Fixed coverity issues of Low impact

src/plugins/intel_npu/src/plugin/npuw/base_sync_infer_request.cpp

@@ -504,8 +504,7 @@ void ov::npuw::IBaseInferRequest::dump_input_tensors(std::size_t idx) {
     // Note:
     // - _name is used for the user option (no leading 00s for indices)
     // - _path is used for disk dump (will have leading 00s for indices)
-    const auto comp_submodel_name = subgr_name(idx);
-    const auto comp_submodel_path = m_npuw_model->m_name + subgr_path_suffix(idx) + iter_path_suffix(idx);
+    const auto& comp_submodel_path = m_npuw_model->m_name + subgr_path_suffix(idx) + iter_path_suffix(idx);
     const auto num_inputs = comp_submodel->inputs().size();
 
     // There's different approaches to dumping normal and spatial subgraphs.
@@ -584,8 +583,7 @@ void ov::npuw::IBaseInferRequest::dump_output_tensors(std::size_t idx) {
     // - _name is used for the user option (no leading 00s for indices)
     // - _path is used for disk dump (will have leading 00s for indices)
     // FIXME: Duplication is evil
-    const auto comp_submodel_name = subgr_name(idx);
-    const auto comp_submodel_path = m_npuw_model->m_name + subgr_path_suffix(idx) + iter_path_suffix(idx);
+    const auto& comp_submodel_path = m_npuw_model->m_name + subgr_path_suffix(idx) + iter_path_suffix(idx);
     const std::size_t num_outputs = comp_submodel->outputs().size();
 
     // Same approach as in above. Spatial tensors require special handling

src/plugins/intel_npu/src/plugin/npuw/compiled_model.cpp

@@ -695,7 +695,7 @@ std::shared_ptr<ov::npuw::CompiledModel> ov::npuw::CompiledModel::deserialize(
         read(stream, key);
         ov::Any val;
         read_any(stream, val);
-        compiled->m_non_npuw_props[key] = val;
+        compiled->m_non_npuw_props[key] = std::move(val);
     }
     compiled->implement_properties();
 
@@ -831,7 +831,7 @@ void ov::npuw::CompiledModel::detach_memory() {
 
 std::string ov::npuw::CompiledModel::global_mem_device() const {
     // Force globally set device if set
-    const std::string device_alloc = m_cfg.get<::intel_npu::NPUW_WEIGHTS_BANK_ALLOC>();
+    const std::string& device_alloc = m_cfg.get<::intel_npu::NPUW_WEIGHTS_BANK_ALLOC>();
     if (!device_alloc.empty()) {
         return device_alloc;
     }
@@ -852,7 +852,7 @@ std::string ov::npuw::CompiledModel::global_mem_device() const {
 
 std::string ov::npuw::CompiledModel::funcall_mem_device(const std::size_t idx) const {
     // Force globally set device if set
-    const std::string device_alloc = m_cfg.get<::intel_npu::NPUW_WEIGHTS_BANK_ALLOC>();
+    const std::string& device_alloc = m_cfg.get<::intel_npu::NPUW_WEIGHTS_BANK_ALLOC>();
     if (!device_alloc.empty()) {
         return device_alloc;
     }

src/plugins/intel_npu/src/plugin/npuw/llm_compiled_model.cpp

@@ -225,7 +225,7 @@ std::shared_ptr<ov::Model> redirect_new_kv_to_output(const std::shared_ptr<ov::M
 
 std::shared_ptr<ov::Model> cvt_value_tensors_layout(std::shared_ptr<ov::Model> model) {
     ov::preprocess::PrePostProcessor ppp(model);
-    for (auto tensor : model->outputs()) {
+    for (const auto& tensor : model->outputs()) {
         if (tensor.get_any_name().find("value") != std::string::npos) {
             // NB: [batch, num_heads, seq_len, emb_size] -> [batch, num_heads, emb_size, seq_len]
             ppp.output(tensor.get_any_name()).model().set_layout(ov::Layout("BHSE"));
@@ -311,7 +311,7 @@ std::optional<NPUDesc> extract_npu_descriptor(const std::shared_ptr<const ov::IP
     const std::string arch = plugin->get_property(ov::device::architecture.name(), ov::AnyMap{}).as<std::string>();
     const int64_t max_tiles = plugin->get_property(ov::intel_npu::max_tiles.name(), ov::AnyMap{}).as<int64_t>();
     bool compiler_dq = false;
-    const auto supported_properties =
+    const auto& supported_properties =
         plugin->get_property(ov::supported_properties.name(), ov::AnyMap{}).as<std::vector<ov::PropertyName>>();
     if (std::find(supported_properties.begin(), supported_properties.end(), "NPU_COMPILER_DYNAMIC_QUANTIZATION") !=
         supported_properties.end()) {
@@ -506,9 +506,9 @@ ov::npuw::LLMCompiledModel::LLMCompiledModel(const std::shared_ptr<ov::Model>& m
         generate_config_opt.value_or(get_default_generate_config(kvcache_model, npudesc, generate_hint))
             .as<ov::AnyMap>();
 
-    auto prefill_config_addition_value =
+    const auto& prefill_config_addition_value =
         prefill_config_addition.has_value() ? prefill_config_addition.value().as<ov::AnyMap>() : ov::AnyMap{};
-    auto generate_config_addition_value =
+    const auto& generate_config_addition_value =
         generate_config_addition.has_value() ? generate_config_addition.value().as<ov::AnyMap>() : ov::AnyMap{};
 
     merge_config_with(prefill_config, other_props);

src/plugins/intel_npu/src/plugin/npuw/llm_infer_request.cpp

@@ -78,8 +78,8 @@ void copy_columns_by_row_chunks(ov::SoPtr<ov::ITensor> src, ov::SoPtr<ov::ITenso
     OPENVINO_ASSERT(src_shape == dst->get_shape());
     OPENVINO_ASSERT(src->get_byte_size() == dst->get_byte_size());
 
-    const auto src_strides = src->get_strides();
-    const auto dst_strides = dst->get_strides();
+    const auto& src_strides = src->get_strides();
+    const auto& dst_strides = dst->get_strides();
     const auto elem_size = src->get_byte_size() / src->get_size();
 
     const auto C = src_shape[1];

src/plugins/intel_npu/src/plugin/npuw/serialization.hpp

@@ -159,7 +159,7 @@ void read(std::istream& stream, std::vector<T>& var) {
     for (std::size_t i = 0; i < var_size; ++i) {
         T elem;
         read(stream, elem);
-        var.push_back(elem);
+        var.push_back(std::move(elem));
     }
 }
 
@@ -180,7 +180,7 @@ void read(std::istream& stream, std::unordered_set<T>& var) {
     for (std::size_t i = 0; i < var_size; ++i) {
         T elem;
         read(stream, elem);
-        var.insert(elem);
+        var.insert(std::move(elem));
     }
 }