ThreadIdOffsetの最大値が複数NUMAノード持つCPUの時に正しく指定されていないのを修正した。

source/misc.cpp

@@ -472,6 +472,111 @@ const std::string config_info()
 	return config;
 }
 
+// -----------------
+//    CPU Threads
+// -----------------
+
+#if defined(_WIN32)
+
+// NUMAノード数、NUMAノードを考慮したコア数、スレッド数を取得する。
+bool get_cpu_info(int* nodes, int* cores, int* threads)
+{
+	*nodes = *cores = *threads = 0;
+
+	DWORD returnLength = 0;
+	DWORD byteOffset = 0;
+
+	// Early exit if the needed API is not available at runtime
+	HMODULE k32 = GetModuleHandle(TEXT("Kernel32.dll"));
+	auto fun1 = (fun1_t)(void(*)())GetProcAddress(k32, "GetLogicalProcessorInformationEx");
+	if (!fun1)
+		return false;
+
+	// First call to GetLogicalProcessorInformationEx() to get returnLength.
+	// We expect the call to fail due to null buffer.
+	if (fun1(RelationAll, nullptr, &returnLength))
+		return false;
+
+	// Once we know returnLength, allocate the buffer
+	SYSTEM_LOGICAL_PROCESSOR_INFORMATION_EX *buffer, *ptr;
+	ptr = buffer = (SYSTEM_LOGICAL_PROCESSOR_INFORMATION_EX*)malloc(returnLength);
+
+	// Second call, now we expect to succeed
+	if (!fun1(RelationAll, buffer, &returnLength))
+	{
+		free(buffer);
+		return false;
+	}
+
+	while (byteOffset < returnLength)
+	{
+		// NUMA NODEの数
+		if (ptr->Relationship == RelationNumaNode)
+			(*nodes)++;
+
+		else if (ptr->Relationship == RelationProcessorCore)
+		{
+			// 物理コアの数
+			(*cores)++;
+
+			// 論理コア数の加算。HT対応なら2を足す。HT非対応なら1を足す。
+			*threads += (ptr->Processor.Flags == LTP_PC_SMT) ? 2 : 1;
+		}
+
+		ASSERT_LV3(ptr->Size);
+		byteOffset += ptr->Size;
+		ptr = (SYSTEM_LOGICAL_PROCESSOR_INFORMATION_EX*)(((char*)ptr) + ptr->Size);
+	}
+
+	free(buffer);
+
+	return true;
+}
+
+int total_thread_count()
+{
+	int n, c, t;
+
+	if (get_cpu_info(&n, &c, &t))
+		return t;
+
+	return std::thread::hardware_concurrency();
+}
+
+#elif defined(__linux__)
+
+int total_thread_count() {
+    int threads = 0;
+
+    // /proc/cpuinfo を開く
+    std::ifstream cpuinfo("/proc/cpuinfo");
+
+    if (!cpuinfo) {
+        std::cerr << "Failed to open /proc/cpuinfo" << std::endl;
+        return std::thread::hardware_concurrency();
+    }
+
+    std::string line;
+
+    // 各行を読み込みながらコア数をカウント
+    while (std::getline(cpuinfo, line)) {
+        if (line.find("processor") == 0) {
+            threads++;
+        }
+    }
+
+	return threads;
+}
+
+#else
+
+int total_thread_count()
+{
+	return std::thread::hardware_concurrency();
+}
+
+#endif
+
 // --------------------
 //  統計情報
 // --------------------
@@ -861,53 +966,9 @@ namespace WinProcGroup {
 		// 物理コア数
 		int cores = 0;
 
-		DWORD returnLength = 0;
-		DWORD byteOffset = 0;
-
-		// Early exit if the needed API is not available at runtime
-		HMODULE k32 = GetModuleHandle(TEXT("Kernel32.dll"));
-		auto fun1 = (fun1_t)(void(*)())GetProcAddress(k32, "GetLogicalProcessorInformationEx");
-		if (!fun1)
+		if (!get_cpu_info(&nodes, &cores, &threads))
 			return -1;
 
-		// First call to GetLogicalProcessorInformationEx() to get returnLength.
-		// We expect the call to fail due to null buffer.
-		if (fun1(RelationAll, nullptr, &returnLength))
-			return -1;
-
-		// Once we know returnLength, allocate the buffer
-		SYSTEM_LOGICAL_PROCESSOR_INFORMATION_EX *buffer, *ptr;
-		ptr = buffer = (SYSTEM_LOGICAL_PROCESSOR_INFORMATION_EX*)malloc(returnLength);
-
-		// Second call, now we expect to succeed
-		if (!fun1(RelationAll, buffer, &returnLength))
-		{
-			free(buffer);
-			return -1;
-		}
-
-		while (byteOffset < returnLength)
-		{
-			// NUMA NODEの数
-			if (ptr->Relationship == RelationNumaNode)
-				nodes++;
-
-			else if (ptr->Relationship == RelationProcessorCore)
-			{
-				// 物理コアの数
-				cores++;
-
-				// 論理コア数の加算。HT対応なら2を足す。HT非対応なら1を足す。
-				threads += (ptr->Processor.Flags == LTP_PC_SMT) ? 2 : 1;
-			}
-
-			ASSERT_LV3(ptr->Size);
-			byteOffset += ptr->Size;
-			ptr = (SYSTEM_LOGICAL_PROCESSOR_INFORMATION_EX*)(((char*)ptr) + ptr->Size);
-		}
-
-		free(buffer);
-
 		std::vector<int> groups;
 
 		// Run as many threads as possible on the same node until core limit is

source/misc.h

@@ -43,6 +43,13 @@ void prefetch(const void* addr);
 // cin/coutへの入出力をファイルにリダイレクトを開始/終了する。
 void start_logger(const std::string& fname);
 
+// -----------------
+//    CPU Threads
+// -----------------
+
+// NUMAノード環境にも対応したCPUスレッド数の取得。
+int total_thread_count();
+
 // --------------------
 //  Large Page確保
 // --------------------

source/usi_option.cpp

@@ -226,7 +226,7 @@ namespace USI {
 		// (プロセッサグループは64論理コアごとに1つ作られる。上のケースでは、ThreadIdOffset = 0,0,64,64でも同じ意味。)
 		//	※　1つのPCで複数の思考エンジンを同時に起動して対局させる場合はこれを適切に設定すべき。
 
-		o["ThreadIdOffset"] << Option(0, 0, std::thread::hardware_concurrency() - 1);
+		o["ThreadIdOffset"] << Option(0, 0, total_thread_count() - 1);
 #endif
 
 #if defined(_WIN64)