[libcxx-commits] [libcxx] [libc++] Fixed get count threads multi-CPU system with NUMA architecture (#72267) (PR #72270)

Louis Dionne via libcxx-commits libcxx-commits at lists.llvm.org
Wed Nov 15 08:33:05 PST 2023


================
@@ -81,9 +81,58 @@ thread::hardware_concurrency() noexcept
         return 0;
     return static_cast<unsigned>(result);
 #elif defined(_LIBCPP_WIN32API)
-    SYSTEM_INFO info;
-    GetSystemInfo(&info);
-    return info.dwNumberOfProcessors;
+    // This implementation supports both conventional single-cpu PC configurations
+    // and multi-cpu system on NUMA (Non-uniform_memory_access) architecture
+    DWORD length = 0;
+    unsigned concurrency = 0;
----------------
ldionne wrote:

You could simplify like this:

```
const auto single_cpu_concurrency = []() noexcept -> unsigned int {
  SYSTEM_INFO info;
  GetSystemInfo(&info);
  return info.dwNumberOfProcessors;
};

if (GetLogicalProcessorInformationEx(RelationAll, nullptr, &length) != FALSE)
  return single_cpu_concurrency();
if (GetLastError() != ERROR_INSUFFICIENT_BUFFER)
  return single_cpu_concurrency();
[...]

DWORD i = 0;
unsigned int concurrency = 0;
while (i < length) {
  [...]
}
return concurrency;
```

In other words, there is no reason to pass `concurrency` by reference to the lambda, it just makes the control flow more complicated to follow.

https://github.com/llvm/llvm-project/pull/72270


More information about the libcxx-commits mailing list