From ae1beb296e08ac37818fe44273cb30f92b2eee41 Mon Sep 17 00:00:00 2001
From: Sun Xiaoxia <xiaoxia.sun@intel.com>
Date: Wed, 17 May 2023 08:08:38 +0000
Subject: [PATCH] remove definition of static CPU (#17393)

* remove definition of static CPU

* revert old format

* definition of cpu_info covers macOS

* add lock in is_cpu_map_available
---
 src/inference/src/streams_executor.hpp |  2 ++
 src/inference/src/system_conf.cpp      | 18 ++++++++++++++----
 2 files changed, 16 insertions(+), 4 deletions(-)

diff --git a/src/inference/src/streams_executor.hpp b/src/inference/src/streams_executor.hpp
index 0fd7bc424df..d2af0454850 100644
--- a/src/inference/src/streams_executor.hpp
+++ b/src/inference/src/streams_executor.hpp
@@ -31,6 +31,8 @@ public:
     int _num_threads = 0;
 };
 
+CPU& cpu_info();
+
 #ifdef __linux__
 /**
  * @brief      Parse processors infomation on Linux
diff --git a/src/inference/src/system_conf.cpp b/src/inference/src/system_conf.cpp
index 19b42b9d3e2..3117ae1a0ba 100644
--- a/src/inference/src/system_conf.cpp
+++ b/src/inference/src/system_conf.cpp
@@ -151,6 +151,11 @@ bool check_open_mp_env_vars(bool include_omp_num_threads) {
     return false;
 }
 
+CPU& cpu_info() {
+    static CPU cpu;
+    return cpu;
+}
+
 #if defined(__APPLE__) || defined(__EMSCRIPTEN__)
 // for Linux and Windows the getNumberOfCPUCores (that accounts only for physical cores) implementation is OS-specific
 // (see cpp files in corresponding folders), for __APPLE__ it is default :
@@ -181,10 +186,9 @@ void set_cpu_used(const std::vector<int>& cpu_ids, const int used) {}
 
 #else
 
-static CPU cpu;
-
 #    ifndef _WIN32
 int get_number_of_cpu_cores(bool bigCoresOnly) {
+    CPU& cpu = cpu_info();
     unsigned numberOfProcessors = cpu._processors;
     unsigned totalNumberOfCpuCores = cpu._cores;
     IE_ASSERT(totalNumberOfCpuCores != 0);
@@ -217,6 +221,7 @@ int get_number_of_cpu_cores(bool bigCoresOnly) {
 
 #        if !((OV_THREAD == OV_THREAD_TBB || OV_THREAD == OV_THREAD_TBB_AUTO))
 std::vector<int> get_available_numa_nodes() {
+    CPU& cpu = cpu_info();
     std::vector<int> nodes((0 == cpu._numa_nodes) ? 1 : cpu._numa_nodes);
     std::iota(std::begin(nodes), std::end(nodes), 0);
     return nodes;
@@ -225,16 +230,19 @@ std::vector<int> get_available_numa_nodes() {
 #    endif
 
 std::vector<std::vector<int>> get_proc_type_table() {
+    CPU& cpu = cpu_info();
     std::lock_guard<std::mutex> lock{cpu._cpu_mutex};
     return cpu._proc_type_table;
 }
 
 bool is_cpu_map_available() {
+    CPU& cpu = cpu_info();
+    std::lock_guard<std::mutex> lock{cpu._cpu_mutex};
     return cpu._proc_type_table.size() > 0 && cpu._num_threads == cpu._proc_type_table[0][ALL_PROC];
 }
 
 int get_num_numa_nodes() {
-    return cpu._numa_nodes;
+    return cpu_info()._numa_nodes;
 }
 
 std::vector<std::vector<int>> reserve_available_cpus(const std::vector<std::vector<int>> streams_info_table) {
@@ -242,6 +250,7 @@ std::vector<std::vector<int>> reserve_available_cpus(const std::vector<std::vect
     int info_table_size = static_cast<int>(streams_info_table.size());
     std::vector<std::vector<int>> stream_ids;
     std::vector<std::vector<std::vector<int>>> res_stream_ids;
+    CPU& cpu = cpu_info();
     stream_ids.assign(info_table_size, std::vector<int>());
     res_stream_ids.assign(info_table_size, std::vector<std::vector<int>>());
 
@@ -275,6 +284,7 @@ std::vector<std::vector<int>> reserve_available_cpus(const std::vector<std::vect
 }
 
 void set_cpu_used(const std::vector<int>& cpu_ids, const int used) {
+    CPU& cpu = cpu_info();
     std::lock_guard<std::mutex> lock{cpu._cpu_mutex};
     const auto cpu_size = static_cast<int>(cpu_ids.size());
     for (int i = 0; i < cpu_size; i++) {
@@ -286,7 +296,7 @@ void set_cpu_used(const std::vector<int>& cpu_ids, const int used) {
     if (used == NOT_USED || used >= PLUGIN_USED_START) {
         std::vector<int> all_table;
         int start = cpu._numa_nodes > 1 ? 1 : 0;
-        if (is_cpu_map_available()) {
+        if (cpu._proc_type_table.size() > 0 && cpu._num_threads == cpu._proc_type_table[0][ALL_PROC]) {
             cpu._proc_type_table.assign(cpu._proc_type_table.size(), std::vector<int>(PROC_TYPE_TABLE_SIZE, 0));
             all_table.resize(PROC_TYPE_TABLE_SIZE, 0);
             for (int i = 0; i < cpu._processors; i++) {