[NPU] Use different graph init (#26903)

pereanub · web-flow · commit 0ebff040fd22 · 2024-10-08T12:55:47.000Z
### Details: - *Use different graph init in case graph file schema or elf is used* ### Tickets: - *CVS-154233*
diff --git a/src/plugins/intel_npu/src/backend/include/zero_executor.hpp b/src/plugins/intel_npu/src/backend/include/zero_executor.hpp
@@ -61,6 +61,8 @@ class ZeroExecutor final : public IExecutor {
     }
 
 private:
+    void initialize_graph_through_command_list() const;
+
     const Config _config;
     Logger _logger;
 
@@ -72,7 +74,6 @@ class ZeroExecutor final : public IExecutor {
     const uint32_t _group_ordinal;
 
     ze_graph_handle_t _graph = nullptr;
-    ze_graph_properties_t _props{};
 
     std::vector<ArgumentDescriptor> _input_descriptors;
     std::vector<ArgumentDescriptor> _output_descriptors;
diff --git a/src/plugins/intel_npu/src/backend/include/zero_remote_tensor.hpp b/src/plugins/intel_npu/src/backend/include/zero_remote_tensor.hpp
@@ -45,7 +45,7 @@ class ZeroRemoteTensor : public RemoteTensor {
     void* _mem = nullptr;
     void* _data = nullptr;
 
-    bool _external_memory_support = true;
+    bool _external_memory_support = false;
 };
 
 }  // namespace intel_npu
diff --git a/src/plugins/intel_npu/src/backend/include/zero_types.hpp b/src/plugins/intel_npu/src/backend/include/zero_types.hpp
@@ -16,7 +16,7 @@
 /**
  * @brief Last version of Table of Graph Extension functions used within plugin
  */
-using ze_graph_dditable_ext_last_t = ze_graph_dditable_ext_1_7_t;
+using ze_graph_dditable_ext_last_t = ze_graph_dditable_ext_1_8_t;
 /**
  * @brief Last version of the Command Queue functions used within plugin
  */
@@ -157,10 +157,23 @@ struct ze_graph_dditable_ext_decorator final {
     }
 
     // version 1.7
-    ze_result_t ZE_APICALL pfnGetNativeBinary2(ze_graph_handle_t hGraph, size_t* pSize, uint8_t** pGraphNativeBinary) {
+    ze_result_t ZE_APICALL pfnGetNativeBinary2(ze_graph_handle_t hGraph,
+                                               size_t* pSize,
+                                               const uint8_t** pGraphNativeBinary) {
         throwWhenUnsupported("pfnGetNativeBinary2", ZE_GRAPH_EXT_VERSION_1_7);
         return _impl->pfnGetNativeBinary2(hGraph, pSize, pGraphNativeBinary);
     }
+
+    // version 1.8
+    ze_result_t ZE_APICALL pfnGetProperties2(ze_graph_handle_t hGraph, ze_graph_properties_2_t* pGraphProperties) {
+        throwWhenUnsupported("ze_pfnGraphGetProperties_ext_2_t", ZE_GRAPH_EXT_VERSION_1_8);
+        return _impl->pfnGetProperties2(hGraph, pGraphProperties);
+    }
+
+    ze_result_t ZE_APICALL pfnGraphInitialize(ze_graph_handle_t hGraph) {
+        throwWhenUnsupported("ze_pfnGraphGetProperties_ext_2_t", ZE_GRAPH_EXT_VERSION_1_8);
+        return _impl->pfnGraphInitialize(hGraph);
+    }
 };
 
 /**
diff --git a/src/plugins/intel_npu/src/backend/src/zero_executor.cpp b/src/plugins/intel_npu/src/backend/src/zero_executor.cpp
@@ -37,23 +37,6 @@ ZeroExecutor::ZeroExecutor(const std::shared_ptr<const ZeroInitStructsHolder>& i
                                                      _initStructs->getCommandQueueDdiTable(),
                                                      _config,
                                                      group_ordinal)} {
-    _logger.debug("ZeroExecutor::ZeroExecutor init start - create graph_command_list");
-    OV_ITT_SCOPED_TASK(itt::domains::LevelZeroBackend, "Executor::ZeroExecutor");
-    CommandList graph_command_list(_initStructs->getDevice(),
-                                   _initStructs->getContext(),
-                                   _graph_ddi_table_ext,
-                                   _config,
-                                   _group_ordinal);
-    _logger.debug("ZeroExecutor::ZeroExecutor - create graph_command_queue");
-    CommandQueue graph_command_queue(_initStructs->getDevice(),
-                                     _initStructs->getContext(),
-                                     ZE_COMMAND_QUEUE_PRIORITY_NORMAL,
-                                     _initStructs->getCommandQueueDdiTable(),
-                                     _config,
-                                     _group_ordinal);
-    _logger.debug("ZeroExecutor::ZeroExecutor - create fence");
-    Fence fence(graph_command_queue, _config);
-
     _logger.debug("ZeroExecutor::ZeroExecutor - create graph");
     OV_ITT_TASK_CHAIN(ZERO_EXECUTOR_GRAPH, itt::domains::LevelZeroBackend, "Executor::ZeroExecutor", "graphCreate");
 
@@ -79,7 +62,10 @@ ZeroExecutor::ZeroExecutor(const std::shared_ptr<const ZeroInitStructsHolder>& i
 
     OV_ITT_TASK_NEXT(ZERO_EXECUTOR_GRAPH, "pfnGetProperties");
     _logger.debug("performing pfnGetProperties");
-    zeroUtils::throwOnFail("pfnGetProperties", _graph_ddi_table_ext.pfnGetProperties(_graph, &_props));
+    ze_graph_properties_t props{};
+    props.stype = ZE_STRUCTURE_TYPE_GRAPH_PROPERTIES;
+
+    zeroUtils::throwOnFail("pfnGetProperties", _graph_ddi_table_ext.pfnGetProperties(_graph, &props));
     auto targetDriverExtVersion = _graph_ddi_table_ext.version();
     if (targetDriverExtVersion <= ZE_GRAPH_EXT_VERSION_1_1) {
         OPENVINO_THROW("Incompatibility between the NPU plugin and driver! The driver version is too old, please "
@@ -88,8 +74,9 @@ ZeroExecutor::ZeroExecutor(const std::shared_ptr<const ZeroInitStructsHolder>& i
 
     OV_ITT_TASK_NEXT(ZERO_EXECUTOR_GRAPH, "pfnGetArgumentProperties3");
     _logger.debug("performing pfnGetArgumentProperties3");
-    for (uint32_t index = 0; index < _props.numGraphArgs; ++index) {
-        ze_graph_argument_properties_3_t arg3;
+    for (uint32_t index = 0; index < props.numGraphArgs; ++index) {
+        ze_graph_argument_properties_3_t arg3{};
+        arg3.stype = ZE_STRUCTURE_TYPE_GRAPH_ARGUMENT_PROPERTIES;
         zeroUtils::throwOnFail("pfnGetArgumentProperties3",
                                _graph_ddi_table_ext.pfnGetArgumentProperties3(_graph, index, &arg3));
 
@@ -100,6 +87,51 @@ ZeroExecutor::ZeroExecutor(const std::shared_ptr<const ZeroInitStructsHolder>& i
         }
     }
 
+    if (_graph_ddi_table_ext.version() < ZE_GRAPH_EXT_VERSION_1_8) {
+        initialize_graph_through_command_list();
+    } else {
+        ze_graph_properties_2_t properties = {};
+        properties.stype = ZE_STRUCTURE_TYPE_GRAPH_PROPERTIES;
+        _graph_ddi_table_ext.pfnGetProperties2(_graph, &properties);
+
+        if (properties.initStageRequired & ZE_GRAPH_STAGE_INITIALIZE) {
+            OV_ITT_TASK_NEXT(ZERO_EXECUTOR_GRAPH, "pfnGraphInitialize");
+            _graph_ddi_table_ext.pfnGraphInitialize(_graph);
+        }
+
+        if (properties.initStageRequired & ZE_GRAPH_STAGE_COMMAND_LIST_INITIALIZE) {
+            initialize_graph_through_command_list();
+        }
+    }
+
+    if (config.has<WORKLOAD_TYPE>()) {
+        setWorkloadType(config.get<WORKLOAD_TYPE>());
+    }
+}
+
+void ZeroExecutor::initialize_graph_through_command_list() const {
+    OV_ITT_TASK_CHAIN(ZERO_EXECUTOR_GRAPH,
+                      itt::domains::LevelZeroBackend,
+                      "Executor::ZeroExecutor",
+                      "initialize_graph_through_command_list");
+
+    _logger.debug("ZeroExecutor::ZeroExecutor init start - create graph_command_list");
+    OV_ITT_SCOPED_TASK(itt::domains::LevelZeroBackend, "Executor::ZeroExecutor");
+    CommandList graph_command_list(_initStructs->getDevice(),
+                                   _initStructs->getContext(),
+                                   _graph_ddi_table_ext,
+                                   _config,
+                                   _group_ordinal);
+    _logger.debug("ZeroExecutor::ZeroExecutor - create graph_command_queue");
+    CommandQueue graph_command_queue(_initStructs->getDevice(),
+                                     _initStructs->getContext(),
+                                     ZE_COMMAND_QUEUE_PRIORITY_NORMAL,
+                                     _initStructs->getCommandQueueDdiTable(),
+                                     _config,
+                                     _group_ordinal);
+    _logger.debug("ZeroExecutor::ZeroExecutor - create fence");
+    Fence fence(graph_command_queue, _config);
+
     OV_ITT_TASK_NEXT(ZERO_EXECUTOR_GRAPH, "appendGraphInitialize");
     _logger.debug("ZeroExecutor::ZeroExecutor - performing appendGraphInitialize");
     graph_command_list.appendGraphInitialize(_graph);
@@ -112,10 +144,6 @@ ZeroExecutor::ZeroExecutor(const std::shared_ptr<const ZeroInitStructsHolder>& i
     _logger.debug("ZeroExecutor::ZeroExecutor - performing hostSynchronize");
     fence.hostSynchronize();
     _logger.debug("ZeroExecutor::ZeroExecutor - hostSynchronize completed");
-
-    if (config.has<WORKLOAD_TYPE>()) {
-        setWorkloadType(config.get<WORKLOAD_TYPE>());
-    }
 }
 
 void ZeroExecutor::setWorkloadType(const ov::WorkloadType workloadType) const {
diff --git a/src/plugins/intel_npu/src/backend/src/zero_remote_tensor.cpp b/src/plugins/intel_npu/src/backend/src/zero_remote_tensor.cpp
@@ -44,9 +44,16 @@ ZeroRemoteTensor::ZeroRemoteTensor(std::shared_ptr<ov::IRemoteContext> context,
     ze_device_external_memory_properties_t desc = {};
     desc.stype = ZE_STRUCTURE_TYPE_DEVICE_EXTERNAL_MEMORY_PROPERTIES;
     auto res = zeDeviceGetExternalMemoryProperties(_init_structs->getDevice(), &desc);
-    if (res != ZE_RESULT_SUCCESS || (desc.memoryAllocationImportTypes != ZE_EXTERNAL_MEMORY_TYPE_FLAG_DMA_BUF &&
-                                     desc.memoryAllocationImportTypes != ZE_EXTERNAL_MEMORY_TYPE_FLAG_OPAQUE_WIN32)) {
-        _external_memory_support = false;
+    if (res == ZE_RESULT_SUCCESS) {
+#ifdef _WIN32
+        if (desc.memoryAllocationImportTypes & ZE_EXTERNAL_MEMORY_TYPE_FLAG_OPAQUE_WIN32) {
+            _external_memory_support = true;
+        }
+#else
+        if (desc.memoryAllocationImportTypes & ZE_EXTERNAL_MEMORY_TYPE_FLAG_DMA_BUF) {
+            _external_memory_support = true;
+        }
+#endif
     }
 
     allocate(byte_size);
diff --git a/src/plugins/intel_npu/src/compiler/include/zero_compiler_in_driver.hpp b/src/plugins/intel_npu/src/compiler/include/zero_compiler_in_driver.hpp
@@ -132,14 +132,14 @@ class LevelZeroCompilerInDriver final : public ICompiler {
     void getNativeBinary(ze_graph_dditable_ext_curr_t& graphDdiTableExt,
                          ze_graph_handle_t graphHandle,
                          std::vector<uint8_t>& blob,
-                         uint8_t*& blobPtr,
+                         const uint8_t*& blobPtr,
                          size_t& blobSize) const;
 
     template <typename T = TableExtension, typename std::enable_if_t<!UseCopyForNativeBinary(T), bool> = true>
     void getNativeBinary(ze_graph_dditable_ext_curr_t& graphDdiTableExt,
                          ze_graph_handle_t graphHandle,
                          std::vector<uint8_t>& /* unusedBlob */,
-                         uint8_t*& blobPtr,
+                         const uint8_t*& blobPtr,
                          size_t& blobSize) const;
 
     template <typename T = TableExtension, typename std::enable_if_t<SupportAPIGraphQueryNetworkV2(T), bool> = true>
diff --git a/src/plugins/intel_npu/src/compiler/src/driver_compiler_adapter.cpp b/src/plugins/intel_npu/src/compiler/src/driver_compiler_adapter.cpp
@@ -70,6 +70,12 @@ LevelZeroCompilerAdapter::LevelZeroCompilerAdapter(std::shared_ptr<IEngineBacken
                                                                                               zeContext,
                                                                                               graph_ddi_table_ext);
         break;
+    case ZE_GRAPH_EXT_VERSION_1_8:
+        apiAdapter = std::make_shared<LevelZeroCompilerInDriver<ze_graph_dditable_ext_1_8_t>>(driverHandle,
+                                                                                              deviceHandle,
+                                                                                              zeContext,
+                                                                                              graph_ddi_table_ext);
+        break;
     default:
         apiAdapter = std::make_shared<LevelZeroCompilerInDriver<ze_graph_dditable_ext_1_2_t>>(driverHandle,
                                                                                               deviceHandle,
diff --git a/src/plugins/intel_npu/src/compiler/src/zero_compiler_in_driver.cpp b/src/plugins/intel_npu/src/compiler/src/zero_compiler_in_driver.cpp
@@ -371,7 +371,7 @@ template <typename T, std::enable_if_t<UseCopyForNativeBinary(T), bool>>
 void LevelZeroCompilerInDriver<TableExtension>::getNativeBinary(ze_graph_dditable_ext_curr_t& graphDdiTableExt,
                                                                 ze_graph_handle_t graphHandle,
                                                                 std::vector<uint8_t>& blob,
-                                                                uint8_t*& blobPtr,
+                                                                const uint8_t*& blobPtr,
                                                                 size_t& blobSize) const {
     // Get blob size first
     auto result = _graphDdiTableExt.pfnGetNativeBinary(graphHandle, &blobSize, nullptr);
@@ -408,7 +408,7 @@ template <typename T, std::enable_if_t<!UseCopyForNativeBinary(T), bool>>
 void LevelZeroCompilerInDriver<TableExtension>::getNativeBinary(ze_graph_dditable_ext_curr_t& graphDdiTableExt,
                                                                 ze_graph_handle_t graphHandle,
                                                                 std::vector<uint8_t>& /* unusedBlob */,
-                                                                uint8_t*& blobPtr,
+                                                                const uint8_t*& blobPtr,
                                                                 size_t& blobSize) const {
     // Get blob ptr and size
     auto result = _graphDdiTableExt.pfnGetNativeBinary2(graphHandle, &blobSize, &blobPtr);
@@ -431,7 +431,7 @@ CompiledNetwork LevelZeroCompilerInDriver<TableExtension>::getCompiledNetwork(
         _logger.info("LevelZeroCompilerInDriver getCompiledNetwork get blob from graphHandle");
         ze_graph_handle_t graphHandle = static_cast<ze_graph_handle_t>(networkDescription.metadata.graphHandle);
 
-        uint8_t* blobPtr = nullptr;
+        const uint8_t* blobPtr = nullptr;
         size_t blobSize = -1;
         std::vector<uint8_t> blob;
 
@@ -1243,6 +1243,7 @@ template class LevelZeroCompilerInDriver<ze_graph_dditable_ext_1_4_t>;
 template class LevelZeroCompilerInDriver<ze_graph_dditable_ext_1_5_t>;
 template class LevelZeroCompilerInDriver<ze_graph_dditable_ext_1_6_t>;
 template class LevelZeroCompilerInDriver<ze_graph_dditable_ext_1_7_t>;
+template class LevelZeroCompilerInDriver<ze_graph_dditable_ext_1_8_t>;
 
 }  // namespace driverCompilerAdapter
 }  // namespace intel_npu
diff --git a/src/plugins/intel_npu/thirdparty/level-zero-ext b/src/plugins/intel_npu/thirdparty/level-zero-ext
@@ -1 +1 @@
-Subproject commit 816b5ce120096cbc115b56ed43f8a030eb420b19
+Subproject commit cdb761dd63b1d47230d501e631a2d725db09ba0d
