Fix processor group binding under Windows.

Starting with Windows Build 20348 the behavior of the numa API has been changed:
https://docs.microsoft.com/en-us/windows/win32/procthread/numa-support

Old code only worked because there was probably a limit on how many
cores/threads can reside within one NUMA node, and the OS creates extra NUMA
nodes when necessary, however the actual mechanism of core binding is
done by "Processor Groups"(https://docs.microsoft.com/en-us/windows/win32/procthread/processor-groups). With a newer OS, one NUMA node can have many
such "Processor Groups" and we should just consistently use the number
of groups to bind the threads instead of deriving the topology from
the number of NUMA nodes.

This change is required to spread threads on all cores on Windows 11 with
a 3990X CPU. It has only 1 NUMA node with 2 groups of 64 threads each.

closes https://github.com/official-stockfish/Stockfish/pull/3787

No functional change.

diff --git a/src/misc.cpp b/src/misc.cpp
index f9c123374f25162dc23e1af70dc7a1fe4076a5af..0af20e10d58f2b014f050ddcec8c23bd29cfc763 100644 (file)
--- a/src/misc.cpp
+++ b/src/misc.cpp
@@ -502,7 +502,7 @@ void bindThisThread(size_t) {}
 int best_group(size_t idx) {
 
   int threads = 0;
 int best_group(size_t idx) {
 
   int threads = 0;

-  int nodes = 0;
+  int groups = 0;

   int cores = 0;
   DWORD returnLength = 0;
   DWORD byteOffset = 0;
   int cores = 0;
   DWORD returnLength = 0;
   DWORD byteOffset = 0;


@@ -530,8 +530,8 @@ int best_group(size_t idx) {
 
   while (byteOffset < returnLength)
   {
 
   while (byteOffset < returnLength)
   {

-      if (ptr->Relationship == RelationNumaNode)
-          nodes++;
+      if (ptr->Relationship == RelationGroup)
+          groups += ptr->Group.MaximumGroupCount;

 
       else if (ptr->Relationship == RelationProcessorCore)
       {
 
       else if (ptr->Relationship == RelationProcessorCore)
       {


@@ -546,23 +546,23 @@ int best_group(size_t idx) {
 
   free(buffer);
 
 
   free(buffer);
 

-  std::vector<int> groups;
+  std::vector<int> core_groups;

 
 

-  // Run as many threads as possible on the same node until core limit is
-  // reached, then move on filling the next node.
-  for (int n = 0; n < nodes; n++)
-      for (int i = 0; i < cores / nodes; i++)
-          groups.push_back(n);
+  // Run as many threads as possible on the same group until core limit is
+  // reached, then move on filling the next group.
+  for (int n = 0; n < groups; n++)
+      for (int i = 0; i < cores / groups; i++)
+          core_groups.push_back(n);

 
   // In case a core has more than one logical processor (we assume 2) and we
   // have still threads to allocate, then spread them evenly across available
 
   // In case a core has more than one logical processor (we assume 2) and we
   // have still threads to allocate, then spread them evenly across available

-  // nodes.
+  // groups.

   for (int t = 0; t < threads - cores; t++)
   for (int t = 0; t < threads - cores; t++)

-      groups.push_back(t % nodes);
+      core_groups.push_back(t % groups);

 
   // If we still have more threads than the total number of logical processors
   // then return -1 and let the OS to decide what to do.
 
   // If we still have more threads than the total number of logical processors
   // then return -1 and let the OS to decide what to do.

-  return idx < groups.size() ? groups[idx] : -1;
+  return idx < core_groups.size() ? core_groups[idx] : -1;

 }
 
 
 }