Only use current thread with TBB when -nstreams=0 (#20975)

* only use current thread when -nstreams=0 with TBB

* add comments

* fix building issue

* fix comments

* only use current thread when -nstreams=0 with TBB

* add comments

* fix building issue

* fix comments

src/inference/dev_api/openvino/runtime/threading/istreams_executor.hpp

@@ -194,6 +194,11 @@ public:
                                     int threads_per_stream,
                                     PreferredCoreType core_type,
                                     bool cpu_pinning);
+        /**
+         * @brief Set _streams_info_table and _cpu_reservation in cpu streams executor config when nstreams = 0,
+         *        that is, only create one thread with TBB
+         */
+        void set_config_zero_stream();
     };
 
     /**

src/inference/src/dev/threading/cpu_streams_executor.cpp

@@ -186,7 +186,10 @@ struct CPUStreamsExecutor::Impl {
             int max_threads_per_core;
             StreamCreateType stream_type;
             const auto org_proc_type_table = get_org_proc_type_table();
-            const auto stream_id = _streamId >= _impl->_config._streams ? _impl->_config._streams - 1 : _streamId;
+            const auto stream_id =
+                _impl->_config._streams == 0
+                    ? 0
+                    : (_streamId >= _impl->_config._streams ? _impl->_config._streams - 1 : _streamId);
 
             get_cur_stream_info(stream_id,
                                 _impl->_config._cpu_reservation,

src/inference/src/dev/threading/istreams_executor.cpp

@@ -659,5 +659,22 @@ void IStreamsExecutor::Config::update_executor_config(int stream_nums,
     }
 }
 
+void IStreamsExecutor::Config::set_config_zero_stream() {
+    std::vector<std::vector<int>> proc_type_table = get_proc_type_table();
+    int core_type = MAIN_CORE_PROC;
+    int numa_id = 0;
+    int socket_id = 0;
+
+    if (proc_type_table.size() > 0) {
+        core_type = proc_type_table[0][MAIN_CORE_PROC] > 0
+                        ? MAIN_CORE_PROC
+                        : (proc_type_table[0][EFFICIENT_CORE_PROC] > 0 ? EFFICIENT_CORE_PROC : HYPER_THREADING_PROC);
+        numa_id = std::max(0, proc_type_table[0][PROC_NUMA_NODE_ID]);
+        socket_id = std::max(0, proc_type_table[0][PROC_SOCKET_ID]);
+    }
+    _streams_info_table.push_back({1, core_type, 1, numa_id, socket_id});
+    _cpu_reservation = false;
+}
+
 }  // namespace threading
 }  // namespace ov

src/plugins/intel_cpu/src/compiled_model.cpp

@@ -247,7 +247,7 @@ ov::Any CompiledModel::get_property(const std::string& name) const {
     } else if (name == ov::optimal_number_of_infer_requests) {
         const auto streams = config.streamExecutorConfig._streams;
         return decltype(ov::optimal_number_of_infer_requests)::value_type(
-            streams);  // ov::optimal_number_of_infer_requests has no negative values
+            streams > 0 ? streams : 1);  // ov::optimal_number_of_infer_requests has no negative values
     } else if (name == ov::num_streams) {
         const auto streams = config.streamExecutorConfig._streams;
         return decltype(ov::num_streams)::value_type(

src/plugins/intel_cpu/src/plugin.cpp

@@ -330,6 +330,8 @@ void Engine::get_performance_streams(Config& config, const std::shared_ptr<ov::M
 
     if (!((0 == config.streamExecutorConfig._streams) && config.streamExecutorConfig._streams_changed)) {
         get_num_streams(streams, model, config);
+    } else {
+        config.streamExecutorConfig.set_config_zero_stream();
     }
 
     OPENVINO_SUPPRESS_DEPRECATED_START