[CPU] Enable CPU pinning on windows in multiple sockets machine (#29283)

### Details:
 - *Enable CPU pinning on windows in multiple sockets machine*

### Tickets:
 - *CVS-163881*

Author: sunxiaoxia2022

src/inference/dev_api/openvino/runtime/system_conf.hpp

@@ -188,6 +188,13 @@ OPENVINO_RUNTIME_API int get_num_numa_nodes();
  */
 OPENVINO_RUNTIME_API int get_num_sockets();
 
+/**
+ * @brief      Get numa node id of cpu_id
+ * @ingroup    ov_dev_api_system_conf
+ * @return     Numa node id
+ */
+OPENVINO_RUNTIME_API int get_numa_node_id(int cpu_id);
+
 /**
  * @brief      Returns a table of number of processor types on Linux/Windows
  * @ingroup    ov_dev_api_system_conf

src/inference/src/dev/threading/cpu_streams_executor_internal.cpp

@@ -57,10 +57,6 @@ void get_cur_stream_info(const int stream_id,
 
 #if defined(__APPLE__)
     pinning = false;
-#elif defined(_WIN32)
-    if (proc_type_table.size() > 1) {
-        pinning = false;
-    }
 #endif
     if (pinning) {
         stream_type = STREAM_WITH_OBSERVE;

src/inference/src/dev/threading/thread_affinity.cpp

@@ -126,7 +126,19 @@ bool pin_thread_to_vacant_core(int thrIdx,
                                int ncores,
                                const CpuSet& procMask,
                                const std::vector<int>& cpu_ids) {
-    return 0 != SetThreadAffinityMask(GetCurrentThread(), DWORD_PTR(1) << cpu_ids[thrIdx]);
+    auto proc_type_table = get_proc_type_table();
+    if (proc_type_table.size() > 1) {
+        int cores_in_numa = proc_type_table[1][MAIN_CORE_PROC] + proc_type_table[1][HYPER_THREADING_PROC];
+        GROUP_AFFINITY group;
+        group.Group = get_numa_node_id(cpu_ids[thrIdx]);
+        group.Mask = DWORD_PTR(1) << (cpu_ids[thrIdx] % cores_in_numa);
+        group.Reserved[0] = 0;
+        group.Reserved[1] = 0;
+        group.Reserved[2] = 0;
+        return 0 != SetThreadGroupAffinity(GetCurrentThread(), &group, NULL);
+    } else {
+        return 0 != SetThreadAffinityMask(GetCurrentThread(), DWORD_PTR(1) << cpu_ids[thrIdx]);
+    }
 }
 bool pin_current_thread_by_mask(int ncores, const CpuSet& procMask) {
     DWORD_PTR mask = *procMask.get();

src/inference/src/system_conf.cpp

@@ -277,6 +277,9 @@ int get_num_numa_nodes() {
 int get_num_sockets() {
     return -1;
 }
+int get_numa_node_id(int cpu_id) {
+    return -1;
+}
 void reserve_available_cpus(const std::vector<std::vector<int>> streams_info_table,
                             std::vector<std::vector<int>>& stream_processors,
                             const int cpu_status) {}
@@ -335,6 +338,9 @@ int get_num_numa_nodes() {
 int get_num_sockets() {
     return cpu_info()._sockets;
 }
+int get_numa_node_id(int cpu_id) {
+    return -1;
+}
 void reserve_available_cpus(const std::vector<std::vector<int>> streams_info_table,
                             std::vector<std::vector<int>>& stream_processors,
                             const int cpu_status) {}
@@ -459,6 +465,11 @@ int get_num_sockets() {
     return cpu_info()._sockets;
 }
 
+int get_numa_node_id(int cpu_id) {
+    CPU& cpu = cpu_info();
+    return cpu._cpu_mapping_table[cpu_id][CPU_MAP_NUMA_NODE_ID];
+}
+
 void reserve_available_cpus(const std::vector<std::vector<int>> streams_info_table,
                             std::vector<std::vector<int>>& stream_processors,
                             const int cpu_status) {

src/inference/tests/unit/cpu_stream_info_test.cpp

@@ -283,7 +283,6 @@ LinuxCpuStreamTypeCase _2sockets_72cores_binding_9streams = {
      {1, ALL_PROC, 4, -1, -1},
      {0, MAIN_CORE_PROC, 2, 0, 0},
      {0, MAIN_CORE_PROC, 2, 1, 1}},
-#    if defined(__linux__)
     {
         STREAM_WITH_OBSERVE,
         STREAM_WITH_OBSERVE,
@@ -295,19 +294,6 @@ LinuxCpuStreamTypeCase _2sockets_72cores_binding_9streams = {
         STREAM_WITH_OBSERVE,
         STREAM_WITH_OBSERVE,
     },
-#    else
-    {
-        STREAM_WITH_NUMA_ID,
-        STREAM_WITH_NUMA_ID,
-        STREAM_WITH_NUMA_ID,
-        STREAM_WITH_NUMA_ID,
-        STREAM_WITH_NUMA_ID,
-        STREAM_WITH_NUMA_ID,
-        STREAM_WITH_NUMA_ID,
-        STREAM_WITH_NUMA_ID,
-        STREAM_WITHOUT_PARAM,
-    },
-#    endif
     {4, 4, 4, 4, 4, 4, 4, 4, 4},
     {
         MAIN_CORE_PROC,

src/plugins/intel_cpu/src/cpu_map_scheduling.cpp

@@ -78,12 +78,7 @@ bool check_cpu_pinning(const bool cpu_pinning,
 #if defined(__APPLE__)
     result_value = false;
 #elif defined(_WIN32)
-    auto proc_type_table = get_proc_type_table();
-    if (proc_type_table.size() == 1) {
-        result_value = cpu_pinning_changed ? cpu_pinning : cpu_reservation;
-    } else {
-        result_value = false;
-    }
+    result_value = cpu_pinning_changed ? cpu_pinning : cpu_reservation;
 #else
     // The following code disables pinning in case stream contains both Pcore and Ecore
     auto hyper_cores_in_stream = [&]() {

src/plugins/intel_cpu/tests/unit/streams_info/cpu_pinning_test.cpp

@@ -131,7 +131,7 @@ CpuPinningTestCase cpu_pinning_win_mock_set_true_2 = {
     },
     {{48, 48, 0, 0, -1, -1}, {24, 24, 0, 0, 0, 0}, {24, 24, 0, 0, 1, 1}},
     {{1, MAIN_CORE_PROC, 24, 0, 0}},
-    false,
+    true,
 };
 CpuPinningTestCase cpu_pinning_win_mock_set_false = {
     false,