Revert and fix earlier windows NUMA patch

[stockfish] / src / misc.cpp

diff --git a/src/misc.cpp b/src/misc.cpp
index 918dc7a953484924603886991a11bf5bd160be45..4dfa9f0ce2f4bf6572f825e0741db602da4d3d55 100644 (file)
--- a/src/misc.cpp
+++ b/src/misc.cpp
@@ -36,6 +36,7 @@ typedef bool(*fun1_t)(LOGICAL_PROCESSOR_RELATIONSHIP,
                       PSYSTEM_LOGICAL_PROCESSOR_INFORMATION_EX, PDWORD);
 typedef bool(*fun2_t)(USHORT, PGROUP_AFFINITY);
 typedef bool(*fun3_t)(HANDLE, CONST GROUP_AFFINITY*, PGROUP_AFFINITY);
+typedef bool(*fun4_t)(USHORT, PGROUP_AFFINITY, USHORT, PUSHORT);
 }
 #endif
 
@@ -51,7 +52,7 @@ typedef bool(*fun3_t)(HANDLE, CONST GROUP_AFFINITY*, PGROUP_AFFINITY);
 #include <sys/mman.h>
 #endif
 
-#if defined(__APPLE__) || defined(__ANDROID__) || defined(__OpenBSD__) || (defined(__GLIBCXX__) && !defined(_GLIBCXX_HAVE_ALIGNED_ALLOC) && !defined(_WIN32))
+#if defined(__APPLE__) || defined(__ANDROID__) || defined(__OpenBSD__) || (defined(__GLIBCXX__) && !defined(_GLIBCXX_HAVE_ALIGNED_ALLOC) && !defined(_WIN32)) || defined(__e2k__)
 #define POSIXALIGNEDALLOC
 #include <stdlib.h>
 #endif
@@ -110,7 +111,14 @@ public:
 
     static Logger l;
 
-    if (!fname.empty() && !l.file.is_open())
+    if (l.file.is_open())
+    {
+        cout.rdbuf(l.out.buf);
+        cin.rdbuf(l.in.buf);
+        l.file.close();
+    }
+
+    if (!fname.empty())
     {
         l.file.open(fname, ifstream::out);
 
@@ -123,12 +131,6 @@ public:
         cin.rdbuf(&l.in);
         cout.rdbuf(&l.out);
     }
-    else if (fname.empty() && l.file.is_open())
-    {
-        cout.rdbuf(l.out.buf);
-        cin.rdbuf(l.in.buf);
-        l.file.close();
-    }
   }
 };
 
@@ -192,6 +194,18 @@ std::string compiler_info() {
      compiler += "(version ";
      compiler += stringify(_MSC_FULL_VER) "." stringify(_MSC_BUILD);
      compiler += ")";
+  #elif defined(__e2k__) && defined(__LCC__)
+    #define dot_ver2(n) \
+      compiler += (char)'.'; \
+      compiler += (char)('0' + (n) / 10); \
+      compiler += (char)('0' + (n) % 10);
+
+     compiler += "MCST LCC ";
+     compiler += "(version ";
+     compiler += std::to_string(__LCC__ / 100);
+     dot_ver2(__LCC__ % 100)
+     dot_ver2(__LCC_MINOR__)
+     compiler += ")";
   #elif __GNUC__
      compiler += "g++ (GNUC) ";
      compiler += make_version_string(__GNUC__, __GNUC_MINOR__, __GNUC_PATCHLEVEL__);
@@ -366,6 +380,7 @@ void std_aligned_free(void* ptr) {
 static void* aligned_large_pages_alloc_windows(size_t allocSize) {
 
   #if !defined(_WIN64)
+    (void)allocSize; // suppress unused-parameter compiler warning
     return nullptr;
   #else
 
@@ -481,11 +496,11 @@ void bindThisThread(size_t) {}
 
 #else
 
-/// best_group() retrieves logical processor information using Windows specific
-/// API and returns the best group id for the thread with index idx. Original
+/// best_node() retrieves logical processor information using Windows specific
+/// API and returns the best node id for the thread with index idx. Original
 /// code from Texel by Peter Österlund.
 
-int best_group(size_t idx) {
+int best_node(size_t idx) {
 
   int threads = 0;
   int nodes = 0;
@@ -557,22 +572,35 @@ int best_group(size_t idx) {
 void bindThisThread(size_t idx) {
 
   // Use only local variables to be thread-safe
-  int group = best_group(idx);
+  int node = best_node(idx);
 
-  if (group == -1)
+  if (node == -1)
       return;
 
   // Early exit if the needed API are not available at runtime
   HMODULE k32 = GetModuleHandle("Kernel32.dll");
   auto fun2 = (fun2_t)(void(*)())GetProcAddress(k32, "GetNumaNodeProcessorMaskEx");
   auto fun3 = (fun3_t)(void(*)())GetProcAddress(k32, "SetThreadGroupAffinity");
+  auto fun4 = (fun4_t)(void(*)())GetProcAddress(k32, "GetNumaNodeProcessorMask2");
 
   if (!fun2 || !fun3)
       return;
 
-  GROUP_AFFINITY affinity;
-  if (fun2(group, &affinity))
-      fun3(GetCurrentThread(), &affinity, nullptr);
+  if (!fun4) {
+      GROUP_AFFINITY affinity;
+      if (fun2(node, &affinity))
+          fun3(GetCurrentThread(), &affinity, nullptr);
+  } else {
+      // If a numa node has more than one processor group, we assume they are
+      // sized equal and we spread threads evenly across the groups.
+      USHORT elements, returnedElements;
+      elements = GetMaximumProcessorGroupCount();
+      GROUP_AFFINITY *affinity = (GROUP_AFFINITY*)malloc(
+          elements * sizeof(GROUP_AFFINITY));
+      if (fun4(node, affinity, elements, &returnedElements))
+          fun3(GetCurrentThread(), &affinity[idx % returnedElements], nullptr);
+      free(affinity);
+  }
 }
 
 #endif