mask_type init_thread_affinity_mask(
std::size_t num_thread
, bool numa_sensitive
)
{ // {{{
std::size_t num_of_cores = hardware_concurrency();
std::size_t affinity = num_thread % num_of_cores;
ULONG numa_nodes = 1;
if (GetNumaHighestNodeNumber(&numa_nodes))
++numa_nodes;
std::size_t num_of_cores_per_numa_node = num_of_cores / numa_nodes;
ULONGLONG node_affinity_mask = 0;
ULONGLONG mask = 0x01LL;
if (numa_sensitive) {
UCHAR numa_node = UCHAR(affinity % numa_nodes);
if (!GetNumaNodeProcessorMask(numa_node, &node_affinity_mask))
{
HPX_THROW_EXCEPTION(kernel_error
, "hpx::threads::windows_topology::init_thread_affinity_mask"
, boost::str(boost::format(
"failed to initialize thread %1% affinity mask")
% num_thread));
}
mask = least_significant_bit(node_affinity_mask) <<
(affinity / numa_nodes);
}
else {
UCHAR numa_node = UCHAR(get_numa_node_number(num_thread));
if (!GetNumaNodeProcessorMask(numa_node, &node_affinity_mask))
{
HPX_THROW_EXCEPTION(kernel_error
, "hpx::threads::windows_topology::init_thread_affinity_mask"
, boost::str(boost::format(
"failed to initialize thread %1% affinity mask")
% num_thread));
}
mask = least_significant_bit(node_affinity_mask) <<
(affinity % num_of_cores_per_numa_node);
}
while (!(mask & node_affinity_mask)) {
mask <<= 1LL;
if (0 == mask)
mask = 0x01LL;
}
return static_cast<mask_type>(mask);
} // }}}
mask_type get_thread_affinity_mask_from_lva(
naming::address::address_type lva
, error_code& ec = throws
) const
{ // {{{
PSAPI_WORKING_SET_EX_INFORMATION info;
info.VirtualAddress = reinterpret_cast<void*>(lva);
if (!QueryWorkingSetEx(GetCurrentProcess(), &info, sizeof(info)))
{
HPX_THROWS_IF(ec, kernel_error
, "hpx::threads::windows_topology::get_thread_affinity_mask_from_lva"
, "failed to get thread affinity mask");
}
if (ec)
return 0;
ULONGLONG mask = 0;
if (!GetNumaNodeProcessorMask(info.VirtualAttributes.Node, &mask))
{
HPX_THROWS_IF(ec, kernel_error
, "hpx::threads::windows_topology::get_thread_affinity_mask_from_lva"
, "failed to get thread affinity mask");
}
if (ec)
return 0;
else if (&ec != &throws)
ec = make_success_code();
return static_cast<std::size_t>(mask);
} // }}}
uint64_t osNumaMask(void)
{
uint64_t numaMask;
if (!GetNumaNodeProcessorMask(0, &numaMask))
{
return 1;
}
return numaMask;
}
mask_type init_numa_node_affinity_mask(
std::size_t num_thread
, bool numa_sensitive
)
{ // {{{
std::size_t num_of_cores = hardware_concurrency();
UCHAR affinity = UCHAR(num_thread % num_of_cores);
ULONG numa_nodes = 1;
if (GetNumaHighestNodeNumber(&numa_nodes))
++numa_nodes;
ULONGLONG mask = 0;
if (numa_sensitive) {
UCHAR numa_node = affinity % numa_nodes;
if (!GetNumaNodeProcessorMask(numa_node, &mask))
{
HPX_THROW_EXCEPTION(kernel_error
, "hpx::threads::windows_topology::init_numa_node_affinity_mask"
, boost::str(boost::format(
"failed to initialize NUMA node affinity mask for "
"thread %1%")
% num_thread));
}
return static_cast<mask_type>(mask);
}
UCHAR numa_node = UCHAR(get_numa_node_number(num_thread));
if (!GetNumaNodeProcessorMask(numa_node, &mask))
{
HPX_THROW_EXCEPTION(kernel_error
, "hpx::threads::windows_topology::init_numa_node_affinity_mask"
, boost::str(boost::format(
"failed to initialize NUMA node affinity mask for "
"thread %1%")
% num_thread));
}
return static_cast<mask_type>(mask);
} // }}}
