Use vclAllocateExecutionCreate3 to get metadata

Signed-off-by: Xin Wang <[email protected]>

Author: XinWangIntel

src/plugins/intel_npu/src/compiler_adapter/include/npu_driver_compiler.h

@@ -23,7 +23,7 @@ extern "C" {
 #endif
 
 #define VCL_COMPILER_VERSION_MAJOR  7
-#define VCL_COMPILER_VERSION_MINOR  4
+#define VCL_COMPILER_VERSION_MINOR  5
 #define VCL_PROFILING_VERSION_MAJOR 2
 #define VCL_PROFILING_VERSION_MINOR 0
 
@@ -272,6 +272,13 @@ VCL_APIEXPORT vcl_result_t VCL_APICALL vclAllocatedExecutableCreate2(vcl_compile
                                                                      uint8_t** blobBuffer,
                                                                      uint64_t* blobSize);
 
+VCL_APIEXPORT vcl_result_t VCL_APICALL vclAllocatedExecutableCreate3(vcl_compiler_handle_t compiler,
+                                                                     vcl_executable_desc_t desc,
+                                                                     vcl_allocator2_t* allocator,
+                                                                     uint8_t** blobBuffer,
+                                                                     uint64_t* blobSize,
+                                                                     void* metadata);
+
 ///////////////////////////////////////////////////////////////////////////////
 /// @brief Destroys the executable and releases the cached blob.
 VCL_APIEXPORT vcl_result_t VCL_APICALL vclExecutableDestroy(vcl_executable_handle_t executable);

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

@@ -34,6 +34,7 @@ namespace intel_npu {
 
 //unsupported symbols with older ze_loader versions
 #define vcl_weak_symbols_list()                             \
+    vcl_symbol_statement(vclAllocatedExecutableCreate3)     \
     vcl_symbol_statement(vclAllocatedExecutableCreate2)     \
     vcl_symbol_statement(vclGetCompilerSupportedOptions)    \
     vcl_symbol_statement(vclGetCompilerIsOptionSupported)

src/plugins/intel_npu/src/compiler_adapter/src/plugin_compiler_adapter.cpp

@@ -106,17 +106,20 @@ std::shared_ptr<IGraph> PluginCompilerAdapter::compile(const std::shared_ptr<con
     ov::Tensor tensor = make_tensor_from_vector(networkDesc.compiledNetwork);
     ze_graph_handle_t graphHandle = nullptr;
 
-    NetworkMetadata networkMeta;
+    NetworkMetadata networkMeta = std::move(networkDesc.metadata);
     if (_zeGraphExt) {
         // Depending on the config, we may get an error when trying to get the graph handle from the compiled
         // network
         try {
             graphHandle =
                 _zeGraphExt->getGraphHandle(*reinterpret_cast<const uint8_t*>(tensor.data()), tensor.get_byte_size());
 #ifdef VCL_FOR_COMPILER
-            // For VCL, we need to get metadata from driver parser
-            networkMeta = _zeGraphExt->getNetworkMeta(graphHandle);
-            networkMeta.name = model->get_friendly_name();
+            if (networkMeta.inputs.empty() && networkMeta.outputs.empty()) {
+                // If the metadata is empty, we can try to get it from the driver parser
+                _logger.info("Metadata is empty, trying to get it from the driver parser");
+                networkMeta = _zeGraphExt->getNetworkMeta(graphHandle);
+                networkMeta.name = model->get_friendly_name();
+            }
 #endif
         } catch (...) {
             _logger.info("Failed to obtain the level zero graph handle. Inference requests for this model are not "

src/plugins/intel_npu/src/compiler_adapter/src/vcl_api.cpp

@@ -208,9 +208,27 @@ NetworkDescription VCLCompilerImpl::compile(const std::shared_ptr<const ov::Mode
                                      buildFlags.c_str(),
                                      buildFlags.size()};
     _logger.debug("compiler vcl version: %d.%d", _vclVersion.major, _vclVersion.minor);
-    if (_vclVersion.major >= 7 && _vclVersion.minor >= 4) {
+    if (_vclVersion.major >= 7 && _vclVersion.minor >= 5) {
+        // For VCL 7.5 and later, we can use vclAllocatedExecutableCreate2
+        _logger.debug("Using vclAllocatedExecutableCreate2 for 7.5 <= VCL");
+        vcl_allocator_vector allocator;
+        uint8_t* blob = nullptr;
+        size_t size = 0;
+        NetworkMetadata metadata;
+
+        THROW_ON_FAIL_FOR_VCL(
+            "vclAllocatedExecutableCreate3",
+            vclAllocatedExecutableCreate3(_compilerHandle, exeDesc, &allocator, &blob, &size, &metadata),
+            _logHandle);
+        if (size == 0 || blob == nullptr) {
+            OPENVINO_THROW("Failed to create VCL executable, size is zero or blob is null");
+        }
+
+        _logger.debug("compile end, blob size:%d", allocator.m_vec.size());
+        return NetworkDescription(std::move(allocator.m_vec), std::move(metadata));
+    } else if (_vclVersion.major >= 7 && _vclVersion.minor >= 4) {
         // For VCL 7.4 and later, we can use vclAllocatedExecutableCreate2
-        _logger.debug("Using vclAllocatedExecutableCreate2 for VCL 7.4+");
+        _logger.debug("Using vclAllocatedExecutableCreate2 for 7.4 <= VCL < 7.5");
         vcl_allocator_vector allocator;
         uint8_t* blob = nullptr;
         size_t size = 0;