ie_plugin.hpp

Go to the documentation of this file.

 // Copyright (C) 2018-2019 Intel Corporation


 // SPDX-License-Identifier: Apache-2.0


 //


 


 /**


  * @brief A header file for Main Inference Engine API


  * @file ie_plugin.hpp


  */


 #pragma once


 

 #include <ie_icnn_network.hpp>


 #include <ie_iextension.h>


 #include "ie_api.h"


 #include "details/ie_no_copy.hpp"


 #include "ie_error.hpp"


 #include "ie_version.hpp"


 #include "ie_iexecutable_network.hpp"


 #include <string>


 #include <vector>


 #include <memory>


 #include <map>


 #include <set>


 

 

 #if defined(_WIN32)


     #ifdef IMPLEMENT_INFERENCE_ENGINE_PLUGIN


         #define INFERENCE_PLUGIN_API(type) extern "C"   __declspec(dllexport) type


     #else


         #define INFERENCE_PLUGIN_API(type) extern "C" type


     #endif


 #elif(__GNUC__ >= 4)


     #ifdef IMPLEMENT_INFERENCE_ENGINE_PLUGIN


         #define INFERENCE_PLUGIN_API(type) extern "C"   __attribute__((visibility("default"))) type


     #else


         #define INFERENCE_PLUGIN_API(type) extern "C" type


     #endif


 #else


     #define INFERENCE_PLUGIN_API(TYPE) extern "C" TYPE


 #endif


 

 namespace InferenceEngine {

 


 /**


  * @brief Responce structure encapsulating information about supported layer


  */


 struct QueryNetworkResult {

     std::set<std::string> supportedLayers;

     StatusCode rc;

     ResponseDesc resp;

 };

 


 /**


  * @brief This class is a main plugin interface


  */


 class IInferencePlugin : public details::IRelease {

 public:


     /**


      * @brief Returns plugin version information


      * @param versionInfo Pointer to version info. Is set by plugin


      */


     virtual void GetVersion(const Version *&versionInfo) noexcept = 0;

 


     /**


      * @brief Sets logging callback


      * Logging is used to track what is going on inside


      * @param listener Logging sink


      */


     virtual void SetLogCallback(IErrorListener &listener) noexcept = 0;

 


     /**


      * @deprecated use LoadNetwork with four parameters (executable network, cnn network, config, response)


      * @brief Loads a pre-built network with weights to the engine. In case of success the plugin will


      *        be ready to infer


      * @param network Network object acquired from CNNNetReader


      * @param resp Pointer to the response message that holds a description of an error if any occurred


      * @return Status code of the operation. OK if succeeded


      */


     virtual StatusCode LoadNetwork(ICNNNetwork &network, ResponseDesc *resp) noexcept = 0;

 


     /**


      * @brief Creates an executable network from a network object. User can create as many networks as they need and use


      *        them simultaneously (up to the limitation of the hardware resources)


      * @param ret Reference to a shared ptr of the returned network interface


      * @param network Network object acquired from CNNNetReader


      * @param config Map of pairs: (config parameter name, config parameter value) relevant only for this load operation


      * @param resp Pointer to the response message that holds a description of an error if any occurred


      * @return Status code of the operation. OK if succeeded


      */


     virtual StatusCode


     LoadNetwork(IExecutableNetwork::Ptr &ret, ICNNNetwork &network, const std::map<std::string, std::string> &config,

                 ResponseDesc *resp) noexcept = 0;

 


     /**


      * @brief Creates an executable network from a previously exported network


      * @param ret Reference to a shared ptr of the returned network interface


      * @param modelFileName Path to the location of the exported file


      * @param config Map of pairs: (config parameter name, config parameter value) relevant only for this load operation*


      * @param resp Pointer to the response message that holds a description of an error if any occurred


      * @return Status code of the operation. OK if succeeded


      */


     virtual StatusCode


     ImportNetwork(IExecutableNetwork::Ptr &ret, const std::string &modelFileName,

                   const std::map<std::string, std::string> &config, ResponseDesc *resp) noexcept = 0;

 


     /**


      * @deprecated Uses Infer() working with multiple inputs and outputs


      * @brief Infers an image(s).


      * Input and output dimensions depend on the topology.


      *     As an example for classification topologies use a 4D Blob as input (batch, channels, width,


      *             height) and get a 1D blob as output (scoring probability vector). To Infer a batch,


      *             use a 4D Blob as input and get a 2D blob as output in both cases the method will


      *             allocate the resulted blob


      * @param input Any TBlob<> object that contains the data to infer. The type of TBlob must match the network input precision and size.


      * @param result Related TBlob<> object that contains the result of the inference action, typically this is a float blob.


                The blob does not need to be allocated or initialized, the engine allocates the relevant data.


      * @param resp Pointer to the response message that holds a description of an error if any occurred


      * @return Status code of the operation. OK if succeeded


      */


     virtual StatusCode Infer(const Blob &input, Blob &result, ResponseDesc *resp) noexcept = 0;

 


     /**


      * @deprecated Loads IExecutableNetwork to create IInferRequest.


      * @brief Infers tensors. Input and output dimensions depend on the topology.


      *     As an example for classification topologies use a 4D Blob as input (batch, channels, width,


      *             height) and get a 1D blob as output (scoring probability vector). To Infer a batch,


      *             use a 4D Blob as input and get a 2D blob as output in both cases the method will


      *             allocate the resulted blob


      * @param input Map of input blobs accessed by input names


      * @param result Map of output blobs accessed by output names


      * @param resp Pointer to the response message that holds a description of an error if any occurred


      * @return Status code of the operation. OK if succeeded


      */


     virtual StatusCode Infer(const BlobMap &input, BlobMap &result, ResponseDesc *resp) noexcept = 0;

 


     /**


      * @deprecated Uses IInferRequest to get performance measures


      * @brief Queries performance measures per layer to get feedback of what is the most time consuming layer


      *  Note: not all plugins provide meaningful data


      * @param perfMap Map of layer names to profiling information for that layer


      * @param resp Pointer to the response message that holds a description of an error if any occurred


      * @return Status code of the operation. OK if succeeded


      */


     virtual StatusCode GetPerformanceCounts(std::map<std::string, InferenceEngineProfileInfo> &perfMap,

                                             ResponseDesc *resp) const noexcept = 0;

 


     /**


      * @brief Registers extension within the plugin


      * @param extension Pointer to already loaded extension


      * @param resp Pointer to the response message that holds a description of an error if any occurred


      * @return Status code of the operation. OK if succeeded


      */


     virtual StatusCode AddExtension(InferenceEngine::IExtensionPtr extension,

                                     InferenceEngine::ResponseDesc *resp) noexcept = 0;

 


     /**


      * @brief Sets configuration for plugin, acceptable keys can be found in ie_plugin_config.hpp


      * @param config Map of pairs: (config parameter name, config parameter value)


      * @param resp Pointer to the response message that holds a description of an error if any occurred


      */


     virtual StatusCode SetConfig(const std::map<std::string, std::string> &config, ResponseDesc *resp) noexcept = 0;

 

 


     /**


      * @depricated Use the version with config parameter


      * @brief Query plugin if it supports specified network


      * @param network Network object to query


      * @param resp Pointer to the response message that holds a description of an error if any occurred


      */


     virtual void QueryNetwork(const ICNNNetwork& /*network*/, QueryNetworkResult& res) const noexcept {

         res.rc = InferenceEngine::NOT_IMPLEMENTED;

     }

 


     /**


      * @brief Query plugin if it supports specified network with specified configuration


      * @param network Network object to query


      * @param config Map of pairs: (config parameter name, config parameter value)


      * @param resp Pointer to the response message that holds a description of an error if any occurred


      */


     virtual void QueryNetwork(const ICNNNetwork& /*network*/,

                               const std::map<std::string, std::string> &/*config*/, QueryNetworkResult& res) const noexcept {

         res.rc = InferenceEngine::NOT_IMPLEMENTED;

     }

 };

 


 /**


  * @brief Creates the default instance of the interface (per plugin)


  * @param plugin Pointer to the plugin


  * @param resp Pointer to the response message that holds a description of an error if any occurred


  * @return Status code of the operation. OK if succeeded


  */


 INFERENCE_PLUGIN_API(StatusCode) CreatePluginEngine(IInferencePlugin *&plugin, ResponseDesc *resp) noexcept;

 }  // namespace InferenceEngine