1 /* 2 * Copyright (c) 2023 Huawei Device Co., Ltd. 3 * Licensed under the Apache License, Version 2.0 (the "License"); 4 * you may not use this file except in compliance with the License. 5 * You may obtain a copy of the License at 6 * 7 * http://www.apache.org/licenses/LICENSE-2.0 8 * 9 * Unless required by applicable law or agreed to in writing, software 10 * distributed under the License is distributed on an "AS IS" BASIS, 11 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 12 * See the License for the specific language governing permissions and 13 * limitations under the License. 14 */ 15 16 #ifndef NEURAL_NETWORK_RUNTIME_NNCOMPILER_H 17 #define NEURAL_NETWORK_RUNTIME_NNCOMPILER_H 18 19 #include "compiler.h" 20 21 #include "mindir.h" 22 #include "device.h" 23 #include "inner_model.h" 24 #include "prepared_model.h" 25 #include "nnexecutor.h" 26 27 namespace OHOS { 28 namespace NeuralNetworkRuntime { 29 30 class NNCompiler : public Compiler { 31 public: 32 NNCompiler() = delete; 33 NNCompiler(std::shared_ptr<Device> device, size_t backendID); 34 NNCompiler(const void* model, std::shared_ptr<Device> device, size_t backendID); 35 ~NNCompiler() override; 36 37 size_t GetBackendID() const override; 38 39 OH_NN_ReturnCode SetCacheDir(const std::string& cacheModelPath, uint32_t version) override; 40 OH_NN_ReturnCode SetPerformance(OH_NN_PerformanceMode performance) override; 41 OH_NN_ReturnCode SetPriority(OH_NN_Priority priority) override; 42 OH_NN_ReturnCode SetEnableFp16(bool isFp16) override; 43 44 bool IsBuild() const override; 45 OH_NN_ReturnCode Build() override; 46 47 OH_NN_ReturnCode SaveToCacheFile() const override; 48 OH_NN_ReturnCode RestoreFromCacheFile() override; 49 OH_NN_ReturnCode SaveToCacheBuffer(const void* buffer, size_t length, size_t* modelSize) const override; 50 OH_NN_ReturnCode RestoreFromCacheBuffer(const void* buffer, size_t length) override; 51 52 OH_NN_ReturnCode SetExtensionConfig(const std::unordered_map<std::string, std::vector<char>>& configs) override; 53 OH_NN_ReturnCode SetOptions(const std::vector<std::shared_ptr<void>>& options) override; 54 OH_NN_ReturnCode GetModelName(std::string& modelName) override; 55 size_t GetModelSize() override; 56 size_t GetOnlineModelID() override; 57 58 NNExecutor* CreateExecutor(); 59 60 private: 61 void ReleaseBuffer(std::vector<Buffer>& buffers) const; 62 void ReleaseBufferByDevice(std::vector<Buffer>& buffers) const; 63 OH_NN_ReturnCode SerializeTensorsToBuffer( 64 const std::vector<std::pair<std::shared_ptr<TensorDesc>, OH_NN_TensorType>>& tensorDescs, 65 Buffer& buffer) const; 66 OH_NN_ReturnCode DeserializedTensorsFromBuffer( 67 const Buffer& buffer, std::vector<std::pair<std::shared_ptr<TensorDesc>, OH_NN_TensorType>>& tensorDescs); 68 69 OH_NN_ReturnCode OnlineBuild(); 70 OH_NN_ReturnCode NormalBuild(); 71 OH_NN_ReturnCode BuildOfflineModel(); 72 OH_NN_ReturnCode CheckModelParameter() const; 73 OH_NN_ReturnCode IsOfflineModel(bool& isOfflineModel) const; 74 OH_NN_ReturnCode IsSupportedModel(const std::shared_ptr<mindspore::lite::LiteGraph>& liteGraph, 75 bool& isSupportedModel) const; 76 77 size_t GetModelSizeFromCache(std::string& path, const std::string& modelName); 78 size_t GetModelSizeFromFile(std::string& path); 79 size_t GetModelSizeFromModel(InnerModel* innerModel); 80 OH_NN_ReturnCode GetNNRtModelIDFromCache(const std::string& path, const std::string& modelName, 81 size_t& nnrtModelID); 82 OH_NN_ReturnCode GetNNRtModelIDFromModel(InnerModel* innerModel, size_t& nnrtModelID); 83 size_t GetOnlineModelID(const std::shared_ptr<mindspore::lite::LiteGraph>& liteGraph); 84 std::vector<mindspore::lite::LiteGraph::Node*> GetNodeIndices( 85 const std::shared_ptr<mindspore::lite::LiteGraph>& liteGraph, size_t layer); 86 size_t DataTypeSize(mindspore::lite::DataType dataType); 87 size_t GetFileSize(const char* fileName); 88 89 private: 90 bool m_isBuild {false}; 91 bool m_enableFp16 {false}; 92 std::string m_cachePath; 93 uint32_t m_cacheVersion {0}; 94 std::shared_ptr<Device> m_device {nullptr}; 95 size_t m_backendID {0}; 96 OH_NN_Priority m_priority {OH_NN_PRIORITY_NONE}; 97 OH_NN_PerformanceMode m_performance {OH_NN_PERFORMANCE_NONE}; 98 std::shared_ptr<PreparedModel> m_preparedModel {nullptr}; 99 void* m_metaGraph {nullptr}; 100 InnerModel* m_innerModel {nullptr}; 101 std::shared_ptr<mindspore::lite::LiteGraph> m_liteGraph {nullptr}; 102 std::vector<std::pair<std::shared_ptr<TensorDesc>, OH_NN_TensorType>> m_inputTensorDescs; 103 std::vector<std::pair<std::shared_ptr<TensorDesc>, OH_NN_TensorType>> m_outputTensorDescs; 104 ExtensionConfig m_extensionConfig; 105 }; 106 } // NeuralNetworkRuntime 107 } // OHOS 108 109 #endif // NEURAL_NETWORK_RUNTIME_NNCOMPILER_H