[GNA] Fix accuracy degradation in compact mode (#12150)

src/plugins/intel_gna/gna_plugin.cpp

@@ -892,18 +892,21 @@ void GNAPlugin::LoadNetwork(const CNNNetwork& _network) {
     std::vector<CNNLayerPtr> sortedNoMem;
     std::unordered_map<std::string, std::vector<InferenceEngine::CNNLayerPtr>> memoryPairs;
     // find all memory layers pairs and mark which one used as outputs
-    uint16_t id = 0;
+    int id = 0;
     for (auto& layer : sortedNet) {
         // set order id for layers to use it in compact mode
+        LayerInfo layerInfo(layer);
         IE_SUPPRESS_DEPRECATED_START
-        layer->userValue.v_int = id++;
+        // Increase lifetime of input data buffer needed by DelayedCopy layer to avoid overwritting buffer
+        // by downstream layers.
+        layer->userValue.v_int = layerInfo.isCopyDelayed() ? std::numeric_limits<int>::max() : id++;
         IE_SUPPRESS_DEPRECATED_END
         auto generic = dynamic_cast<GenericLayer*>(layer.get());
         if (generic == nullptr) {
             sortedNoMem.push_back(layer);
             continue;
         }
-        LayerInfo layerInfo(layer);
+
         if (layerInfo.isMemory()) {
             // collect all memory pairs
             auto id = generic->GetParamAsString("id");