeDevice, |
| 26 | const TensorInfo& commonTensorInfo, |
| 27 | const std::vector<T>& inputData, |
| 28 | const std::vector<T>& constantData, |
| 29 | const std::vector<T>& expectedOutputData) |
| 30 | { |
| 31 | // Create runtime in which test will run |
| 32 | IRuntime::CreationOptions options; |
| 33 | IRuntimePtr runtime(IRuntime::Create(options)); |
| 34 | |
| 35 | // Builds up the structure of the network. |
| 36 | INetworkPtr net(INetwork::Create()); |
| 37 | |
| 38 | IConnectableLayer* input = net->AddInputLayer(0); |
| 39 | IConnectableLayer* constant = net->AddConstantLayer(ConstTensor(commonTensorInfo, constantData)); |
| 40 | IConnectableLayer* add = net->AddAdditionLayer(); |
| 41 | IConnectableLayer* output = net->AddOutputLayer(0); |
| 42 | |
| 43 | input->GetOutputSlot(0).Connect(add->GetInputSlot(0)); |
| 44 | constant->GetOutputSlot(0).Connect(add->GetInputSlot(1)); |
| 45 | add->GetOutputSlot(0).Connect(output->GetInputSlot(0)); |
| 46 | |
| 47 | // Sets the tensors in the network. |
| 48 | input->GetOutputSlot(0).SetTensorInfo(commonTensorInfo); |
| 49 | constant->GetOutputSlot(0).SetTensorInfo(commonTensorInfo); |
| 50 | add->GetOutputSlot(0).SetTensorInfo(commonTensorInfo); |
| 51 | |
| 52 | // optimize the network |
| 53 | IOptimizedNetworkPtr optNet = Optimize(*net, computeDevice, runtime->GetDeviceSpec()); |
| 54 | |
| 55 | // Loads it into the runtime. |
| 56 | NetworkId netId; |
| 57 | runtime->LoadNetwork(netId, std::move(optNet)); |
| 58 | |
| 59 | // Creates structures for input & output. |
| 60 | std::vector<T> outputData(inputData.size()); |
| 61 | |
| 62 | InputTensors inputTensors |
| 63 | { |
| 64 | {0, ConstTensor(runtime->GetInputTensorInfo(netId, 0), inputData.data())} |
| 65 | }; |
| 66 | OutputTensors outputTensors |
| 67 | { |
| 68 | {0, Tensor(runtime->GetOutputTensorInfo(netId, 0), outputData.data())} |
| 69 | }; |
| 70 | |
| 71 | // Does the inference. |
| 72 | runtime->EnqueueWorkload(netId, inputTensors, outputTensors); |
| 73 | |
| 74 | // Checks the results. |
| 75 | return outputData == expectedOutputData; |
| 76 | } |
| 77 | |
| 78 | inline bool ConstantUsageFloat32Test(const std::vector<BackendId>& backends) |
| 79 | { |
| 80 | const TensorInfo commonTensorInfo({ 2, 3 }, DataType::Float32); |
| 81 | |
| 82 | return ConstantUsageTest(backends, |
| 83 | commonTensorInfo, |
| 84 | std::vector<float>{ 1.f, 2.f, 3.f, 4.f, 5.f, 6.f }, // Input. |
| 85 | std::vector<float>{ 6.f, 5.f, 4.f, 3.f, 2.f, 1.f }, // Const input. |
| 86 | std::vector<float>{ 7.f, 7.f, 7.f, 7.f, 7.f, 7.f } // Expected output. |
| 87 | ); |
| 88 | } |
| 89 | |
| 90 | inline bool ConstantUsageUint8Test(const std::vector<BackendId>& backends) |
| 91 | { |
Derek Lamberti | f90c56d | 2020-01-10 17:14:08 +0000 | [diff] [blame] | 92 | TensorInfo commonTensorInfo({ 2, 3 }, DataType::QAsymmU8); |
Aron Virginas-Tar | 7010400 | 2018-10-24 15:33:28 +0100 | [diff] [blame] | 93 | |
| 94 | const float scale = 0.023529f; |
| 95 | const int8_t offset = -43; |
| 96 | |
| 97 | commonTensorInfo.SetQuantizationScale(scale); |
| 98 | commonTensorInfo.SetQuantizationOffset(offset); |
| 99 | |
| 100 | return ConstantUsageTest(backends, |
| 101 | commonTensorInfo, |
Aron Virginas-Tar | 48623a0 | 2019-10-22 10:00:28 +0100 | [diff] [blame] | 102 | armnnUtils::QuantizedVector<uint8_t>({ 1.f, 2.f, 3.f, 4.f, 5.f, 6.f }, scale, offset), // Input. |
| 103 | armnnUtils::QuantizedVector<uint8_t>({ 6.f, 5.f, 4.f, 3.f, 2.f, 1.f }, scale, offset), // Const input. |
| 104 | armnnUtils::QuantizedVector<uint8_t>({ 7.f, 7.f, 7.f, 7.f, 7.f, 7.f }, scale, offset) // Expected output. |
Aron Virginas-Tar | 7010400 | 2018-10-24 15:33:28 +0100 | [diff] [blame] | 105 | ); |
| 106 | } |
| 107 | |
Aron Virginas-Tar | f97f6da | 2019-10-01 18:35:44 +0100 | [diff] [blame] | 108 | // Utility template for comparing tensor elements |
| 109 | template<DataType ArmnnType, typename T = ResolveType<ArmnnType>> |
Jan Eilers | bca73e1 | 2020-03-11 12:52:46 +0000 | [diff] [blame] | 110 | bool Compare(T a, T b, float tolerance = 0.000001f) |
Nattapat Chaimanowong | 1fcb4ff | 2019-01-24 15:25:26 +0000 | [diff] [blame] | 111 | { |
Aron Virginas-Tar | f97f6da | 2019-10-01 18:35:44 +0100 | [diff] [blame] | 112 | if (ArmnnType == DataType::Boolean) |
| 113 | { |
| 114 | // NOTE: Boolean is represented as uint8_t (with zero equals |
| 115 | // false and everything else equals true), therefore values |
| 116 | // need to be casted to bool before comparing them |
| 117 | return static_cast<bool>(a) == static_cast<bool>(b); |
| 118 | } |
| 119 | |
| 120 | // NOTE: All other types can be cast to float and compared with |
| 121 | // a certain level of tolerance |
Aron Virginas-Tar | f97f6da | 2019-10-01 18:35:44 +0100 | [diff] [blame] | 122 | return std::fabs(static_cast<float>(a) - static_cast<float>(b)) <= tolerance; |
| 123 | } |
Nattapat Chaimanowong | 1fcb4ff | 2019-01-24 15:25:26 +0000 | [diff] [blame] | 124 | |
Ferran Balaguer | 83239f9 | 2019-09-19 11:49:25 +0100 | [diff] [blame] | 125 | // Utility function to find the number of instances of a substring within a string. |
| 126 | int SubStringCounter(std::string& string, std::string&& substring) |
| 127 | { |
| 128 | std::size_t found = 0; |
| 129 | int count = 0; |
| 130 | // Look for the substring starting from where we last found the substring |
| 131 | while((found = string.find(substring, found)) != std::string::npos) |
| 132 | { |
| 133 | count++; |
| 134 | // Offset by substring length to avoid finding the same substring twice |
| 135 | found += substring.length(); |
| 136 | } |
| 137 | return count; |
| 138 | } |
| 139 | |
Nattapat Chaimanowong | 1fcb4ff | 2019-01-24 15:25:26 +0000 | [diff] [blame] | 140 | template<DataType ArmnnIType, DataType ArmnnOType, |
| 141 | typename TInput = ResolveType<ArmnnIType>, typename TOutput = ResolveType<ArmnnOType>> |
narpra01 | b9546cf | 2018-11-20 15:21:28 +0000 | [diff] [blame] | 142 | void EndToEndLayerTestImpl(INetworkPtr network, |
kevmay01 | 2b4d88e | 2019-01-24 14:05:09 +0000 | [diff] [blame] | 143 | const std::map<int, std::vector<TInput>>& inputTensorData, |
| 144 | const std::map<int, std::vector<TOutput>>& expectedOutputData, |
Jan Eilers | bca73e1 | 2020-03-11 12:52:46 +0000 | [diff] [blame] | 145 | std::vector<BackendId> backends, |
| 146 | float tolerance = 0.000001f) |
narpra01 | b9546cf | 2018-11-20 15:21:28 +0000 | [diff] [blame] | 147 | { |
| 148 | // Create runtime in which test will run |
| 149 | IRuntime::CreationOptions options; |
| 150 | IRuntimePtr runtime(IRuntime::Create(options)); |
| 151 | |
| 152 | // optimize the network |
| 153 | IOptimizedNetworkPtr optNet = Optimize(*network, backends, runtime->GetDeviceSpec()); |
| 154 | |
| 155 | // Loads it into the runtime. |
| 156 | NetworkId netId; |
| 157 | runtime->LoadNetwork(netId, std::move(optNet)); |
| 158 | |
| 159 | InputTensors inputTensors; |
| 160 | inputTensors.reserve(inputTensorData.size()); |
| 161 | for (auto&& it : inputTensorData) |
| 162 | { |
| 163 | inputTensors.push_back({it.first, |
| 164 | ConstTensor(runtime->GetInputTensorInfo(netId, it.first), it.second.data())}); |
| 165 | } |
| 166 | OutputTensors outputTensors; |
| 167 | outputTensors.reserve(expectedOutputData.size()); |
kevmay01 | 2b4d88e | 2019-01-24 14:05:09 +0000 | [diff] [blame] | 168 | std::map<int, std::vector<TOutput>> outputStorage; |
narpra01 | b9546cf | 2018-11-20 15:21:28 +0000 | [diff] [blame] | 169 | for (auto&& it : expectedOutputData) |
| 170 | { |
kevmay01 | 2b4d88e | 2019-01-24 14:05:09 +0000 | [diff] [blame] | 171 | std::vector<TOutput> out(it.second.size()); |
narpra01 | b9546cf | 2018-11-20 15:21:28 +0000 | [diff] [blame] | 172 | outputStorage.emplace(it.first, out); |
| 173 | outputTensors.push_back({it.first, |
| 174 | Tensor(runtime->GetOutputTensorInfo(netId, it.first), |
| 175 | outputStorage.at(it.first).data())}); |
| 176 | } |
| 177 | |
| 178 | // Does the inference. |
| 179 | runtime->EnqueueWorkload(netId, inputTensors, outputTensors); |
| 180 | |
| 181 | // Checks the results. |
| 182 | for (auto&& it : expectedOutputData) |
| 183 | { |
kevmay01 | 2b4d88e | 2019-01-24 14:05:09 +0000 | [diff] [blame] | 184 | std::vector<TOutput> out = outputStorage.at(it.first); |
Aron Virginas-Tar | f97f6da | 2019-10-01 18:35:44 +0100 | [diff] [blame] | 185 | for (unsigned int i = 0; i < out.size(); ++i) |
Nattapat Chaimanowong | 1fcb4ff | 2019-01-24 15:25:26 +0000 | [diff] [blame] | 186 | { |
Teresa Charlin | 2e3f4d2 | 2020-07-29 14:29:20 +0100 | [diff] [blame] | 187 | BOOST_CHECK_MESSAGE(Compare<ArmnnOType>(it.second[i], out[i], tolerance) == true, |
| 188 | "Actual output: " << out[i] << ". Expected output:" << it.second[i]); |
| 189 | |
Nattapat Chaimanowong | 1fcb4ff | 2019-01-24 15:25:26 +0000 | [diff] [blame] | 190 | } |
narpra01 | b9546cf | 2018-11-20 15:21:28 +0000 | [diff] [blame] | 191 | } |
| 192 | } |
| 193 | |
David Monahan | 4f1e8e4 | 2019-09-04 09:22:10 +0100 | [diff] [blame] | 194 | inline void ImportNonAlignedInputPointerTest(std::vector<BackendId> backends) |
Ferran Balaguer | dcaa610 | 2019-08-21 13:28:38 +0100 | [diff] [blame] | 195 | { |
| 196 | using namespace armnn; |
| 197 | |
| 198 | // Create runtime in which test will run |
| 199 | IRuntime::CreationOptions options; |
| 200 | IRuntimePtr runtime(armnn::IRuntime::Create(options)); |
| 201 | |
| 202 | // build up the structure of the network |
| 203 | INetworkPtr net(INetwork::Create()); |
| 204 | |
| 205 | IConnectableLayer* input = net->AddInputLayer(0); |
| 206 | |
David Monahan | 3fb7e10 | 2019-08-20 11:25:29 +0100 | [diff] [blame] | 207 | ActivationDescriptor descriptor; |
| 208 | descriptor.m_Function = ActivationFunction::Square; |
| 209 | IConnectableLayer* pooling = net->AddActivationLayer(descriptor); |
Ferran Balaguer | dcaa610 | 2019-08-21 13:28:38 +0100 | [diff] [blame] | 210 | |
| 211 | IConnectableLayer* output = net->AddOutputLayer(0); |
| 212 | |
David Monahan | 3fb7e10 | 2019-08-20 11:25:29 +0100 | [diff] [blame] | 213 | input->GetOutputSlot(0).Connect(pooling->GetInputSlot(0)); |
| 214 | pooling->GetOutputSlot(0).Connect(output->GetInputSlot(0)); |
Ferran Balaguer | dcaa610 | 2019-08-21 13:28:38 +0100 | [diff] [blame] | 215 | |
David Monahan | 3fb7e10 | 2019-08-20 11:25:29 +0100 | [diff] [blame] | 216 | input->GetOutputSlot(0).SetTensorInfo(TensorInfo({ 1, 1, 1, 4 }, DataType::Float32)); |
| 217 | pooling->GetOutputSlot(0).SetTensorInfo(TensorInfo({ 1, 1, 1, 4 }, DataType::Float32)); |
Ferran Balaguer | dcaa610 | 2019-08-21 13:28:38 +0100 | [diff] [blame] | 218 | |
| 219 | // Optimize the network |
| 220 | IOptimizedNetworkPtr optNet = Optimize(*net, backends, runtime->GetDeviceSpec()); |
David Monahan | 3fb7e10 | 2019-08-20 11:25:29 +0100 | [diff] [blame] | 221 | BOOST_CHECK(optNet); |
Ferran Balaguer | dcaa610 | 2019-08-21 13:28:38 +0100 | [diff] [blame] | 222 | |
| 223 | // Loads it into the runtime. |
| 224 | NetworkId netId; |
David Monahan | 4f1e8e4 | 2019-09-04 09:22:10 +0100 | [diff] [blame] | 225 | std::string ignoredErrorMessage; |
| 226 | // Enable Importing |
David Monahan | 3fb7e10 | 2019-08-20 11:25:29 +0100 | [diff] [blame] | 227 | INetworkProperties networkProperties(true, false); |
David Monahan | 4f1e8e4 | 2019-09-04 09:22:10 +0100 | [diff] [blame] | 228 | runtime->LoadNetwork(netId, std::move(optNet), ignoredErrorMessage, networkProperties); |
Ferran Balaguer | dcaa610 | 2019-08-21 13:28:38 +0100 | [diff] [blame] | 229 | |
| 230 | // Creates structures for input & output |
| 231 | std::vector<float> inputData |
| 232 | { |
David Monahan | 3fb7e10 | 2019-08-20 11:25:29 +0100 | [diff] [blame] | 233 | 1.0f, 2.0f, 3.0f, 4.0f |
Ferran Balaguer | dcaa610 | 2019-08-21 13:28:38 +0100 | [diff] [blame] | 234 | }; |
| 235 | |
| 236 | // Misaligned input |
Aron Virginas-Tar | d9f7c8b | 2019-09-13 13:37:03 +0100 | [diff] [blame] | 237 | float* misalignedInputData = reinterpret_cast<float*>(reinterpret_cast<char*>(inputData.data()) + 1); |
Ferran Balaguer | dcaa610 | 2019-08-21 13:28:38 +0100 | [diff] [blame] | 238 | |
David Monahan | 3fb7e10 | 2019-08-20 11:25:29 +0100 | [diff] [blame] | 239 | std::vector<float> outputData(4); |
Ferran Balaguer | dcaa610 | 2019-08-21 13:28:38 +0100 | [diff] [blame] | 240 | |
David Monahan | 4f1e8e4 | 2019-09-04 09:22:10 +0100 | [diff] [blame] | 241 | // Aligned output |
David Monahan | 3fb7e10 | 2019-08-20 11:25:29 +0100 | [diff] [blame] | 242 | float* alignedOutputData = outputData.data(); |
David Monahan | 4f1e8e4 | 2019-09-04 09:22:10 +0100 | [diff] [blame] | 243 | |
| 244 | InputTensors inputTensors |
| 245 | { |
| 246 | {0,armnn::ConstTensor(runtime->GetInputTensorInfo(netId, 0), misalignedInputData)}, |
| 247 | }; |
| 248 | OutputTensors outputTensors |
| 249 | { |
| 250 | {0,armnn::Tensor(runtime->GetOutputTensorInfo(netId, 0), alignedOutputData)} |
| 251 | }; |
| 252 | |
David Monahan | 4f1e8e4 | 2019-09-04 09:22:10 +0100 | [diff] [blame] | 253 | runtime->GetProfiler(netId)->EnableProfiling(true); |
| 254 | |
| 255 | // Do the inference and expect it to fail with a ImportMemoryException |
| 256 | BOOST_CHECK_THROW(runtime->EnqueueWorkload(netId, inputTensors, outputTensors), MemoryImportException); |
| 257 | } |
| 258 | |
Ferran Balaguer | 83239f9 | 2019-09-19 11:49:25 +0100 | [diff] [blame] | 259 | inline void ExportNonAlignedOutputPointerTest(std::vector<BackendId> backends) |
David Monahan | 4f1e8e4 | 2019-09-04 09:22:10 +0100 | [diff] [blame] | 260 | { |
| 261 | using namespace armnn; |
| 262 | |
| 263 | // Create runtime in which test will run |
| 264 | IRuntime::CreationOptions options; |
| 265 | IRuntimePtr runtime(armnn::IRuntime::Create(options)); |
| 266 | |
| 267 | // build up the structure of the network |
| 268 | INetworkPtr net(INetwork::Create()); |
| 269 | |
| 270 | IConnectableLayer* input = net->AddInputLayer(0); |
| 271 | |
David Monahan | 3fb7e10 | 2019-08-20 11:25:29 +0100 | [diff] [blame] | 272 | ActivationDescriptor descriptor; |
| 273 | descriptor.m_Function = ActivationFunction::Square; |
| 274 | IConnectableLayer* pooling = net->AddActivationLayer(descriptor); |
David Monahan | 4f1e8e4 | 2019-09-04 09:22:10 +0100 | [diff] [blame] | 275 | |
| 276 | IConnectableLayer* output = net->AddOutputLayer(0); |
| 277 | |
David Monahan | 3fb7e10 | 2019-08-20 11:25:29 +0100 | [diff] [blame] | 278 | input->GetOutputSlot(0).Connect(pooling->GetInputSlot(0)); |
| 279 | pooling->GetOutputSlot(0).Connect(output->GetInputSlot(0)); |
David Monahan | 4f1e8e4 | 2019-09-04 09:22:10 +0100 | [diff] [blame] | 280 | |
David Monahan | 3fb7e10 | 2019-08-20 11:25:29 +0100 | [diff] [blame] | 281 | input->GetOutputSlot(0).SetTensorInfo(TensorInfo({ 1, 1, 1, 4 }, DataType::Float32)); |
| 282 | pooling->GetOutputSlot(0).SetTensorInfo(TensorInfo({ 1, 1, 1, 4 }, DataType::Float32)); |
David Monahan | 4f1e8e4 | 2019-09-04 09:22:10 +0100 | [diff] [blame] | 283 | |
| 284 | // Optimize the network |
| 285 | IOptimizedNetworkPtr optNet = Optimize(*net, backends, runtime->GetDeviceSpec()); |
David Monahan | 3fb7e10 | 2019-08-20 11:25:29 +0100 | [diff] [blame] | 286 | BOOST_CHECK(optNet); |
David Monahan | 4f1e8e4 | 2019-09-04 09:22:10 +0100 | [diff] [blame] | 287 | |
| 288 | // Loads it into the runtime. |
| 289 | NetworkId netId; |
| 290 | std::string ignoredErrorMessage; |
David Monahan | 3fb7e10 | 2019-08-20 11:25:29 +0100 | [diff] [blame] | 291 | // Enable Importing and Exporting |
David Monahan | 4f1e8e4 | 2019-09-04 09:22:10 +0100 | [diff] [blame] | 292 | INetworkProperties networkProperties(true, true); |
| 293 | runtime->LoadNetwork(netId, std::move(optNet), ignoredErrorMessage, networkProperties); |
| 294 | |
| 295 | // Creates structures for input & output |
| 296 | std::vector<float> inputData |
| 297 | { |
| 298 | 1.0f, 2.0f, 3.0f, 4.0f, 5.0f |
| 299 | }; |
| 300 | |
| 301 | // Aligned input |
David Monahan | 3fb7e10 | 2019-08-20 11:25:29 +0100 | [diff] [blame] | 302 | float* alignedInputData = inputData.data(); |
David Monahan | 4f1e8e4 | 2019-09-04 09:22:10 +0100 | [diff] [blame] | 303 | |
| 304 | std::vector<float> outputData(5); |
| 305 | |
Ferran Balaguer | dcaa610 | 2019-08-21 13:28:38 +0100 | [diff] [blame] | 306 | // Misaligned output |
Aron Virginas-Tar | d9f7c8b | 2019-09-13 13:37:03 +0100 | [diff] [blame] | 307 | float* misalignedOutputData = reinterpret_cast<float*>(reinterpret_cast<char*>(outputData.data()) + 1); |
Ferran Balaguer | dcaa610 | 2019-08-21 13:28:38 +0100 | [diff] [blame] | 308 | |
| 309 | InputTensors inputTensors |
| 310 | { |
David Monahan | 4f1e8e4 | 2019-09-04 09:22:10 +0100 | [diff] [blame] | 311 | {0,armnn::ConstTensor(runtime->GetInputTensorInfo(netId, 0), alignedInputData)}, |
Ferran Balaguer | dcaa610 | 2019-08-21 13:28:38 +0100 | [diff] [blame] | 312 | }; |
| 313 | OutputTensors outputTensors |
| 314 | { |
| 315 | {0,armnn::Tensor(runtime->GetOutputTensorInfo(netId, 0), misalignedOutputData)} |
| 316 | }; |
| 317 | |
Ferran Balaguer | 83239f9 | 2019-09-19 11:49:25 +0100 | [diff] [blame] | 318 | // Do the inference and expect it to fail with a ExportMemoryException |
| 319 | if (backends[0] == Compute::CpuAcc) |
| 320 | { |
| 321 | // For CpuAcc the NeonTensorHandle will throw its own exception on misaligned memory |
| 322 | BOOST_CHECK_THROW(runtime->EnqueueWorkload(netId, inputTensors, outputTensors), MemoryImportException); |
| 323 | } |
| 324 | else |
| 325 | { |
| 326 | BOOST_CHECK_THROW(runtime->EnqueueWorkload(netId, inputTensors, outputTensors), MemoryExportException); |
| 327 | } |
Ferran Balaguer | dcaa610 | 2019-08-21 13:28:38 +0100 | [diff] [blame] | 328 | } |
| 329 | |
| 330 | inline void ImportAlignedPointerTest(std::vector<BackendId> backends) |
| 331 | { |
| 332 | using namespace armnn; |
| 333 | |
| 334 | // Create runtime in which test will run |
| 335 | IRuntime::CreationOptions options; |
| 336 | IRuntimePtr runtime(armnn::IRuntime::Create(options)); |
| 337 | |
| 338 | // build up the structure of the network |
| 339 | INetworkPtr net(INetwork::Create()); |
| 340 | |
| 341 | IConnectableLayer* input = net->AddInputLayer(0); |
| 342 | |
David Monahan | 3fb7e10 | 2019-08-20 11:25:29 +0100 | [diff] [blame] | 343 | ActivationDescriptor descriptor; |
| 344 | descriptor.m_Function = ActivationFunction::Square; |
| 345 | IConnectableLayer* pooling = net->AddActivationLayer(descriptor); |
Ferran Balaguer | dcaa610 | 2019-08-21 13:28:38 +0100 | [diff] [blame] | 346 | |
| 347 | IConnectableLayer* output = net->AddOutputLayer(0); |
| 348 | |
David Monahan | 3fb7e10 | 2019-08-20 11:25:29 +0100 | [diff] [blame] | 349 | input->GetOutputSlot(0).Connect(pooling->GetInputSlot(0)); |
| 350 | pooling->GetOutputSlot(0).Connect(output->GetInputSlot(0)); |
Ferran Balaguer | dcaa610 | 2019-08-21 13:28:38 +0100 | [diff] [blame] | 351 | |
David Monahan | 3fb7e10 | 2019-08-20 11:25:29 +0100 | [diff] [blame] | 352 | input->GetOutputSlot(0).SetTensorInfo(TensorInfo({ 1, 1, 1, 4 }, DataType::Float32)); |
| 353 | pooling->GetOutputSlot(0).SetTensorInfo(TensorInfo({ 1, 1, 1, 4 }, DataType::Float32)); |
Ferran Balaguer | dcaa610 | 2019-08-21 13:28:38 +0100 | [diff] [blame] | 354 | |
| 355 | // Optimize the network |
| 356 | IOptimizedNetworkPtr optNet = Optimize(*net, backends, runtime->GetDeviceSpec()); |
David Monahan | 3fb7e10 | 2019-08-20 11:25:29 +0100 | [diff] [blame] | 357 | BOOST_CHECK(optNet); |
Ferran Balaguer | dcaa610 | 2019-08-21 13:28:38 +0100 | [diff] [blame] | 358 | |
| 359 | // Loads it into the runtime. |
| 360 | NetworkId netId; |
David Monahan | 4f1e8e4 | 2019-09-04 09:22:10 +0100 | [diff] [blame] | 361 | std::string ignoredErrorMessage; |
| 362 | // Enable Importing |
| 363 | INetworkProperties networkProperties(true, true); |
| 364 | runtime->LoadNetwork(netId, std::move(optNet), ignoredErrorMessage, networkProperties); |
Ferran Balaguer | dcaa610 | 2019-08-21 13:28:38 +0100 | [diff] [blame] | 365 | |
| 366 | // Creates structures for input & output |
| 367 | std::vector<float> inputData |
| 368 | { |
| 369 | 1.0f, 2.0f, 3.0f, 4.0f |
| 370 | }; |
| 371 | |
| 372 | std::vector<float> outputData(4); |
| 373 | |
James Conroy | 57d10b7 | 2019-10-25 09:44:14 +0100 | [diff] [blame] | 374 | std::vector<float> expectedOutput |
| 375 | { |
| 376 | 1.0f, 4.0f, 9.0f, 16.0f |
| 377 | }; |
| 378 | |
Ferran Balaguer | dcaa610 | 2019-08-21 13:28:38 +0100 | [diff] [blame] | 379 | InputTensors inputTensors |
| 380 | { |
| 381 | {0,armnn::ConstTensor(runtime->GetInputTensorInfo(netId, 0), inputData.data())}, |
| 382 | }; |
| 383 | OutputTensors outputTensors |
| 384 | { |
| 385 | {0,armnn::Tensor(runtime->GetOutputTensorInfo(netId, 0), outputData.data())} |
| 386 | }; |
| 387 | |
Ferran Balaguer | dcaa610 | 2019-08-21 13:28:38 +0100 | [diff] [blame] | 388 | runtime->GetProfiler(netId)->EnableProfiling(true); |
| 389 | |
| 390 | // Do the inference |
| 391 | runtime->EnqueueWorkload(netId, inputTensors, outputTensors); |
| 392 | |
| 393 | // Retrieve the Profiler.Print() output to get the workload execution |
| 394 | ProfilerManager& profilerManager = armnn::ProfilerManager::GetInstance(); |
| 395 | std::stringstream ss; |
| 396 | profilerManager.GetProfiler()->Print(ss);; |
| 397 | std::string dump = ss.str(); |
| 398 | |
David Monahan | 3fb7e10 | 2019-08-20 11:25:29 +0100 | [diff] [blame] | 399 | // Contains ActivationWorkload |
| 400 | std::size_t found = dump.find("ActivationWorkload"); |
Ferran Balaguer | dcaa610 | 2019-08-21 13:28:38 +0100 | [diff] [blame] | 401 | BOOST_TEST(found != std::string::npos); |
James Conroy | 57d10b7 | 2019-10-25 09:44:14 +0100 | [diff] [blame] | 402 | |
Ferran Balaguer | dcaa610 | 2019-08-21 13:28:38 +0100 | [diff] [blame] | 403 | // Contains SyncMemGeneric |
| 404 | found = dump.find("SyncMemGeneric"); |
| 405 | BOOST_TEST(found != std::string::npos); |
James Conroy | 57d10b7 | 2019-10-25 09:44:14 +0100 | [diff] [blame] | 406 | |
Ferran Balaguer | 83239f9 | 2019-09-19 11:49:25 +0100 | [diff] [blame] | 407 | // Does not contain CopyMemGeneric |
Ferran Balaguer | dcaa610 | 2019-08-21 13:28:38 +0100 | [diff] [blame] | 408 | found = dump.find("CopyMemGeneric"); |
| 409 | BOOST_TEST(found == std::string::npos); |
James Conroy | 57d10b7 | 2019-10-25 09:44:14 +0100 | [diff] [blame] | 410 | |
| 411 | // Check output is as expected |
| 412 | BOOST_TEST(outputData == expectedOutput); |
Ferran Balaguer | dcaa610 | 2019-08-21 13:28:38 +0100 | [diff] [blame] | 413 | } |
| 414 | |
Ferran Balaguer | 83239f9 | 2019-09-19 11:49:25 +0100 | [diff] [blame] | 415 | inline void ImportOnlyWorkload(std::vector<BackendId> backends) |
| 416 | { |
| 417 | using namespace armnn; |
| 418 | |
| 419 | IRuntime::CreationOptions options; |
| 420 | IRuntimePtr runtime(IRuntime::Create(options)); |
| 421 | |
| 422 | // Builds up the structure of the network. |
| 423 | INetworkPtr net(INetwork::Create()); |
| 424 | |
| 425 | IConnectableLayer* input = net->AddInputLayer(0); |
| 426 | |
| 427 | ActivationDescriptor descriptor; |
| 428 | descriptor.m_Function = ActivationFunction::Square; |
| 429 | IConnectableLayer* pooling = net->AddActivationLayer(descriptor); |
| 430 | |
| 431 | IConnectableLayer* output = net->AddOutputLayer(0); |
| 432 | |
| 433 | input->GetOutputSlot(0).Connect(pooling->GetInputSlot(0)); |
| 434 | pooling->GetOutputSlot(0).Connect(output->GetInputSlot(0)); |
| 435 | |
| 436 | input->GetOutputSlot(0).SetTensorInfo(TensorInfo({ 1, 1, 1, 4 }, DataType::Float32)); |
| 437 | pooling->GetOutputSlot(0).SetTensorInfo(TensorInfo({ 1, 1, 1, 4 }, DataType::Float32)); |
| 438 | |
| 439 | // optimize the network |
| 440 | IOptimizedNetworkPtr optNet = Optimize(*net, backends, runtime->GetDeviceSpec()); |
| 441 | |
| 442 | BOOST_TEST_CHECKPOINT("Load Network"); |
| 443 | // Load it into the runtime. It should pass. |
| 444 | NetworkId netId; |
| 445 | std::string ignoredErrorMessage; |
| 446 | INetworkProperties networkProperties(true, false); |
| 447 | BOOST_TEST(runtime->LoadNetwork(netId, std::move(optNet),ignoredErrorMessage, networkProperties) |
| 448 | == Status::Success); |
| 449 | |
| 450 | BOOST_TEST_CHECKPOINT("Generate Data"); |
| 451 | // Creates structures for input & output |
| 452 | std::vector<float> inputData |
| 453 | { |
| 454 | 1.0f, 2.0f, 3.0f, 4.0f |
| 455 | }; |
| 456 | |
| 457 | std::vector<float> outputData(4); |
| 458 | |
| 459 | std::vector<float> expectedOutput |
| 460 | { |
| 461 | 1.0f, 4.0f, 9.0f, 16.0f |
| 462 | }; |
| 463 | |
| 464 | BOOST_TEST_CHECKPOINT("Create Network"); |
| 465 | InputTensors inputTensors |
| 466 | { |
| 467 | {0,armnn::ConstTensor(runtime->GetInputTensorInfo(netId, 0), inputData.data())}, |
| 468 | }; |
| 469 | OutputTensors outputTensors |
| 470 | { |
| 471 | {0,armnn::Tensor(runtime->GetOutputTensorInfo(netId, 0), outputData.data())} |
| 472 | }; |
| 473 | |
| 474 | BOOST_TEST_CHECKPOINT("Get Profiler"); |
| 475 | |
| 476 | runtime->GetProfiler(netId)->EnableProfiling(true); |
| 477 | |
| 478 | BOOST_TEST_CHECKPOINT("Run Inference"); |
| 479 | // Do the inference |
| 480 | runtime->EnqueueWorkload(netId, inputTensors, outputTensors); |
| 481 | |
| 482 | BOOST_TEST_CHECKPOINT("Print Profiler"); |
| 483 | // Retrieve the Profiler.Print() output to get the workload execution |
| 484 | ProfilerManager& profilerManager = armnn::ProfilerManager::GetInstance(); |
| 485 | std::stringstream ss; |
| 486 | profilerManager.GetProfiler()->Print(ss);; |
| 487 | std::string dump = ss.str(); |
| 488 | |
| 489 | // Check there are no SyncMemGeneric workloads as we didn't export |
| 490 | BOOST_TEST_CHECKPOINT("Find SyncMemGeneric"); |
| 491 | int count = SubStringCounter(dump, "SyncMemGeneric"); |
| 492 | BOOST_TEST(count == 0); |
| 493 | |
| 494 | // Should only be 1 CopyMemGeneric for the output as we imported |
| 495 | BOOST_TEST_CHECKPOINT("Find CopyMemGeneric"); |
| 496 | count = SubStringCounter(dump, "CopyMemGeneric"); |
| 497 | BOOST_TEST(count == 1); |
| 498 | |
| 499 | // Check the output is correct |
| 500 | BOOST_CHECK_EQUAL_COLLECTIONS(outputData.begin(), outputData.end(), expectedOutput.begin(), expectedOutput.end()); |
| 501 | } |
| 502 | |
| 503 | inline void ExportOnlyWorkload(std::vector<BackendId> backends) |
| 504 | { |
| 505 | using namespace armnn; |
| 506 | |
| 507 | IRuntime::CreationOptions options; |
| 508 | IRuntimePtr runtime(IRuntime::Create(options)); |
| 509 | |
| 510 | // Builds up the structure of the network. |
| 511 | INetworkPtr net(INetwork::Create()); |
| 512 | |
| 513 | IConnectableLayer* input = net->AddInputLayer(0); |
| 514 | |
| 515 | ActivationDescriptor descriptor; |
| 516 | descriptor.m_Function = ActivationFunction::Square; |
| 517 | IConnectableLayer* pooling = net->AddActivationLayer(descriptor); |
| 518 | |
| 519 | IConnectableLayer* output = net->AddOutputLayer(0); |
| 520 | |
| 521 | input->GetOutputSlot(0).Connect(pooling->GetInputSlot(0)); |
| 522 | pooling->GetOutputSlot(0).Connect(output->GetInputSlot(0)); |
| 523 | |
| 524 | input->GetOutputSlot(0).SetTensorInfo(TensorInfo({ 1, 1, 1, 4 }, DataType::Float32)); |
| 525 | pooling->GetOutputSlot(0).SetTensorInfo(TensorInfo({ 1, 1, 1, 4 }, DataType::Float32)); |
| 526 | |
| 527 | // optimize the network |
| 528 | IOptimizedNetworkPtr optNet = Optimize(*net, backends, runtime->GetDeviceSpec()); |
| 529 | |
| 530 | BOOST_TEST_CHECKPOINT("Load Network"); |
| 531 | // Load it into the runtime. It should pass. |
| 532 | NetworkId netId; |
| 533 | std::string ignoredErrorMessage; |
| 534 | INetworkProperties networkProperties(false, true); |
| 535 | BOOST_TEST(runtime->LoadNetwork(netId, std::move(optNet),ignoredErrorMessage, networkProperties) |
| 536 | == Status::Success); |
| 537 | |
| 538 | BOOST_TEST_CHECKPOINT("Generate Data"); |
| 539 | // Creates structures for input & output |
| 540 | std::vector<float> inputData |
| 541 | { |
| 542 | 1.0f, 2.0f, 3.0f, 4.0f |
| 543 | }; |
| 544 | |
| 545 | std::vector<float> outputData(4); |
| 546 | |
| 547 | std::vector<float> expectedOutput |
| 548 | { |
| 549 | 1.0f, 4.0f, 9.0f, 16.0f |
| 550 | }; |
| 551 | |
| 552 | BOOST_TEST_CHECKPOINT("Create Network"); |
| 553 | InputTensors inputTensors |
| 554 | { |
| 555 | {0,armnn::ConstTensor(runtime->GetInputTensorInfo(netId, 0), inputData.data())}, |
| 556 | }; |
| 557 | OutputTensors outputTensors |
| 558 | { |
| 559 | {0,armnn::Tensor(runtime->GetOutputTensorInfo(netId, 0), outputData.data())} |
| 560 | }; |
| 561 | |
| 562 | BOOST_TEST_CHECKPOINT("Get Profiler"); |
| 563 | |
| 564 | runtime->GetProfiler(netId)->EnableProfiling(true); |
| 565 | |
| 566 | BOOST_TEST_CHECKPOINT("Run Inference"); |
| 567 | // Do the inference |
| 568 | runtime->EnqueueWorkload(netId, inputTensors, outputTensors); |
| 569 | |
| 570 | BOOST_TEST_CHECKPOINT("Print Profiler"); |
| 571 | // Retrieve the Profiler.Print() output to get the workload execution |
| 572 | ProfilerManager& profilerManager = armnn::ProfilerManager::GetInstance(); |
| 573 | std::stringstream ss; |
| 574 | profilerManager.GetProfiler()->Print(ss);; |
| 575 | std::string dump = ss.str(); |
| 576 | |
| 577 | // Check there is a SyncMemGeneric workload as we exported |
| 578 | BOOST_TEST_CHECKPOINT("Find SyncMemGeneric"); |
| 579 | int count = SubStringCounter(dump, "SyncMemGeneric"); |
| 580 | BOOST_TEST(count == 1); |
| 581 | |
| 582 | // Should be 1 CopyMemGeneric for the output as we did not import |
| 583 | BOOST_TEST_CHECKPOINT("Find CopyMemGeneric"); |
| 584 | count = SubStringCounter(dump, "CopyMemGeneric"); |
| 585 | BOOST_TEST(count == 1); |
| 586 | |
| 587 | // Check the output is correct |
| 588 | BOOST_CHECK_EQUAL_COLLECTIONS(outputData.begin(), outputData.end(), expectedOutput.begin(), expectedOutput.end()); |
| 589 | } |
| 590 | |
| 591 | inline void ImportAndExportWorkload(std::vector<BackendId> backends) |
| 592 | { |
| 593 | using namespace armnn; |
| 594 | |
| 595 | IRuntime::CreationOptions options; |
| 596 | IRuntimePtr runtime(IRuntime::Create(options)); |
| 597 | |
| 598 | // Builds up the structure of the network. |
| 599 | INetworkPtr net(INetwork::Create()); |
| 600 | |
| 601 | IConnectableLayer* input = net->AddInputLayer(0); |
| 602 | |
| 603 | ActivationDescriptor descriptor; |
| 604 | descriptor.m_Function = ActivationFunction::Square; |
| 605 | IConnectableLayer* pooling = net->AddActivationLayer(descriptor); |
| 606 | |
| 607 | IConnectableLayer* output = net->AddOutputLayer(0); |
| 608 | |
| 609 | input->GetOutputSlot(0).Connect(pooling->GetInputSlot(0)); |
| 610 | pooling->GetOutputSlot(0).Connect(output->GetInputSlot(0)); |
| 611 | |
| 612 | input->GetOutputSlot(0).SetTensorInfo(TensorInfo({ 1, 1, 1, 4 }, DataType::Float32)); |
| 613 | pooling->GetOutputSlot(0).SetTensorInfo(TensorInfo({ 1, 1, 1, 4 }, DataType::Float32)); |
| 614 | |
| 615 | IOptimizedNetworkPtr optNet = Optimize(*net, backends, runtime->GetDeviceSpec()); |
| 616 | |
| 617 | BOOST_TEST_CHECKPOINT("Load Network"); |
| 618 | // Load it into the runtime. It should pass. |
| 619 | NetworkId netId; |
| 620 | std::string ignoredErrorMessage; |
| 621 | INetworkProperties networkProperties(true, true); |
| 622 | BOOST_TEST(runtime->LoadNetwork(netId, std::move(optNet),ignoredErrorMessage, networkProperties) |
| 623 | == Status::Success); |
| 624 | |
| 625 | BOOST_TEST_CHECKPOINT("Generate Data"); |
| 626 | // Creates structures for input & output |
| 627 | std::vector<float> inputData |
| 628 | { |
| 629 | 1.0f, 2.0f, 3.0f, 4.0f |
| 630 | }; |
| 631 | |
| 632 | std::vector<float> outputData(4); |
| 633 | |
| 634 | std::vector<float> expectedOutput |
| 635 | { |
| 636 | 1.0f, 4.0f, 9.0f, 16.0f |
| 637 | }; |
| 638 | |
| 639 | BOOST_TEST_CHECKPOINT("Create Network"); |
| 640 | InputTensors inputTensors |
| 641 | { |
| 642 | {0,armnn::ConstTensor(runtime->GetInputTensorInfo(netId, 0), inputData.data())}, |
| 643 | }; |
| 644 | OutputTensors outputTensors |
| 645 | { |
| 646 | {0,armnn::Tensor(runtime->GetOutputTensorInfo(netId, 0), outputData.data())} |
| 647 | }; |
| 648 | |
| 649 | BOOST_TEST_CHECKPOINT("Get Profiler"); |
| 650 | |
| 651 | runtime->GetProfiler(netId)->EnableProfiling(true); |
| 652 | |
| 653 | BOOST_TEST_CHECKPOINT("Run Inference"); |
| 654 | // Do the inference |
| 655 | runtime->EnqueueWorkload(netId, inputTensors, outputTensors); |
| 656 | |
| 657 | BOOST_TEST_CHECKPOINT("Print Profiler"); |
| 658 | // Retrieve the Profiler.Print() output to get the workload execution |
| 659 | ProfilerManager& profilerManager = armnn::ProfilerManager::GetInstance(); |
| 660 | std::stringstream ss; |
| 661 | profilerManager.GetProfiler()->Print(ss);; |
| 662 | std::string dump = ss.str(); |
| 663 | |
| 664 | // Check there is a SyncMemGeneric workload as we exported |
| 665 | BOOST_TEST_CHECKPOINT("Find SyncMemGeneric"); |
| 666 | int count = SubStringCounter(dump, "SyncMemGeneric"); |
| 667 | BOOST_TEST(count == 1); |
| 668 | |
| 669 | // Shouldn't be any CopyMemGeneric workloads |
| 670 | BOOST_TEST_CHECKPOINT("Find CopyMemGeneric"); |
| 671 | count = SubStringCounter(dump, "CopyMemGeneric"); |
| 672 | BOOST_TEST(count == 0); |
| 673 | |
| 674 | // Check the output is correct |
| 675 | BOOST_CHECK_EQUAL_COLLECTIONS(outputData.begin(), outputData.end(), expectedOutput.begin(), expectedOutput.end()); |
| 676 | } |
| 677 | |
| 678 | inline void ExportOutputWithSeveralOutputSlotConnectionsTest(std::vector<BackendId> backends) |
| 679 | { |
| 680 | using namespace armnn; |
| 681 | |
| 682 | // Create runtime in which test will run |
| 683 | IRuntime::CreationOptions options; |
| 684 | IRuntimePtr runtime(armnn::IRuntime::Create(options)); |
| 685 | |
| 686 | // build up the structure of the network |
| 687 | INetworkPtr net(INetwork::Create()); |
| 688 | |
| 689 | IConnectableLayer* input = net->AddInputLayer(0); |
| 690 | |
| 691 | ActivationDescriptor descriptor; |
| 692 | descriptor.m_Function = ActivationFunction::Square; |
| 693 | IConnectableLayer* activation = net->AddActivationLayer(descriptor); |
| 694 | |
| 695 | IConnectableLayer* output0 = net->AddOutputLayer(0); |
| 696 | IConnectableLayer* output1 = net->AddOutputLayer(1); |
| 697 | |
| 698 | input->GetOutputSlot(0).Connect(activation->GetInputSlot(0)); |
| 699 | activation->GetOutputSlot(0).Connect(output0->GetInputSlot(0)); |
| 700 | activation->GetOutputSlot(0).Connect(output1->GetInputSlot(0)); |
| 701 | |
| 702 | input->GetOutputSlot(0).SetTensorInfo(TensorInfo({ 1, 1, 4, 1 }, DataType::Float32)); |
| 703 | activation->GetOutputSlot(0).SetTensorInfo(TensorInfo({ 1, 1, 4, 1 }, DataType::Float32)); |
| 704 | |
| 705 | // Optimize the network |
| 706 | IOptimizedNetworkPtr optNet = Optimize(*net, backends, runtime->GetDeviceSpec()); |
| 707 | |
| 708 | // Loads it into the runtime. |
| 709 | NetworkId netId; |
| 710 | std::string ignoredErrorMessage; |
| 711 | // Enable Importing |
| 712 | INetworkProperties networkProperties(true, true); |
| 713 | runtime->LoadNetwork(netId, std::move(optNet), ignoredErrorMessage, networkProperties); |
| 714 | |
| 715 | // Creates structures for input & output |
| 716 | std::vector<float> inputData |
| 717 | { |
| 718 | 1.0f, 2.0f, 3.0f, 4.0f |
| 719 | }; |
| 720 | |
| 721 | std::vector<float> outputData0(4); |
| 722 | std::vector<float> outputData1(4); |
| 723 | |
Narumol Prangnawarat | 3b90af6 | 2020-06-26 11:00:21 +0100 | [diff] [blame] | 724 | std::vector<float> expectedOutput |
| 725 | { |
| 726 | 1.0f, 4.0f, 9.0f, 16.0f |
| 727 | }; |
| 728 | |
Ferran Balaguer | 83239f9 | 2019-09-19 11:49:25 +0100 | [diff] [blame] | 729 | InputTensors inputTensors |
| 730 | { |
| 731 | {0,armnn::ConstTensor(runtime->GetInputTensorInfo(netId, 0), inputData.data())}, |
| 732 | }; |
| 733 | OutputTensors outputTensors |
| 734 | { |
| 735 | {0,armnn::Tensor(runtime->GetOutputTensorInfo(netId, 0), outputData0.data())}, |
| 736 | {1,armnn::Tensor(runtime->GetOutputTensorInfo(netId, 1), outputData1.data())} |
| 737 | }; |
| 738 | |
| 739 | // The result of the inference is not important, just the fact that there |
| 740 | // should not be CopyMemGeneric workloads. |
| 741 | runtime->GetProfiler(netId)->EnableProfiling(true); |
| 742 | |
| 743 | // Do the inference |
| 744 | runtime->EnqueueWorkload(netId, inputTensors, outputTensors); |
| 745 | |
| 746 | // Retrieve the Profiler.Print() output to get the workload execution |
| 747 | ProfilerManager& profilerManager = armnn::ProfilerManager::GetInstance(); |
| 748 | std::stringstream ss; |
| 749 | profilerManager.GetProfiler()->Print(ss); |
| 750 | std::string dump = ss.str(); |
| 751 | |
| 752 | std::size_t found = std::string::npos; |
| 753 | |
| 754 | if (backends[0] == Compute::CpuRef) |
| 755 | { |
| 756 | found = dump.find("RefActivationWorkload"); |
| 757 | } |
| 758 | else if (backends[0] == Compute::CpuAcc) |
| 759 | { |
| 760 | found = dump.find("NeonActivationWorkload"); |
| 761 | } |
| 762 | else if (backends[0] == Compute::GpuAcc) |
| 763 | { |
| 764 | found = dump.find("ClActivationWorkload"); |
| 765 | } |
| 766 | |
| 767 | BOOST_TEST(found != std::string::npos); |
| 768 | // No contains SyncMemGeneric |
| 769 | found = dump.find("SyncMemGeneric"); |
| 770 | BOOST_TEST(found == std::string::npos); |
| 771 | // Contains CopyMemGeneric |
| 772 | found = dump.find("CopyMemGeneric"); |
| 773 | BOOST_TEST(found != std::string::npos); |
Narumol Prangnawarat | 3b90af6 | 2020-06-26 11:00:21 +0100 | [diff] [blame] | 774 | |
| 775 | // Check that the outputs are correct |
| 776 | BOOST_CHECK_EQUAL_COLLECTIONS(outputData0.begin(), outputData0.end(), |
| 777 | expectedOutput.begin(), expectedOutput.end()); |
| 778 | BOOST_CHECK_EQUAL_COLLECTIONS(outputData1.begin(), outputData1.end(), |
| 779 | expectedOutput.begin(), expectedOutput.end()); |
Ferran Balaguer | 83239f9 | 2019-09-19 11:49:25 +0100 | [diff] [blame] | 780 | } |
| 781 | |
David Monahan | 0a99a14 | 2020-03-13 07:52:54 +0000 | [diff] [blame] | 782 | inline void StridedSliceInvalidSliceEndToEndTest(std::vector<BackendId> backends) |
| 783 | { |
| 784 | using namespace armnn; |
| 785 | |
| 786 | // Create runtime in which test will run |
| 787 | IRuntime::CreationOptions options; |
| 788 | IRuntimePtr runtime(armnn::IRuntime::Create(options)); |
| 789 | |
| 790 | // build up the structure of the network |
| 791 | INetworkPtr net(INetwork::Create()); |
| 792 | |
| 793 | IConnectableLayer* input = net->AddInputLayer(0); |
| 794 | |
| 795 | // Configure a strided slice with a stride the same size as the input but with a ShrinkAxisMask on the first |
| 796 | // dim of the output to make it too small to hold the specified slice. |
| 797 | StridedSliceDescriptor descriptor; |
| 798 | descriptor.m_Begin = {0, 0}; |
| 799 | descriptor.m_End = {2, 3}; |
| 800 | descriptor.m_Stride = {1, 1}; |
| 801 | descriptor.m_BeginMask = 0; |
| 802 | descriptor.m_EndMask = 0; |
| 803 | descriptor.m_ShrinkAxisMask = 1; |
| 804 | IConnectableLayer* stridedSlice = net->AddStridedSliceLayer(descriptor); |
| 805 | |
| 806 | IConnectableLayer* output0 = net->AddOutputLayer(0); |
| 807 | |
| 808 | input->GetOutputSlot(0).Connect(stridedSlice->GetInputSlot(0)); |
| 809 | stridedSlice->GetOutputSlot(0).Connect(output0->GetInputSlot(0)); |
| 810 | |
| 811 | input->GetOutputSlot(0).SetTensorInfo(TensorInfo({ 2, 3 }, DataType::Float32)); |
| 812 | stridedSlice->GetOutputSlot(0).SetTensorInfo(TensorInfo({ 3 }, DataType::Float32)); |
| 813 | |
| 814 | // Attempt to optimize the network and check that the correct exception is thrown |
| 815 | BOOST_CHECK_THROW(Optimize(*net, backends, runtime->GetDeviceSpec()), armnn::LayerValidationException); |
| 816 | } |
| 817 | |
Nattapat Chaimanowong | 1fcb4ff | 2019-01-24 15:25:26 +0000 | [diff] [blame] | 818 | } // anonymous namespace |