Modifying type of binomial tree used for binomial reduce:

switching:
       0                         0
     / \ \                     / \ \
	1    \ \         -->       4   \ \
  /      \ \                 /     \ \
 3       2  \               3       2 \
             4                         1
(duh).  The first form is the bmtree suitable for bcast, but the latter is better for reduce.
Updating default decision function accordingly.

This commit was SVN r15422.

ompi/mca/coll/tuned/coll_tuned_decision_fixed.c

@@ -351,19 +351,15 @@ int ompi_coll_tuned_reduce_intra_dec_fixed( void *sendbuf, void *recvbuf,
     if ((communicator_size < 8) && (message_size < 512)){
         /* Linear_0K */
         return ompi_coll_tuned_reduce_intra_basic_linear (sendbuf, recvbuf, count, datatype, op, root, comm); 
-    } else if ((communicator_size < 8) && (message_size < 20480)) {
+    } else if (((communicator_size < 8) && (message_size < 20480)) ||
+               (message_size < 2048) || (count <= 1)) {
         /* Binomial_0K */
         segsize = 0;
         return ompi_coll_tuned_reduce_intra_binomial(sendbuf, recvbuf, count, datatype, op, root, comm, segsize, max_requests);
-    } else if ((message_size < 2048) ||
-               (count <= 1)){
-        /* Binary_0K */
-        segsize = 0;
-        return ompi_coll_tuned_reduce_intra_binary(sendbuf, recvbuf, count, datatype, op, root, comm, segsize, max_requests);
     } else if (communicator_size > (a1 * message_size + b1)) {
-        /* Binary_1K */
+        /* Binomial_1K */
         segsize = 1024;
-        return ompi_coll_tuned_reduce_intra_binary(sendbuf, recvbuf, count, datatype, op, root, comm, segsize, max_requests);
+        return ompi_coll_tuned_reduce_intra_binomial(sendbuf, recvbuf, count, datatype, op, root, comm, segsize, max_requests);
     } else if (communicator_size > (a2 * message_size + b2)) {
         /* Pipeline_1K */
         segsize = 1024;

ompi/mca/coll/tuned/coll_tuned_reduce.c

@@ -437,7 +437,7 @@ int ompi_coll_tuned_reduce_intra_binomial( void *sendbuf, void *recvbuf,
     OPAL_OUTPUT((ompi_coll_tuned_stream,"coll:tuned:reduce_intra_binomial rank %d ss %5d",
                  ompi_comm_rank(comm), segsize));
 
-    COLL_TUNED_UPDATE_BMTREE( comm, root );
+    COLL_TUNED_UPDATE_IN_ORDER_BMTREE( comm, root );
 
     /**
      * Determine number of segments and number of elements
@@ -448,7 +448,7 @@ int ompi_coll_tuned_reduce_intra_binomial( void *sendbuf, void *recvbuf,
 
     return ompi_coll_tuned_reduce_generic( sendbuf, recvbuf, count, datatype, 
                                            op, root, comm,
-                                           comm->c_coll_selected_data->cached_bmtree, 
+                                           comm->c_coll_selected_data->cached_in_order_bmtree, 
                                            segcount, max_outstanding_reqs );
 }