Refactoring code

src/plugins/intel_npu/src/al/CMakeLists.txt

@@ -26,6 +26,12 @@ target_link_libraries(${TARGET_NAME}
         openvino::runtime::dev
 )
 
+target_include_directories(${TARGET_NAME}
+    PUBLIC
+        $<TARGET_PROPERTY:LevelZero::LevelZero,INTERFACE_INCLUDE_DIRECTORIES>
+        $<TARGET_PROPERTY:LevelZero::NPUExt,INTERFACE_INCLUDE_DIRECTORIES>
+)
+
 set_target_properties(${TARGET_NAME} PROPERTIES INTERPROCEDURAL_OPTIMIZATION_RELEASE ${ENABLE_LTO})
 ov_add_clang_format_target(${TARGET_NAME}_clang FOR_TARGETS ${TARGET_NAME})
 

src/plugins/intel_npu/src/al/include/igraph.hpp

@@ -2,43 +2,80 @@
 // SPDX-License-Identifier: Apache-2.0
 //
 
-// Compiler Interface
-
 #pragma once
 
+#include <ze_graph_ext.h>
+
 #include <memory>
 #include <vector>
 
-#include "intel_npu/al/config/common.hpp"
 #include "intel_npu/al/icompiler.hpp"
 
 namespace intel_npu {
 
-class IGraph {
+class IExecutor {
 public:
-    IGraph(void* handle, NetworkMetadata metadata) : _handle(handle), _metadata(std::move(metadata)) {}
+    virtual ~IExecutor() = default;
+
+    virtual void setWorkloadType(const ov::WorkloadType workloadType) const = 0;
+};
+
+class IGraph : public std::enable_shared_from_this<IGraph> {
+public:
+    IGraph(ze_graph_handle_t handle, NetworkMetadata metadata) : _handle(handle), _metadata(std::move(metadata)) {}
 
     virtual CompiledNetwork export_blob() const = 0;
 
-    virtual std::vector<ov::ProfilingInfo> process_profiling_output() const = 0;
+    virtual std::vector<ov::ProfilingInfo> process_profiling_output(const std::vector<uint8_t>& profData) const = 0;
 
     virtual void set_argument_value(uint32_t argi, const void* argv) const = 0;
 
-    virtual void initialize() const = 0;
+    virtual void initialize() = 0;
 
     virtual ~IGraph() = default;
 
-    NetworkMetadata get_metadata() const {
+    struct ArgumentDescriptor {
+        ze_graph_argument_properties_3_t info;
+        uint32_t idx;
+    };
+
+    void setWorkloadType(const ov::WorkloadType workloadType) const {
+        if (_executor != nullptr) {
+            _executor->setWorkloadType(workloadType);
+        }
+    }
+
+    const std::shared_ptr<IExecutor>& get_executor() const {
+        return _executor;
+    }
+
+    const NetworkMetadata& get_metadata() const {
         return _metadata;
     }
 
-    void* get_handle() const {
+    ze_graph_handle_t get_handle() const {
         return _handle;
     }
 
+    void update_network_name(std::string_view name) {
+        _metadata.name = name;
+    }
+
+    inline const std::vector<ArgumentDescriptor>& get_input_descriptors() const {
+        return _input_descriptors;
+    }
+    inline const std::vector<ArgumentDescriptor>& get_output_descriptors() const {
+        return _output_descriptors;
+    }
+
 protected:
-    void* _handle = nullptr;
+    ze_graph_handle_t _handle = nullptr;
     NetworkMetadata _metadata;
+
+    std::shared_ptr<IExecutor> _executor;
+
+    std::vector<ArgumentDescriptor> _input_descriptors;
+    std::vector<ArgumentDescriptor> _output_descriptors;
 };
 
 }  // namespace intel_npu

src/plugins/intel_npu/src/al/include/intel_npu/al/icompiled_model.hpp

@@ -7,7 +7,6 @@
 #include <memory>
 #include <vector>
 
-#include "igraph.hpp"
 #include "intel_npu/al/config/common.hpp"
 #include "intel_npu/al/icompiler.hpp"
 #include "openvino/runtime/icompiled_model.hpp"
@@ -18,13 +17,9 @@ class ICompiledModel : public ov::ICompiledModel {
 public:
     using ov::ICompiledModel::ICompiledModel;
 
-    virtual const std::shared_ptr<IGraph>& get_graph() const = 0;
-
     virtual const Config& get_config() const = 0;
 
-    const NetworkMetadata get_network_metadata() const {
-        return get_graph()->get_metadata();
-    }
+    virtual const NetworkMetadata& get_network_metadata() const = 0;
 
 protected:
     std::shared_ptr<const ICompiledModel> shared_from_this() const {

src/plugins/intel_npu/src/al/include/npu.hpp

@@ -6,7 +6,7 @@
 
 #include <cstdint>
 
-#include "IGraph.hpp"
+#include "igraph.hpp"
 #include "intel_npu/al/config/config.hpp"
 #include "intel_npu/al/icompiled_model.hpp"
 #include "intel_npu/al/icompiler.hpp"
@@ -53,11 +53,14 @@ class IEngineBackend : public std::enable_shared_from_this<IEngineBackend> {
 
 //------------------------------------------------------------------------------
 
-class IExecutor {
+class ICompilerAdapter {
 public:
-    virtual ~IExecutor() = default;
+    virtual std::shared_ptr<IGraph> compile(const std::shared_ptr<const ov::Model>& model,
+                                            const Config& config) const = 0;
+    virtual std::shared_ptr<IGraph> parse(const std::vector<uint8_t>& network, const Config& config) const = 0;
+    virtual ov::SupportedOpsMap query(const std::shared_ptr<const ov::Model>& model, const Config& config) const = 0;
 
-    virtual void setWorkloadType(const ov::WorkloadType workloadType) const = 0;
+    virtual ~ICompilerAdapter() = default;
 };
 
 //------------------------------------------------------------------------------
@@ -66,8 +69,6 @@ class IDevice : public std::enable_shared_from_this<IDevice> {
 public:
     using Uuid = ov::device::UUID;
 
-    virtual std::shared_ptr<IExecutor> createExecutor(const std::shared_ptr<IGraph>& graph, const Config& config) = 0;
-
     virtual std::string getName() const = 0;
     virtual std::string getFullDeviceName() const = 0;
     virtual Uuid getUuid() const;
@@ -82,7 +83,6 @@ class IDevice : public std::enable_shared_from_this<IDevice> {
     virtual std::shared_ptr<SyncInferRequest> createInferRequest(
         const std::shared_ptr<const ICompiledModel>& compiledModel,
         const std::shared_ptr<IGraph>& graph,
-        const std::shared_ptr<IExecutor>& executor,
         const Config& config) = 0;
 
     virtual void updateInfo(const Config& config) = 0;

src/plugins/intel_npu/src/backend/CMakeLists.txt

@@ -34,25 +34,21 @@ target_link_libraries(${TARGET_NAME}
 ov_install_static_lib(${TARGET_NAME} ${NPU_INTERNAL_COMPONENT})
 
 if(TARGET ze_loader)
-    if(NOT BUILD_SHARED_LIBS)
-        # Support link of static runtime in case system does not have ze_loader
-        install(TARGETS ze_loader EXPORT OpenVINOTargets
-            RUNTIME DESTINATION ${OV_CPACK_RUNTIMEDIR} COMPONENT ${NPU_PLUGIN_COMPONENT}
-            ARCHIVE DESTINATION ${OV_CPACK_ARCHIVEDIR} COMPONENT ${NPU_PLUGIN_COMPONENT}
-            LIBRARY DESTINATION ${OV_CPACK_LIBRARYDIR} COMPONENT ${NPU_PLUGIN_COMPONENT})
-
-        install(TARGETS utils EXPORT OpenVINOTargets
-            RUNTIME DESTINATION ${OV_CPACK_RUNTIMEDIR} COMPONENT ${NPU_PLUGIN_COMPONENT}
-            ARCHIVE DESTINATION ${OV_CPACK_ARCHIVEDIR} COMPONENT ${NPU_PLUGIN_COMPONENT}
-            LIBRARY DESTINATION ${OV_CPACK_LIBRARYDIR} COMPONENT ${NPU_PLUGIN_COMPONENT})
-
-        # export to local tree to build against static build tree
-        export(TARGETS ze_loader NAMESPACE openvino::
-            APPEND FILE "${CMAKE_BINARY_DIR}/OpenVINOTargets.cmake")
-
-        export(TARGETS utils NAMESPACE openvino::
-            APPEND FILE "${CMAKE_BINARY_DIR}/OpenVINOTargets.cmake")
-    endif()
+    install(TARGETS ze_loader EXPORT OpenVINOTargets
+        RUNTIME DESTINATION ${OV_CPACK_RUNTIMEDIR} COMPONENT ${NPU_PLUGIN_COMPONENT}
+        ARCHIVE DESTINATION ${OV_CPACK_ARCHIVEDIR} COMPONENT ${NPU_PLUGIN_COMPONENT}
+        LIBRARY DESTINATION ${OV_CPACK_LIBRARYDIR} COMPONENT ${NPU_PLUGIN_COMPONENT})
+
+    install(TARGETS utils EXPORT OpenVINOTargets
+        RUNTIME DESTINATION ${OV_CPACK_RUNTIMEDIR} COMPONENT ${NPU_PLUGIN_COMPONENT}
+        ARCHIVE DESTINATION ${OV_CPACK_ARCHIVEDIR} COMPONENT ${NPU_PLUGIN_COMPONENT}
+        LIBRARY DESTINATION ${OV_CPACK_LIBRARYDIR} COMPONENT ${NPU_PLUGIN_COMPONENT})
+
+    export(TARGETS ze_loader NAMESPACE openvino::
+        APPEND FILE "${CMAKE_BINARY_DIR}/OpenVINOTargets.cmake")
+
+    export(TARGETS utils NAMESPACE openvino::
+        APPEND FILE "${CMAKE_BINARY_DIR}/OpenVINOTargets.cmake")
 
     # Support tests to run with ze_loader
     install(TARGETS ze_loader

src/plugins/intel_npu/src/backend/include/zero_device.hpp

@@ -20,8 +20,6 @@ class ZeroDevice : public IDevice {
 public:
     ZeroDevice(const std::shared_ptr<ZeroInitStructsHolder>& initStructs);
 
-    std::shared_ptr<IExecutor> createExecutor(const std::shared_ptr<IGraph>& graph, const Config& config) override;
-
     std::string getName() const override;
     std::string getFullDeviceName() const override;
     Uuid getUuid() const override;
@@ -36,7 +34,6 @@ class ZeroDevice : public IDevice {
 
     std::shared_ptr<SyncInferRequest> createInferRequest(const std::shared_ptr<const ICompiledModel>& compiledModel,
                                                          const std::shared_ptr<IGraph>& graph,
-                                                         const std::shared_ptr<IExecutor>& executor,
                                                          const Config& config) override;
     void updateInfo(const Config& config) override {
         log.setLevel(config.get<LOG_LEVEL>());

src/plugins/intel_npu/src/backend/include/zero_executor.hpp

@@ -9,7 +9,6 @@
 
 #include <mutex>
 
-#include "igraph.hpp"
 #include "intel_npu/utils/logger/logger.hpp"
 #include "npu.hpp"
 #include "openvino/runtime/properties.hpp"
@@ -20,15 +19,18 @@ namespace intel_npu {
 
 class ZeroExecutor final : public IExecutor {
 public:
-    ZeroExecutor(const std::shared_ptr<const ZeroInitStructsHolder>& initStructs,
-                 const std::shared_ptr<IGraph>& graph,
+    ZeroExecutor(ze_graph_handle_t graphHandle,
+                 ze_device_handle_t deviceHandle,
+                 ze_context_handle_t contextHandle,
+                 ze_graph_dditable_ext_curr_t& graphDdiTableExt,
+                 ze_command_queue_npu_dditable_ext_curr_t& commandQueueDdiTable,
                  const Config& config,
-                 uint32_t group_ordinal);
+                 uint32_t groupOrdinal);
 
     ZeroExecutor(const ZeroExecutor&) = delete;
     ZeroExecutor& operator=(const ZeroExecutor&) = delete;
 
-    ~ZeroExecutor() override;
+    ~ZeroExecutor() override{};
 
     struct ArgumentDescriptor {
         ze_graph_argument_properties_3_t info;
@@ -53,9 +55,6 @@ class ZeroExecutor final : public IExecutor {
     const Config _config;
     Logger _logger;
 
-    const std::shared_ptr<const ZeroInitStructsHolder> _initStructs;
-    std::shared_ptr<IGraph> _graph;
-
     std::vector<ArgumentDescriptor> _input_descriptors;
     std::vector<ArgumentDescriptor> _output_descriptors;
 

src/plugins/intel_npu/src/backend/include/zero_infer_request.hpp

@@ -23,7 +23,6 @@ class ZeroInferRequest final : public SyncInferRequest {
     explicit ZeroInferRequest(const std::shared_ptr<ZeroInitStructsHolder>& initStructs,
                               const std::shared_ptr<const ICompiledModel>& compiledModel,
                               const std::shared_ptr<IGraph>& graph,
-                              const std::shared_ptr<const IExecutor>& executor,
                               const Config& config,
                               uint32_t group_ordinal);
 
@@ -87,7 +86,6 @@ class ZeroInferRequest final : public SyncInferRequest {
 
     const std::shared_ptr<ZeroInitStructsHolder> _initStructs;
     const std::shared_ptr<IGraph> _graph;
-    const std::shared_ptr<const IExecutor> _executorPtr;
     const ZeroExecutor* _executor;
     const uint32_t _group_ordinal;
     const Config _config;

src/plugins/intel_npu/src/backend/include/zero_pipeline.hpp

@@ -23,7 +23,6 @@ struct Pipeline {
     Pipeline(const Config& config,
              const std::shared_ptr<ZeroInitStructsHolder>& initStructs,
              const std::shared_ptr<IGraph>& graph,
-             const std::shared_ptr<const IExecutor>& executorPtr,
              zeroProfiling::ProfilingPool& profiling_pool,
              zeroProfiling::ProfilingQuery& profiling_query,
              std::shared_ptr<zeroProfiling::NpuInferProfiling> npu_profiling,

src/plugins/intel_npu/src/backend/include/zero_wrappers.hpp

@@ -15,14 +15,6 @@ namespace intel_npu {
 class CommandList;
 class CommandQueue;
 
-enum stage {
-    UPLOAD,
-    EXECUTE,
-    READBACK,
-
-    COUNT
-};
-
 class EventPool {
 public:
     EventPool() = delete;

src/plugins/intel_npu/src/backend/src/zero_device.cpp

@@ -85,11 +85,6 @@ ZeroDevice::ZeroDevice(const std::shared_ptr<ZeroInitStructsHolder>& initStructs
     log.debug("ZeroDevice::ZeroDevice - init completed");
 }
 
-std::shared_ptr<IExecutor> ZeroDevice::createExecutor(const std::shared_ptr<IGraph>& graph, const Config& config) {
-    OV_ITT_SCOPED_TASK(itt::domains::LevelZeroBackend, "Device::createExecutor");
-    return std::make_shared<ZeroExecutor>(_initStructs, graph, config, _group_ordinal);
-}
-
 std::string ZeroDevice::getName() const {
 //    KMD is setting usDeviceID from VpuFamilyID.h
 #define NPU_3720_P_DEVICE_ID 0x7D1D
@@ -172,9 +167,8 @@ uint32_t ZeroDevice::getGroupOrdinal() const {
 std::shared_ptr<SyncInferRequest> ZeroDevice::createInferRequest(
     const std::shared_ptr<const ICompiledModel>& compiledModel,
     const std::shared_ptr<IGraph>& graph,
-    const std::shared_ptr<IExecutor>& executor,
     const Config& config) {
-    return std::make_shared<ZeroInferRequest>(_initStructs, compiledModel, graph, executor, config, _group_ordinal);
+    return std::make_shared<ZeroInferRequest>(_initStructs, compiledModel, graph, config, _group_ordinal);
 }
 
 ov::SoPtr<ov::IRemoteTensor> ZeroDevice::createRemoteTensor(std::shared_ptr<ov::IRemoteContext> context,