tt_result_t __thread_bind_numa(IN HANDLE hThread, IN USHORT Node)
{
GROUP_AFFINITY ProcessorMask;
// how to handle the arch in which a numa node includes more than
// 64 cpus??
if (!GetNumaNodeProcessorMaskEx(Node, &ProcessorMask)) {
TT_ERROR("fail to get numa node processor mask");
return TT_FAIL;
}
if (!SetThreadGroupAffinity(hThread, &ProcessorMask, NULL)) {
TT_ERROR("fail to bind thread to numa node[%d]", Node);
return TT_FAIL;
}
return TT_SUCCESS;
}
/* static */
void ThreadPool::setThreadNodeAffinity(int numaNode)
{
#if defined(_WIN32_WINNT) && _WIN32_WINNT >= _WIN32_WINNT_WIN7
GROUP_AFFINITY groupAffinity;
if (GetNumaNodeProcessorMaskEx((USHORT)numaNode, &groupAffinity))
{
if (SetThreadAffinityMask(GetCurrentThread(), (DWORD_PTR)groupAffinity.Mask))
return;
}
x265_log(NULL, X265_LOG_ERROR, "unable to set thread affinity to NUMA node %d\n", numaNode);
#elif HAVE_LIBNUMA
if (numa_available() >= 0)
{
numa_run_on_node(numaNode);
numa_set_preferred(numaNode);
numa_set_localalloc();
return;
}
x265_log(NULL, X265_LOG_ERROR, "unable to set thread affinity to NUMA node %d\n", numaNode);
#else
(void)numaNode;
#endif
}
