Use the macro to compute the number of elements in a segment in both

bcast and reduce and update the default values for the variables
as required by the comment in the coll_tuned.h file.

This commit was SVN r12546.

ompi/mca/coll/tuned/coll_tuned_bcast.c

@@ -267,7 +267,7 @@ ompi_coll_tuned_bcast_intra_pipeline( void* buffer,
                                       struct ompi_communicator_t* comm, 
                                       uint32_t segsize )
 {
-    int segcount;
+    int segcount = count;
     size_t typelng;
 
     COLL_TUNED_UPDATE_PIPELINE( comm, root );
@@ -293,7 +293,7 @@ ompi_coll_tuned_bcast_intra_chain( void* buffer,
                                    struct ompi_communicator_t* comm, 
                                    uint32_t segsize, int32_t chains )
 {
-    int segcount;
+    int segcount = count;
     size_t typelng;
 
     COLL_TUNED_UPDATE_CHAIN( comm, root, chains );
@@ -319,7 +319,7 @@ ompi_coll_tuned_bcast_intra_binomial( void* buffer,
                                       struct ompi_communicator_t* comm, 
                                       uint32_t segsize )
 {
-    int segcount;
+    int segcount = count;
     size_t typelng;
 
     COLL_TUNED_UPDATE_BMTREE( comm, root );

ompi/mca/coll/tuned/coll_tuned_reduce.c

@@ -213,7 +213,7 @@ int ompi_coll_tuned_reduce_intra_chain( void *sendbuf, void *recvbuf, int count,
                                         int root, ompi_communicator_t* comm, uint32_t segsize,
                                         int fanout)
 {
-    int segcount;
+    int segcount = count;
     size_t typelng;
 
     OPAL_OUTPUT((ompi_coll_tuned_stream,"coll:tuned:reduce_intra_chain rank %d fo %d ss %5d", ompi_comm_rank(comm), fanout, segsize));
@@ -224,11 +224,7 @@ int ompi_coll_tuned_reduce_intra_chain( void *sendbuf, void *recvbuf, int count,
      * sent per operation
      */
     ompi_ddt_type_size( datatype, &typelng );
-    if( segsize > typelng ) {
+    COLL_TUNED_COMPUTED_SEGCOUNT( segsize, typelng, segcount );
-        segcount     = (int)(segsize / typelng);
-    } else  {
-        segcount     = count;
-    }
 
     return ompi_coll_tuned_reduce_generic( sendbuf, recvbuf, count, datatype, op, root, comm,
                                            comm->c_coll_selected_data->cached_chain, segcount );
@@ -240,7 +236,7 @@ int ompi_coll_tuned_reduce_intra_pipeline( void *sendbuf, void *recvbuf,
                                            ompi_op_t* op, int root,
                                            ompi_communicator_t* comm, uint32_t segsize )
 {
-    int segcount;
+    int segcount = count;
     size_t typelng;
 
     OPAL_OUTPUT((ompi_coll_tuned_stream,"coll:tuned:reduce_intra_pipeline rank %d ss %5d",
@@ -253,11 +249,7 @@ int ompi_coll_tuned_reduce_intra_pipeline( void *sendbuf, void *recvbuf,
      * sent per operation
      */
     ompi_ddt_type_size( datatype, &typelng );
-    if( segsize > typelng ) {
+    COLL_TUNED_COMPUTED_SEGCOUNT( segsize, typelng, segcount );
-        segcount     = (int)(segsize / typelng);
-    } else  {
-        segcount     = count;
-    }
 
     return ompi_coll_tuned_reduce_generic( sendbuf, recvbuf, count, datatype, op, root, comm,
                                            comm->c_coll_selected_data->cached_pipeline, segcount );
@@ -268,7 +260,7 @@ int ompi_coll_tuned_reduce_intra_binary( void *sendbuf, void *recvbuf,
                                          ompi_op_t* op, int root,
                                          ompi_communicator_t* comm, uint32_t segsize )
 {
-    int segcount;
+    int segcount = count;
     size_t typelng;
 
     OPAL_OUTPUT((ompi_coll_tuned_stream,"coll:tuned:reduce_intra_binary rank %d ss %5d",
@@ -281,11 +273,7 @@ int ompi_coll_tuned_reduce_intra_binary( void *sendbuf, void *recvbuf,
      * sent per operation
      */
     ompi_ddt_type_size( datatype, &typelng );
-    if( segsize > typelng ) {
+    COLL_TUNED_COMPUTED_SEGCOUNT( segsize, typelng, segcount );
-        segcount     = (int)(segsize / typelng);
-    } else  {
-        segcount     = count;
-    }
 
     return ompi_coll_tuned_reduce_generic( sendbuf, recvbuf, count, datatype, op, root, comm,
                                            comm->c_coll_selected_data->cached_bintree, segcount );
@@ -296,7 +284,7 @@ int ompi_coll_tuned_reduce_intra_binomial( void *sendbuf, void *recvbuf,
                                            ompi_op_t* op, int root,
                                            ompi_communicator_t* comm, uint32_t segsize )
 {
-    int segcount;
+    int segcount = count;
     size_t typelng;
 
     OPAL_OUTPUT((ompi_coll_tuned_stream,"coll:tuned:reduce_intra_binomial rank %d ss %5d",
@@ -309,11 +297,7 @@ int ompi_coll_tuned_reduce_intra_binomial( void *sendbuf, void *recvbuf,
      * sent per operation
      */
     ompi_ddt_type_size( datatype, &typelng );
-    if( segsize > typelng ) {
+    COLL_TUNED_COMPUTED_SEGCOUNT( segsize, typelng, segcount );
-        segcount     = (int)(segsize / typelng);
-    } else  {
-        segcount     = count;
-    }
 
     return ompi_coll_tuned_reduce_generic( sendbuf, recvbuf, count, datatype, op, root, comm,
                                            comm->c_coll_selected_data->cached_bmtree, segcount );