ENH: add Pstream::broadcastList()

- broadcasts list contiguous content as a two-step process:
    1. broadcast the size, and resize for receiver list
    2. broadcast contiguous contents (if non-empty)

  This avoids serialization/de-serialization memory overhead but at
  the expense of an additional broadcast call.
  The trade-off of the extra broadcast of the size will be less
  important than avoiding a memory peak for large contiguous mesh data.

REVERT: unstable MPI_Mprobe/MPI_Mrecv on intelmpi + PMI-2 (#2796)

- partial revert of commit c6f528588b, for NBX implementation.
  Not yet flagged as causing errors here, but eliminated for
  consistency.

applications/test/IndirectList/Test-IndirectList.C

@@ -183,9 +183,7 @@ int main(int argc, char *argv[])
             Pout<<"recv: " << flatOutput(recv) << endl;
         }
 
-        // MPI barrier
-        bool barrier = true;
-        Pstream::broadcast(barrier);
+        UPstream::barrier(UPstream::worldComm);
     }
 
 

applications/test/globalIndex/Test-globalIndex.C

@@ -204,7 +204,7 @@ int main(int argc, char *argv[])
     labelPair inOut;
     pointField allCcs(globalNumbering.gather(mesh.cellCentres()));
     inOut[0] = allCcs.size();
-    Pstream::broadcast(allCcs);
+    Pstream::broadcastList(allCcs);
     inOut[1] = allCcs.size();
     Pout<< "    " << inOut << endl;
 

src/OpenFOAM/db/IOstreams/Pstreams/Pstream.H

@@ -102,9 +102,8 @@ public:
         //- Broadcast buffer content to all processes in communicator.
         using UPstream::broadcast;
 
-        //- Broadcast content (contiguous or non-contiguous)
-        //- to all processes in communicator.
-        //  For \b non-parallel : do nothing.
+        //- Broadcast content (contiguous or non-contiguous) to all
+        //- communicator ranks. Does nothing in \b non-parallel.
         template<class Type>
         static void broadcast
         (
@@ -112,11 +111,22 @@ public:
             const label comm = UPstream::worldComm
         );
 
-        //- Broadcast multiple items to all processes in communicator.
-        //  For \b non-parallel : do nothing.
+        //- Broadcast multiple items to all communicator ranks.
+        //- Does nothing in \b non-parallel.
         template<class Type, class... Args>
         static void broadcasts(const label comm, Type& arg1, Args&&... args);
 
+        //- Broadcast list content (contiguous or non-contiguous) to all
+        //- communicator ranks. Does nothing in \b non-parallel.
+        //  For contiguous list data, this avoids serialization overhead,
+        //  but at the expense of an additional broadcast call.
+        template<class ListType>
+        static void broadcastList
+        (
+            ListType& list,
+            const label comm = UPstream::worldComm
+        );
+
 
     // Gather
 

src/OpenFOAM/db/IOstreams/Pstreams/PstreamBroadcast.C

@@ -80,4 +80,61 @@ void Foam::Pstream::broadcasts(const label comm, Type& arg1, Args&&... args)
 }
 
 
+template<class ListType>
+void Foam::Pstream::broadcastList(ListType& list, const label comm)
+{
+    if (is_contiguous<typename ListType::value_type>::value)
+    {
+        // List data are contiguous
+        // 1. broadcast the size
+        // 2. resize for receiver list
+        // 3. broadcast contiguous contents
+
+        if (UPstream::is_parallel(comm))
+        {
+            label len(list.size());
+
+            UPstream::broadcast
+            (
+                reinterpret_cast<char*>(&len),
+                sizeof(label),
+                comm,
+                UPstream::masterNo()
+            );
+
+            if (UPstream::is_subrank(comm))
+            {
+                list.resize_nocopy(len);
+            }
+
+            if (len)
+            {
+                UPstream::broadcast
+                (
+                    list.data_bytes(),
+                    list.size_bytes(),
+                    comm,
+                    UPstream::masterNo()
+                );
+            }
+        }
+    }
+    else if (UPstream::is_parallel(comm))
+    {
+        // List data are non-contiguous - serialize/de-serialize
+
+        if (UPstream::master(comm))
+        {
+            OPBstream os(UPstream::masterNo(), comm);
+            os << list;
+        }
+        else  // UPstream::is_subrank(comm)
+        {
+            IPBstream is(UPstream::masterNo(), comm);
+            is >> list;
+        }
+    }
+}
+
+
 // ************************************************************************* //

src/Pstream/mpi/UPstreamWrappingTemplates.C

@@ -613,19 +613,6 @@ void Foam::PstreamDetail::allToAllConsensus
         int flag = 0;
         MPI_Status status;
 
-#if defined(MPI_VERSION) && (MPI_VERSION >= 3)
-        // MPI-3 : eg, openmpi-1.7 (2013) and later
-        MPI_Message message;
-        MPI_Improbe
-        (
-            MPI_ANY_SOURCE,
-            tag,
-            PstreamGlobals::MPICommunicators_[comm],
-           &flag,
-           &message,
-           &status
-        );
-#else
         MPI_Iprobe
         (
             MPI_ANY_SOURCE,
@@ -634,7 +621,6 @@ void Foam::PstreamDetail::allToAllConsensus
            &flag,
            &status
         );
-#endif
 
         if (flag)
         {
@@ -654,17 +640,6 @@ void Foam::PstreamDetail::allToAllConsensus
 
             // Regular blocking receive [the data are small]
 
-#if defined(MPI_VERSION) && (MPI_VERSION >= 3)
-            // MPI-3 : eg, openmpi-1.7 (2013) and later
-            MPI_Mrecv
-            (
-               &recvData[proci],
-                count,          // count=1 (see above)
-                datatype,
-               &message,
-                MPI_STATUS_IGNORE
-            );
-#else
             MPI_Recv
             (
                &recvData[proci],
@@ -675,7 +650,6 @@ void Foam::PstreamDetail::allToAllConsensus
                 PstreamGlobals::MPICommunicators_[comm],
                 MPI_STATUS_IGNORE
             );
-#endif
         }
 
         if (barrier_active)