[svn-r3249] adding back SMP-aware allreduce

src/mpi/coll/allreduce.c

@@ -680,10 +680,68 @@ int MPI_Allreduce ( void *sendbuf, void *recvbuf, int count,
     }
     else
     {
-        if (comm_ptr->comm_kind == MPID_INTRACOMM) 
+        if (comm_ptr->comm_kind == MPID_INTRACOMM) {
             /* intracommunicator */
+#if USE_SMP_COLLECTIVES
+	    MPID_Op *op_ptr;
+	    int is_commutative; 
+
+	    /* is the op commutative? We do SMP optimizations only if it is. */ 
+	    if (HANDLE_GET_KIND(op) == HANDLE_KIND_BUILTIN)
+		is_commutative = 1;
+	    else {
+		MPID_Op_get_ptr(op, op_ptr);
+		is_commutative = (op_ptr->kind == MPID_OP_USER_NONCOMMUTE) ? 0 : 1;
+	    }
+
+            if (MPIR_Comm_is_node_aware(comm_ptr) && is_commutative) {
+		/* on each node, do a reduce to the local root */ 
+                if (comm_ptr->node_comm != NULL)
+                {
+		    /* take care of the MPI_IN_PLACE case. For reduce, 
+		        MPI_IN_PLACE is specified only on the root; 
+                        for allreduce it is specified on all processes. */
+
+		    if ((sendbuf == MPI_IN_PLACE) && (comm_ptr->node_comm->rank != 0)) {
+			/* IN_PLACE and not root of reduce. Data supplied to this
+			allreduce is in recvbuf. Pass that as the sendbuf to reduce. */
+			
+			mpi_errno = MPIR_Reduce(recvbuf, NULL, count, datatype,
+						op, 0, comm_ptr->node_comm);
+		    }
+		    else {
+			mpi_errno = MPIR_Reduce(sendbuf, recvbuf, count, datatype,
+						op, 0, comm_ptr->node_comm);
+		    }
+		    if (mpi_errno) goto fn_fail;
+                }
+
+                /* now do an IN_PLACE allreduce among the local roots of all nodes */
+                if (comm_ptr->node_roots_comm != NULL) {
+                    mpi_errno = MPIR_Allreduce(MPI_IN_PLACE, recvbuf, count, datatype,
+					       op, comm_ptr->node_roots_comm);
+                    if (mpi_errno) goto fn_fail;
+                }
+
+		/* now broadcast the result among local processes */
+                if (comm_ptr->node_comm != NULL) {
+		    MPIU_THREADPRIV_DECL;
+		    MPIU_THREADPRIV_GET;
+		    
+		    MPIR_Nest_incr();
+                    mpi_errno = MPIR_Bcast(recvbuf, count, datatype, 0, comm_ptr->node_comm);
+		    MPIR_Nest_decr();
+		}
+            }
+            else {
+                mpi_errno = MPIR_Allreduce(sendbuf, recvbuf, count, datatype,
+                                       op, comm_ptr);
+            }
+#else
             mpi_errno = MPIR_Allreduce(sendbuf, recvbuf, count, datatype,
                                        op, comm_ptr); 
+#endif
+	}
         else {
             /* intercommunicator */
             mpi_errno = MPIR_Allreduce_inter(sendbuf, recvbuf, count,