[cluster] Make bench compatible

Fix one TODO.

Takes care of output from bench.
Sum nodes over ranks.

src/cluster.cpp

@@ -266,6 +266,12 @@ void pick_moves(MoveInfo& mi) {
   MPI_Bcast(&mi, 1, MIDatatype, 0, MoveComm);
 }
 
+void sum(uint64_t& val) {
+
+  const uint64_t send = val;
+  MPI_Reduce(&send, &val, 1, MPI_UINT64_T, MPI_SUM, 0, MoveComm);
+}
+
 }
 
 #endif // USE_MPI

src/cluster.h

@@ -74,6 +74,7 @@ int rank();
 inline bool is_root() { return rank() == 0; }
 void save(Thread* thread, TTEntry* tte, Key k, Value v, Bound b, Depth d, Move m, Value ev);
 void pick_moves(MoveInfo& mi);
+void sum(uint64_t& val);
 void sync_start();
 void sync_stop();
 
@@ -94,6 +95,7 @@ inline void save(Thread* thread, TTEntry* tte, Key k, Value v, Bound b, Depth d,
   tte->save(k, v, b, d, m, ev);
 }
 inline void pick_moves(MoveInfo&) { }
+void sum(uint64_t& val) { }
 inline void sync_start() { }
 inline void sync_stop() { }
 

src/uci.cpp

@@ -141,8 +141,6 @@ namespace {
   // a list of UCI commands is setup according to bench parameters, then
   // it is run one by one printing a summary at the end.
 
-  // TODO make (output?) cluster compatible
-
   void bench(Position& pos, istream& args, StateListPtr& states) {
 
     string token;
@@ -160,7 +158,8 @@ namespace {
 
         if (token == "go")
         {
-            cerr << "\nPosition: " << cnt++ << '/' << num << endl;
+            if (Cluster::is_root())
+                cerr << "\nPosition: " << cnt++ << '/' << num << endl;
             go(pos, is, states);
             Threads.main()->wait_for_search_finished();
             nodes += Threads.nodes_searched();
@@ -174,10 +173,12 @@ namespace {
 
     dbg_print(); // Just before exiting
 
-    cerr << "\n==========================="
-         << "\nTotal time (ms) : " << elapsed
-         << "\nNodes searched  : " << nodes
-         << "\nNodes/second    : " << 1000 * nodes / elapsed << endl;
+    Cluster::sum(nodes);
+    if (Cluster::is_root())
+        cerr << "\n==========================="
+             << "\nTotal time (ms) : " << elapsed
+             << "\nNodes searched  : " << nodes
+             << "\nNodes/second    : " << 1000 * nodes / elapsed << endl;
   }
 
 } // namespace