telsoa01 | c577f2c | 2018-08-31 09:22:23 +0100 | [diff] [blame] | 1 | // |
| 2 | // Copyright © 2017 Arm Ltd. All rights reserved. |
David Beck | ecb56cd | 2018-09-05 12:52:57 +0100 | [diff] [blame] | 3 | // SPDX-License-Identifier: MIT |
telsoa01 | c577f2c | 2018-08-31 09:22:23 +0100 | [diff] [blame] | 4 | // |
| 5 | #pragma once |
| 6 | |
| 7 | #include "armnnOnnxParser/IOnnxParser.hpp" |
| 8 | #include "google/protobuf/repeated_field.h" |
| 9 | #include <unordered_map> |
| 10 | |
| 11 | #include <onnx/onnx.pb.h> |
| 12 | |
| 13 | |
| 14 | namespace armnn |
| 15 | { |
| 16 | class TensorInfo; |
Tee Jung | 7ff9a60 | 2019-11-01 07:04:42 +0000 | [diff] [blame] | 17 | enum class ActivationFunction; |
telsoa01 | c577f2c | 2018-08-31 09:22:23 +0100 | [diff] [blame] | 18 | } |
| 19 | |
| 20 | namespace armnnOnnxParser |
| 21 | { |
| 22 | |
telsoa01 | c577f2c | 2018-08-31 09:22:23 +0100 | [diff] [blame] | 23 | using ModelPtr = std::unique_ptr<onnx::ModelProto>; |
| 24 | |
| 25 | class OnnxParser : public IOnnxParser |
| 26 | { |
| 27 | |
| 28 | using OperationParsingFunction = void(OnnxParser::*)(const onnx::NodeProto& NodeProto); |
| 29 | |
| 30 | public: |
| 31 | |
| 32 | using GraphPtr = std::unique_ptr<onnx::GraphProto>; |
| 33 | |
| 34 | /// Create the network from a protobuf binary file on disk |
| 35 | virtual armnn::INetworkPtr CreateNetworkFromBinaryFile(const char* graphFile) override; |
| 36 | |
| 37 | /// Create the network from a protobuf text file on disk |
| 38 | virtual armnn::INetworkPtr CreateNetworkFromTextFile(const char* graphFile) override; |
| 39 | |
| 40 | /// Create the network directly from protobuf text in a string. Useful for debugging/testing |
| 41 | virtual armnn::INetworkPtr CreateNetworkFromString(const std::string& protoText) override; |
| 42 | |
| 43 | /// Retrieve binding info (layer id and tensor info) for the network input identified by the given layer name |
| 44 | virtual BindingPointInfo GetNetworkInputBindingInfo(const std::string& name) const override; |
| 45 | |
| 46 | /// Retrieve binding info (layer id and tensor info) for the network output identified by the given layer name |
| 47 | virtual BindingPointInfo GetNetworkOutputBindingInfo(const std::string& name) const override; |
| 48 | |
| 49 | public: |
| 50 | |
| 51 | OnnxParser(); |
| 52 | |
| 53 | static ModelPtr LoadModelFromBinaryFile(const char * fileName); |
| 54 | static ModelPtr LoadModelFromTextFile(const char * fileName); |
| 55 | static ModelPtr LoadModelFromString(const std::string& inputString); |
| 56 | |
| 57 | ///Retrieve inputs names |
| 58 | static std::vector<std::string> GetInputs(ModelPtr& model); |
| 59 | |
| 60 | ///Retrieve outputs names |
| 61 | static std::vector<std::string> GetOutputs(ModelPtr& model); |
| 62 | |
| 63 | private: |
| 64 | |
| 65 | /// Parses a ModelProto loaded into memory from one of the other CreateNetwork* |
| 66 | armnn::INetworkPtr CreateNetworkFromModel(onnx::ModelProto& model); |
| 67 | |
| 68 | ///Parse every node and make the connection between the resulting tensors |
| 69 | void LoadGraph(); |
| 70 | |
| 71 | void SetupInfo(const google::protobuf::RepeatedPtrField<onnx::ValueInfoProto >* list); |
| 72 | |
| 73 | std::vector<armnn::TensorInfo> ComputeOutputInfo(std::vector<std::string> outNames, |
| 74 | const armnn::IConnectableLayer* layer, |
| 75 | std::vector<armnn::TensorShape> inputShapes); |
| 76 | |
| 77 | void DetectFullyConnected(); |
| 78 | |
| 79 | template <typename Location> |
| 80 | void GetInputAndParam(const onnx::NodeProto& node, |
| 81 | std::string* inputName, |
| 82 | std::string* constName, |
| 83 | const Location& location); |
| 84 | |
| 85 | template <typename Location> |
| 86 | void To1DTensor(const std::string &name, const Location& location); |
| 87 | |
| 88 | //Broadcast Preparation functions |
| 89 | std::pair<std::string, std::string> AddPrepareBroadcast(const std::string& input0, const std::string& input1); |
| 90 | void PrependForBroadcast(const std::string& outputName, const std::string& input0, const std::string& input1); |
| 91 | |
| 92 | void CreateConstantLayer(const std::string& tensorName, const std::string& layerName); |
| 93 | void CreateReshapeLayer(const std::string& inputName, |
| 94 | const std::string& outputName, |
| 95 | const std::string& layerName); |
| 96 | |
| 97 | void ParseBatchNormalization(const onnx::NodeProto& node); |
| 98 | void ParseConstant(const onnx::NodeProto& nodeProto); |
| 99 | |
| 100 | void ParseMaxPool(const onnx::NodeProto& nodeProto); |
| 101 | void ParseAveragePool(const onnx::NodeProto& nodeProto); |
| 102 | void ParseGlobalAveragePool(const onnx::NodeProto& node); |
| 103 | |
| 104 | void AddPoolingLayer(const onnx::NodeProto& nodeProto, armnn::Pooling2dDescriptor& desc); |
| 105 | |
| 106 | void ParseReshape(const onnx::NodeProto& nodeProto); |
Tee Jung | 7ff9a60 | 2019-11-01 07:04:42 +0000 | [diff] [blame] | 107 | |
| 108 | void ParseActivation(const onnx::NodeProto& nodeProto, const armnn::ActivationFunction func); |
| 109 | void ParseSigmoid(const onnx::NodeProto& nodeProto); |
| 110 | void ParseTanh(const onnx::NodeProto& nodeProto); |
telsoa01 | c577f2c | 2018-08-31 09:22:23 +0100 | [diff] [blame] | 111 | void ParseRelu(const onnx::NodeProto& nodeProto); |
Tee Jung | 7ff9a60 | 2019-11-01 07:04:42 +0000 | [diff] [blame] | 112 | void ParseLeakyRelu(const onnx::NodeProto& nodeProto); |
telsoa01 | c577f2c | 2018-08-31 09:22:23 +0100 | [diff] [blame] | 113 | |
| 114 | void AddConvLayerWithDepthwiseConv(const onnx::NodeProto& node, const armnn::Convolution2dDescriptor& convDesc); |
| 115 | void ParseConv(const onnx::NodeProto& nodeProto); |
| 116 | |
| 117 | void ParseAdd(const onnx::NodeProto& nodeProto); |
| 118 | void AddFullyConnected(const onnx::NodeProto& matmulNode, const onnx::NodeProto* addNode = nullptr); |
| 119 | |
| 120 | void RegisterInputSlots(armnn::IConnectableLayer* layer, const std::vector<std::string>& tensorIndexes); |
| 121 | void RegisterOutputSlots(armnn::IConnectableLayer* layer, const std::vector<std::string>& tensorIndexes); |
| 122 | |
| 123 | void SetupInputLayers(); |
| 124 | void SetupOutputLayers(); |
| 125 | |
| 126 | void ResetParser(); |
| 127 | void Cleanup(); |
| 128 | |
| 129 | std::pair<armnn::ConstTensor, std::unique_ptr<float[]>> CreateConstTensor(const std::string name); |
| 130 | |
| 131 | template <typename TypeList, typename Location> |
| 132 | void ValidateInputs(const onnx::NodeProto& node, |
| 133 | TypeList validInputs, |
| 134 | const Location& location); |
| 135 | |
| 136 | /// The network we're building. Gets cleared after it is passed to the user |
| 137 | armnn::INetworkPtr m_Network; |
| 138 | |
| 139 | ///Ptr to the graph we're building the network from |
| 140 | GraphPtr m_Graph; |
| 141 | |
| 142 | ///Map of the information for every tensor |
| 143 | struct OnnxTensor |
| 144 | { |
| 145 | std::unique_ptr<armnn::TensorInfo> m_info; |
| 146 | std::unique_ptr<const onnx::TensorProto> m_tensor; |
| 147 | onnx::TensorProto::DataType m_dtype; |
| 148 | |
| 149 | OnnxTensor() : m_info(nullptr), m_tensor(nullptr), m_dtype(onnx::TensorProto::FLOAT) { } |
| 150 | bool isConstant() { return m_tensor != nullptr; } |
| 151 | |
| 152 | }; |
| 153 | |
| 154 | std::unordered_map<std::string, OnnxTensor> m_TensorsInfo; |
| 155 | |
| 156 | /// map of onnx operation names to parsing member functions |
| 157 | static const std::map<std::string, OperationParsingFunction> m_ParserFunctions; |
| 158 | |
| 159 | /// A mapping of an output slot to each of the input slots it should be connected to |
| 160 | /// The outputSlot is from the layer that creates this tensor as one of its ouputs |
| 161 | /// The inputSlots are from the layers that use this tensor as one of their inputs |
| 162 | struct TensorSlots |
| 163 | { |
| 164 | armnn::IOutputSlot* outputSlot; |
| 165 | std::vector<armnn::IInputSlot*> inputSlots; |
| 166 | |
| 167 | TensorSlots() : outputSlot(nullptr) { } |
| 168 | }; |
| 169 | ///Map of the tensor names to their connections for the connections of the layers of the graph |
| 170 | std::unordered_map<std::string, TensorSlots> m_TensorConnections; |
| 171 | |
| 172 | //Map of the tensor names to their node and index in graph.node() |
| 173 | std::unordered_map<std::string, std::pair<const onnx::NodeProto*, int>> m_OutputsMap; |
| 174 | |
| 175 | /// Number of times a specific node (identified by his index number) was used as input |
| 176 | /// and list of the nodes it was fused with |
| 177 | struct UsageSummary |
| 178 | { |
| 179 | std::vector<size_t> fusedWithNodes; |
| 180 | size_t inputForNodes; |
| 181 | |
| 182 | UsageSummary() : fusedWithNodes({}), inputForNodes(0) { } |
| 183 | |
| 184 | }; |
| 185 | |
| 186 | std::vector<UsageSummary> m_OutputsFusedAndUsed; |
| 187 | }; |
| 188 | } |