move countBits to impl::popcnt

test/cpumask_test.cpp

diff --git a/test/cpumask_test.cpp b/test/cpumask_test.cpp
index c651066..95ec88f 100644
--- a/test/cpumask_test.cpp
+++ b/test/cpumask_test.cpp
@@ -5,16 +5,6 @@
 
 using namespace Xbyak::util;
 
-uint32_t popcnt(uint32_t v)
-{
-	uint32_t n = 0;
-	while (v) {
-		if (v & 1) n++;
-		v >>= 1;
-	}
-	return n;
-}
-
 CYBOZU_TEST_AUTO(pattern)
 {
 	const uint32_t bitN = XBYAK_CPUMASK_BITN;

xbyak/xbyak_util.h

diff --git a/xbyak/xbyak_util.h b/xbyak/xbyak_util.h
index 976e603..1072c2a 100644
--- a/xbyak/xbyak_util.h
+++ b/xbyak/xbyak_util.h
@@ -1124,31 +1124,33 @@
 	friend void impl::initCpuTopology(CpuTopology&, const Cpu&);
 	std::vector<LogicalCpu> logicalCpus_;
 	size_t physicalCoreNum_;
-	size_t socketNum_;
 	uint32_t lineSize_;
 	bool isHybrid_;
 };
 
 namespace impl {
+
+inline uint32_t popcnt(uint64_t mask)
+{
+#if defined(_M_X64) || defined(_M_AMD64)
+	return (int)__popcnt64(mask);
+#elif defined(__GNUC__) || defined(__clang__)
+	return __builtin_popcountll(mask);
+#else
+	uint32_t count = 0;
+	while (mask) {
+		count += (mask & 1);
+		mask >>= 1;
+	}
+	return count;
+#endif
+}
+
 #ifdef _WIN32
 inline void initCpuTopology(CpuTopology& cpuTopo, const Cpu& cpu)
 {
 	typedef SYSTEM_LOGICAL_PROCESSOR_INFORMATION_EX processorInfo;
 
-	// Helper lambda to count set bits in a mask
-	auto countBits = [](KAFFINITY mask) -> int {
-#if defined(_M_X64) || defined(_M_AMD64)
-		return (int)__popcnt64(mask);
-#else
-		int count = 0;
-		while (mask) {
-			count += (mask & 1);
-			mask >>= 1;
-		}
-		return count;
-#endif
-	};
-
 	// First pass: Get processor core information
 	DWORD len = 0;
 	GetLogicalProcessorInformationEx(RelationProcessorCore, nullptr, &len);