fix test, clean header and clang-format

Author: DanLiu2Intel

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

@@ -4,7 +4,7 @@
 
 #pragma once
 
-#include "intel_npu/common/graph_arguments.hpp"
+#include "intel_npu/common/dynamic_graph_arguments.hpp"
 #include "intel_npu/common/network_metadata.hpp"
 #include "zero_pipeline.hpp"
 

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

@@ -4,7 +4,6 @@
 
 #pragma once
 
-#include "intel_npu/common/graph_arguments.hpp"
 #include "intel_npu/common/igraph.hpp"
 #include "intel_npu/utils/zero/zero_tensor.hpp"
 #include "intel_npu/utils/zero/zero_wrappers.hpp"

src/plugins/intel_npu/src/backend/src/zero_dynamic_infer_request.cpp

@@ -283,8 +283,7 @@ void ZeroDynamicInferRequest::predict_shapes(std::vector<MemRefType>& outputProp
     }
 }
 
-void ZeroDynamicInferRequest::check_tensor_and_predicted_shapes(
-    const std::vector<MemRefType>& outputProps) {
+void ZeroDynamicInferRequest::check_tensor_and_predicted_shapes(const std::vector<MemRefType>& outputProps) {
     if (outputProps.empty()) {
         _logger.debug("check_tensor_and_predicted_shapes - no output props to check, skip check");
         return;

src/plugins/intel_npu/src/backend/src/zero_dynamic_pipeline.cpp

@@ -9,7 +9,7 @@
 
 #include <sstream>
 
-#include "intel_npu/common/dynamic_graph_vm_impl.hpp"
+#include "intel_npu/common/dynamic_graph_arguments_impl.hpp"
 #include "intel_npu/common/itt.hpp"
 #include "intel_npu/config/options.hpp"
 #include "intel_npu/prefix.hpp"
@@ -201,7 +201,7 @@ void DynamicPipeline::execute_vm_runtime(npu_vm_runtime_handle_t vmRuntime,
                                          ze_command_queue_handle_t commandQueue,
                                          ze_fence_handle_t fence,
                                          ze_event_handle_t event) {
-    _logger.debug("execute_vm_runtime - started");
+    _logger.debug("Start to execute graph with runtime engine");
 
     const bool firstExecution = (args._impl == nullptr);
     std::shared_ptr<DynamicGraphArgumentsImpl> argsImpl =
@@ -239,7 +239,7 @@ void DynamicPipeline::execute_vm_runtime(npu_vm_runtime_handle_t vmRuntime,
     }
 
     if (!firstExecution && noTensorChange) {
-        _logger.debug("execute_vm_runtime - reuse command list (no tensor change)");
+        _logger.debug("Reuse command list without update since no tensor change detected");
         auto result = zeCommandQueueExecuteCommandLists(commandQueue,
                                                         static_cast<uint32_t>(commandLists.size()),
                                                         commandLists.data(),
@@ -250,16 +250,16 @@ void DynamicPipeline::execute_vm_runtime(npu_vm_runtime_handle_t vmRuntime,
         return;
     }
 
-    _logger.debug("execute_vm_runtime - reset command lists");
-    // Reset commandLists since there are tensors with new shapes (or first execution); can not reuse via update.
+    _logger.debug("Reset command list to run with runtime");
+    // Reset commandLists since there are tensor with new shapes or it is the first execution, can not reuse command
+    // list with update
     for (auto& cmdList : commandLists) {
         zeCommandListReset(cmdList);
     }
 
     // Lazily create the VM execution context (owned by argsImpl, destroyed with it).
     if (params->executionContext == nullptr) {
-        if (npuVMRuntimeCreateExecutionContext(vmRuntime, &params->executionContext) !=
-            NPU_VM_RUNTIME_RESULT_SUCCESS) {
+        if (npuVMRuntimeCreateExecutionContext(vmRuntime, &params->executionContext) != NPU_VM_RUNTIME_RESULT_SUCCESS) {
             OPENVINO_THROW("Failed to create a VM execution context");
         }
         _logger.debug("execute_vm_runtime - execution context created");
@@ -278,6 +278,7 @@ void DynamicPipeline::execute_vm_runtime(npu_vm_runtime_handle_t vmRuntime,
     params->inferenceFence = fence;
     params->event = event;
 
+    _logger.debug("Execute graph with runtime engine");
     if (npuVMRuntimeExecute(vmRuntime, params) != NPU_VM_RUNTIME_RESULT_SUCCESS) {
         OPENVINO_THROW("Failed to execute VM runtime engine");
     }
@@ -286,7 +287,7 @@ void DynamicPipeline::execute_vm_runtime(npu_vm_runtime_handle_t vmRuntime,
         args._impl = argsImpl;
     }
 
-    _logger.debug("execute_vm_runtime - completed");
+    _logger.debug("Completed to execute graph with runtime engine");
 }
 
 void DynamicPipeline::predict_output_shape(const IGraph& graph,
@@ -301,8 +302,7 @@ void DynamicPipeline::predict_output_shape(const IGraph& graph,
     std::vector<npu_vm_runtime_mem_ref_handle_t> inputs;
     inputs.reserve(inputDescriptors.size());
     for (auto& in : inputDescriptors) {
-        std::shared_ptr<DynamicGraphMemRefImpl> inImpl =
-            std::static_pointer_cast<DynamicGraphMemRefImpl>(in._impl);
+        std::shared_ptr<DynamicGraphMemRefImpl> inImpl = std::static_pointer_cast<DynamicGraphMemRefImpl>(in._impl);
         if (inImpl == nullptr) {
             inImpl = std::make_shared<DynamicGraphMemRefImpl>();
             in._impl = inImpl;
@@ -314,8 +314,7 @@ void DynamicPipeline::predict_output_shape(const IGraph& graph,
     std::vector<npu_vm_runtime_mem_ref_handle_t> outputs;
     outputs.reserve(outputDescriptors.size());
     for (auto& out : outputDescriptors) {
-        std::shared_ptr<DynamicGraphMemRefImpl> outImpl =
-            std::static_pointer_cast<DynamicGraphMemRefImpl>(out._impl);
+        std::shared_ptr<DynamicGraphMemRefImpl> outImpl = std::static_pointer_cast<DynamicGraphMemRefImpl>(out._impl);
         if (outImpl == nullptr) {
             outImpl = std::make_shared<DynamicGraphMemRefImpl>();
             out._impl = outImpl;
@@ -335,8 +334,7 @@ void DynamicPipeline::predict_output_shape(const IGraph& graph,
     }
 
     for (auto& out : outputDescriptors) {
-        std::shared_ptr<DynamicGraphMemRefImpl> outImpl =
-            std::static_pointer_cast<DynamicGraphMemRefImpl>(out._impl);
+        std::shared_ptr<DynamicGraphMemRefImpl> outImpl = std::static_pointer_cast<DynamicGraphMemRefImpl>(out._impl);
         OPENVINO_ASSERT(outImpl != nullptr,
                         "MemRefType implementation is broken, unknown error happens in shape prediction.");
         outImpl->alignWithHandle(out);

…ude/intel_npu/common/graph_arguments.hpp …l_npu/common/dynamic_graph_arguments.hppsrc/plugins/intel_npu/src/common/include/intel_npu/common/graph_arguments.hpp renamed to src/plugins/intel_npu/src/common/include/intel_npu/common/dynamic_graph_arguments.hpp src/plugins/intel_npu/src/common/include/intel_npu/common/graph_arguments.hpp renamed to src/plugins/intel_npu/src/common/include/intel_npu/common/dynamic_graph_arguments.hpp

@@ -4,14 +4,17 @@
 
 #pragma once
 
-#include "intel_npu/common/igraph.hpp"
-#include "intel_npu/utils/zero/zero_init.hpp"
+#include <cstdint>
+#include <iosfwd>
+#include <memory>
+#include <string>
+#include <vector>
 
-namespace intel_npu {
+#include "openvino/core/except.hpp"
+#include "openvino/core/shape.hpp"
+#include "openvino/runtime/itensor.hpp"
 
-/// Host-side description of a single VM-runtime MemRef. Used as the per-IO entry inside
-/// @ref GraphArguments. The opaque @c _impl pointer carries a @c DynamicGraphMemRefImpl
-/// (defined in dynamic_graph_vm_impl.hpp) which owns the underlying VM MemRef handle.
+namespace intel_npu {
 struct MemRefType {
     const void* _basePtr;
     const void* _data;
@@ -46,9 +49,6 @@ struct MemRefType {
     std::string toString();
 };
 
-/// Runtime carrier for VM-pipeline inputs/outputs. Owned by @c PipelinedCommandLists.
-/// The opaque @c _impl holds a @c DynamicGraphArgumentsImpl (VM execution context + cached
-/// mem-ref vectors), created lazily on first VM execution.
 struct GraphArguments {
     std::vector<MemRefType> _inputs;
     std::vector<MemRefType> _outputs;

…tel_npu/common/dynamic_graph_vm_impl.hpp …/common/dynamic_graph_arguments_impl.hppsrc/plugins/intel_npu/src/common/include/intel_npu/common/dynamic_graph_vm_impl.hpp renamed to src/plugins/intel_npu/src/common/include/intel_npu/common/dynamic_graph_arguments_impl.hpp src/plugins/intel_npu/src/common/include/intel_npu/common/dynamic_graph_vm_impl.hpp renamed to src/plugins/intel_npu/src/common/include/intel_npu/common/dynamic_graph_arguments_impl.hpp

@@ -7,7 +7,7 @@
 #include <memory>
 #include <vector>
 
-#include "intel_npu/common/graph_arguments.hpp"
+#include "intel_npu/common/dynamic_graph_arguments.hpp"
 #include "intel_npu/utils/vm/npu_vm_runtime_api.hpp"
 #include "openvino/core/except.hpp"
 

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

@@ -21,7 +21,6 @@
 // to dereference / call into the VM runtime. Keeping it forward-declared here avoids
 // pulling the VM headers into every consumer of the IGraph interface.
 struct _npu_vm_runtime_handle_t;
-using npu_vm_runtime_handle_t = struct _npu_vm_runtime_handle_t*;
 
 namespace intel_npu {
 
@@ -42,6 +41,7 @@ class IGraph : public std::enable_shared_from_this<IGraph> {
     virtual std::vector<ov::ProfilingInfo> process_profiling_output(const std::vector<uint8_t>& profData) const;
 
     virtual void set_argument_value(uint32_t id, const void* data) const;
+
     virtual void set_argument_value_with_strides(uint32_t id,
                                                  const void* data,
                                                  const std::vector<size_t>& strides) const;
@@ -87,7 +87,7 @@ class IGraph : public std::enable_shared_from_this<IGraph> {
     /// Return the VM-runtime engine handle backing this graph, or nullptr if the graph does
     /// not use the VM-runtime (e.g. static driver-only graphs). Used by the dynamic pipeline
     /// to drive @c npuVMRuntimeExecute.
-    virtual npu_vm_runtime_handle_t get_vm_runtime_handle() const;
+    virtual _npu_vm_runtime_handle_t* get_vm_runtime_handle() const;
 
     /// Return the number of subgraphs this graph contains. Defaults to 1 for static graphs.
     virtual uint64_t get_num_subgraphs() const;

…l_npu/src/common/src/graph_arguments.cpp …c/common/src/dynamic_graph_arguments.cppsrc/plugins/intel_npu/src/common/src/graph_arguments.cpp renamed to src/plugins/intel_npu/src/common/src/dynamic_graph_arguments.cpp src/plugins/intel_npu/src/common/src/graph_arguments.cpp renamed to src/plugins/intel_npu/src/common/src/dynamic_graph_arguments.cpp

@@ -2,11 +2,9 @@
 // SPDX-License-Identifier: Apache-2.0
 //
 
-#include "intel_npu/common/graph_arguments.hpp"
-
 #include <sstream>
 
-#include "openvino/core/except.hpp"
+#include "intel_npu/common/dynamic_graph_arguments.hpp"
 
 namespace intel_npu {
 

src/plugins/intel_npu/src/common/src/igraph.cpp

@@ -28,7 +28,7 @@ std::optional<std::string_view> IGraph::get_compatibility_descriptor() const {
     OPENVINO_THROW("get_compatibility_descriptor not implemented");
 }
 
-npu_vm_runtime_handle_t IGraph::get_vm_runtime_handle() const {
+_npu_vm_runtime_handle_t* IGraph::get_vm_runtime_handle() const {
     return nullptr;
 }
 

src/plugins/intel_npu/src/compiler_adapter/include/dynamic_graph.hpp

@@ -8,8 +8,7 @@
 
 #include <mutex>
 
-#include "intel_npu/common/dynamic_graph_vm_impl.hpp"
-#include "intel_npu/common/graph_arguments.hpp"
+#include "intel_npu/common/igraph.hpp"
 #include "intel_npu/common/network_metadata.hpp"
 #include "intel_npu/utils/vm/npu_vm_runtime_api.hpp"
 #include "intel_npu/utils/zero/zero_init.hpp"
@@ -18,12 +17,7 @@
 namespace intel_npu {
 class DynamicGraph final : public IGraph {
 public:
-    using MemRefTypeImpl = DynamicGraphMemRefImpl;
-    using GraphArgumentsImpl = DynamicGraphArgumentsImpl;
-
     class Impl {
-        using MemRefTypeImpl = DynamicGraph::MemRefTypeImpl;
-
     public:
         virtual void initialize(std::optional<ov::Tensor>& blob, NetworkMetadata& metadata) = 0;
         virtual uint64_t getNumSubgraphs() = 0;