Skip to content

WA for 6740e on Windows Server #32304

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 4 commits into
base: master
Choose a base branch
from
+1,297 −45
Open

WA for 6740e on Windows Server #32304

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
07cd3d8
WA for 6740e on Windows Server
wangleis Oct 6, 2025
9d2ffa4
update for test case
wangleis Oct 7, 2025
01a442a
support multi numa nodes on single socket in win11
wangleis Oct 7, 2025
5ae64cf
update for test case
wangleis Oct 8, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
127 changes: 82 additions & 45 deletions src/inference/src/os/win/win_system_conf.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -68,28 +68,33 @@ void parse_processor_info_win(const char* base_ptr,
std::vector<int> proc_info;
std::unordered_set<int> l3_set;

std::vector<int> proc_init_line({0, 0, 0, 0, 0, -1, -1});
std::vector<int> proc_init_line({0, 0, 0, 0, 0, 0, 0});
std::vector<int> cpu_init_line(CPU_MAP_TABLE_SIZE, -1);

constexpr int initial_core_type = -1;
constexpr int initial_numa_mask = -1;
constexpr int group_with_1_core = 1;
constexpr int group_with_2_cores = 2;
constexpr int group_with_4_cores = 4;

char* info_ptr = (char*)base_ptr;
int list_len = 0;
int base_proc = 0;
int base_proc_socket = 0;
int group = 0;
int proc_group = 0;

int group_start = 0;
int group_end = 0;
int group_id = 0;
int group_type = initial_core_type;

int num_package = 0;
int cur_numa_mask = initial_numa_mask;

_processors = 0;
_sockets = 0;
_numa_nodes = 0;
_cores = 0;
_blocked_cores = 0;

Expand All @@ -112,6 +117,27 @@ void parse_processor_info_win(const char* base_ptr,
return;
};

auto create_new_proc_line = [&]() {
_proc_type_table[0][PROC_NUMA_NODE_ID] = _numa_nodes;
_proc_type_table[0][PROC_SOCKET_ID] = _sockets;
_proc_type_table.push_back(_proc_type_table[0]);
_proc_type_table[0] = proc_init_line;
return;
};

auto check_numa_node = [&]() {
if (l3_set.size() < 64) {
if (cur_numa_mask == initial_numa_mask) {
cur_numa_mask = info->Processor.GroupMask->Group;
} else if (cur_numa_mask != info->Processor.GroupMask->Group) {
create_new_proc_line();
_numa_nodes++;
cur_numa_mask = info->Processor.GroupMask->Group;
}
}
return;
};

_proc_type_table.push_back(proc_init_line);

for (; info_ptr < base_ptr + len; info_ptr += (DWORD)info->Size) {
Expand All @@ -121,31 +147,39 @@ void parse_processor_info_win(const char* base_ptr,
MaskToList(info->Processor.GroupMask->Mask);
if (num_package > 0) {
_sockets++;
l3_set.clear();
_numa_nodes++;
cur_numa_mask = initial_numa_mask;
if (_processors < 64) {
l3_set.clear();
} else {
base_proc_socket = _processors;
}
_proc_type_table.push_back(_proc_type_table[0]);
_proc_type_table[0] = proc_init_line;
}
num_package++;
} else if (info->Relationship == RelationProcessorCore) {
MaskToList(info->Processor.GroupMask->Mask);

if (0 == list[0]) {
base_proc = _processors;
if (0 == list[0] || proc_group != info->Processor.GroupMask->Group) {
base_proc = _processors - base_proc_socket - list[0];
proc_group = info->Processor.GroupMask->Group;
}

if (group_with_2_cores == list_len) {
proc_info = cpu_init_line;
proc_info[CPU_MAP_PROCESSOR_ID] = list[0] + base_proc;
proc_info[CPU_MAP_NUMA_NODE_ID] = _sockets;
proc_info[CPU_MAP_PROCESSOR_ID] = list[0] + base_proc_socket + base_proc;
check_numa_node();
proc_info[CPU_MAP_NUMA_NODE_ID] = _numa_nodes;
proc_info[CPU_MAP_SOCKET_ID] = _sockets;
proc_info[CPU_MAP_CORE_ID] = _cores;
proc_info[CPU_MAP_CORE_TYPE] = HYPER_THREADING_PROC;
proc_info[CPU_MAP_GROUP_ID] = group;
_cpu_mapping_table.push_back(proc_info);

proc_info = cpu_init_line;
proc_info[CPU_MAP_PROCESSOR_ID] = list[1] + base_proc;
proc_info[CPU_MAP_NUMA_NODE_ID] = _sockets;
proc_info[CPU_MAP_PROCESSOR_ID] = list[1] + base_proc_socket + base_proc;
proc_info[CPU_MAP_NUMA_NODE_ID] = _numa_nodes;
proc_info[CPU_MAP_SOCKET_ID] = _sockets;
proc_info[CPU_MAP_CORE_ID] = _cores;
proc_info[CPU_MAP_CORE_TYPE] = MAIN_CORE_PROC;
Expand All @@ -158,8 +192,9 @@ void parse_processor_info_win(const char* base_ptr,

} else {
proc_info = cpu_init_line;
proc_info[CPU_MAP_PROCESSOR_ID] = list[0] + base_proc;
proc_info[CPU_MAP_NUMA_NODE_ID] = _sockets;
proc_info[CPU_MAP_PROCESSOR_ID] = list[0] + base_proc_socket + base_proc;
check_numa_node();
proc_info[CPU_MAP_NUMA_NODE_ID] = _numa_nodes;
proc_info[CPU_MAP_SOCKET_ID] = _sockets;
proc_info[CPU_MAP_CORE_ID] = _cores;
if ((_processors > group_start) && (_processors <= group_end)) {
Expand All @@ -185,51 +220,48 @@ void parse_processor_info_win(const char* base_ptr,
MaskToList(info->Cache.GroupMask.Mask);

if (list_len == group_with_1_core) {
if (_processors <= list[list_len - 1] + base_proc) {
group_start = list[0];
group_end = list[list_len - 1];
group_id = group;
group_type = MAIN_CORE_PROC;
if (_cpu_mapping_table.size() > list[0] + base_proc_socket + base_proc &&
_cpu_mapping_table[list[0] + base_proc_socket + base_proc][CPU_MAP_CORE_TYPE] ==
initial_core_type) {
_cpu_mapping_table[list[0] + base_proc_socket + base_proc][CPU_MAP_CORE_TYPE] = MAIN_CORE_PROC;
_cpu_mapping_table[list[0] + base_proc_socket + base_proc][CPU_MAP_GROUP_ID] = group++;
_proc_type_table[0][MAIN_CORE_PROC]++;
}
for (int m = 0; m < _processors - list[0]; m++) {
if (_cpu_mapping_table[list[m] + base_proc][CPU_MAP_CORE_TYPE] == initial_core_type) {
_cpu_mapping_table[list[m] + base_proc][CPU_MAP_CORE_TYPE] = MAIN_CORE_PROC;
_cpu_mapping_table[list[m] + base_proc][CPU_MAP_GROUP_ID] = group++;
_proc_type_table[0][MAIN_CORE_PROC]++;
}
}
} else if (initial_core_type == _cpu_mapping_table[list[0] + base_proc][CPU_MAP_CORE_TYPE]) {
if (l3_set.size() == 0 || l3_set.count(list[0])) {
if (_processors <= list[list_len - 1] + base_proc) {
group_start = list[0];
group_end = list[list_len - 1];
} else if (initial_core_type ==
_cpu_mapping_table[list[0] + base_proc_socket + base_proc][CPU_MAP_CORE_TYPE]) {
if (l3_set.size() == 0 || l3_set.count(list[0]) ||
list[0] + base_proc_socket + base_proc > l3_set.size()) {
if (_processors <= list[list_len - 1] + base_proc_socket + base_proc) {
group_start = list[0] + base_proc_socket + base_proc;
group_end = list[list_len - 1] + base_proc_socket + base_proc;
group_type = (group_with_4_cores == list_len) ? EFFICIENT_CORE_PROC : MAIN_CORE_PROC;
} else {
group_type = (group_type == initial_core_type) ? MAIN_CORE_PROC : group_type;
}
group_id = group++;
for (int m = 0; m < _processors - list[0]; m++) {
if (_cpu_mapping_table[list[m] + base_proc][CPU_MAP_CORE_TYPE] == initial_core_type) {
_cpu_mapping_table[list[m] + base_proc][CPU_MAP_CORE_TYPE] = group_type;
_cpu_mapping_table[list[m] + base_proc][CPU_MAP_GROUP_ID] = group_id;
for (int m = 0; m < _processors - base_proc_socket - base_proc - list[0]; m++) {
if (_cpu_mapping_table[list[m] + base_proc_socket + base_proc][CPU_MAP_CORE_TYPE] ==
initial_core_type) {
_cpu_mapping_table[list[m] + base_proc_socket + base_proc][CPU_MAP_CORE_TYPE] = group_type;
_cpu_mapping_table[list[m] + base_proc_socket + base_proc][CPU_MAP_GROUP_ID] = group_id;
_proc_type_table[0][group_type]++;
}
}
} else {
if (_processors <= list[list_len - 1] + base_proc) {
if (_processors <= list[list_len - 1] + base_proc_socket + base_proc) {
group_start = list[0];
group_end = list[list_len - 1];
group_id = (group_with_2_cores == list_len) ? CPU_BLOCKED : group++;
group_type = LP_EFFICIENT_CORE_PROC;
}
for (int m = 0; m < _processors - list[0]; m++) {
_cpu_mapping_table[list[m] + base_proc][CPU_MAP_CORE_TYPE] = group_type;
_cpu_mapping_table[list[m] + base_proc][CPU_MAP_GROUP_ID] = group_id;
for (int m = 0; m < _processors - base_proc_socket - base_proc - list[0]; m++) {
_cpu_mapping_table[list[m] + base_proc_socket + base_proc][CPU_MAP_CORE_TYPE] = group_type;
_cpu_mapping_table[list[m] + base_proc_socket + base_proc][CPU_MAP_GROUP_ID] = group_id;
if (group_id == CPU_BLOCKED) {
_cpu_mapping_table[list[m] + base_proc][CPU_MAP_USED_FLAG] = CPU_BLOCKED;
_cpu_mapping_table[list[m] + base_proc_socket + base_proc][CPU_MAP_USED_FLAG] = CPU_BLOCKED;
_blocked_cores++;
} else {
_cpu_mapping_table[list[m] + base_proc][CPU_MAP_USED_FLAG] = NOT_USED;
_cpu_mapping_table[list[m] + base_proc_socket + base_proc][CPU_MAP_USED_FLAG] = NOT_USED;
_proc_type_table[0][LP_EFFICIENT_CORE_PROC]++;
}
}
Expand All @@ -240,26 +272,31 @@ void parse_processor_info_win(const char* base_ptr,
l3_set.insert(list.begin(), list.end());
}
}
_sockets++;
_processors -= _blocked_cores;
_cores -= _blocked_cores;
_proc_type_table[0][ALL_PROC] -= _blocked_cores;
if (_sockets > 1) {
_proc_type_table.push_back(_proc_type_table[0]);
_proc_type_table[0] = proc_init_line;
if (_proc_type_table.size() > 1) {
create_new_proc_line();

for (int m = 1; m <= _sockets; m++) {
for (int m = 1; m <= _proc_type_table.size() - 1; m++) {
for (int n = 0; n <= HYPER_THREADING_PROC; n++) {
_proc_type_table[0][n] += _proc_type_table[m][n];
}
_proc_type_table[m][PROC_SOCKET_ID] = m - 1;
_proc_type_table[m][PROC_NUMA_NODE_ID] = m - 1;
_proc_type_table[0][PROC_SOCKET_ID] =
_proc_type_table[0][PROC_SOCKET_ID] == _proc_type_table[m][PROC_SOCKET_ID]
? _proc_type_table[m][PROC_SOCKET_ID]
: -1;
_proc_type_table[0][PROC_NUMA_NODE_ID] =
_proc_type_table[0][PROC_NUMA_NODE_ID] == _proc_type_table[m][PROC_NUMA_NODE_ID]
? _proc_type_table[m][PROC_NUMA_NODE_ID]
: -1;
}
} else {
_proc_type_table[0][PROC_SOCKET_ID] = 0;
_proc_type_table[0][PROC_NUMA_NODE_ID] = 0;
}
_numa_nodes = _sockets;
_sockets++;
_numa_nodes++;
}

int get_number_of_cpu_cores(bool bigCoresOnly) {
Expand Down
Loading
Loading