OMPI/COLL/Tuned: add command line params for thresholds to decide if small/intermediate MSGs alltoall algorithm will be used.

cmr=v1.8.3:reviewer=miked

This commit was SVN r32735.

ompi/mca/coll/tuned/coll_tuned.h

@@ -88,6 +88,8 @@ extern char* ompi_coll_tuned_dynamic_rules_filename;
 extern int   ompi_coll_tuned_init_tree_fanout;
 extern int   ompi_coll_tuned_init_chain_fanout;
 extern int   ompi_coll_tuned_init_max_requests;
+extern int   ompi_coll_tuned_alltoall_small_msg;
+extern int   ompi_coll_tuned_alltoall_intermediate_msg;
 
 /* forced algorithm choices */
 /* this structure is for storing the indexes to the forced algorithm mca params... */

ompi/mca/coll/tuned/coll_tuned_component.c

@@ -50,6 +50,8 @@ char* ompi_coll_tuned_dynamic_rules_filename = (char*) NULL;
 int   ompi_coll_tuned_init_tree_fanout = 4;
 int   ompi_coll_tuned_init_chain_fanout = 4;
 int   ompi_coll_tuned_init_max_requests = 128;
+int   ompi_coll_tuned_alltoall_small_msg = 1000;
+int   ompi_coll_tuned_alltoall_intermediate_msg = 2000;
 
 /* forced alogrithm variables */
 /* indices for the MCA parameters */
@@ -148,6 +150,24 @@ static int tuned_register(void)
                                            MCA_BASE_VAR_SCOPE_READONLY,
                                            &ompi_coll_tuned_init_chain_fanout);
 
+    ompi_coll_tuned_alltoall_small_msg = 1000;
+    (void) mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
+                                           "alltoall_small_msg",
+                                           "threshold (if supported) to decide if small MSGs alltoall algorithm will be used",
+                                           MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
+                                           OPAL_INFO_LVL_6,
+                                           MCA_BASE_VAR_SCOPE_READONLY,
+                                           &ompi_coll_tuned_alltoall_small_msg);
+
+    ompi_coll_tuned_alltoall_intermediate_msg = 2000;
+    (void) mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
+                                           "alltoall_intermediate_msg",
+                                           "threshold (if supported) to decide if intermediate MSGs alltoall algorithm will be used",
+                                           MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
+                                           OPAL_INFO_LVL_6,
+                                           MCA_BASE_VAR_SCOPE_READONLY,
+                                           &ompi_coll_tuned_alltoall_intermediate_msg);
+
     ompi_coll_tuned_use_dynamic_rules = false;
     (void) mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
                                            "use_dynamic_rules",

ompi/mca/coll/tuned/coll_tuned_decision_fixed.c

@@ -121,12 +121,13 @@ int ompi_coll_tuned_alltoall_intra_dec_fixed(void *sbuf, int scount,
     ompi_datatype_type_size(sdtype, &dsize);
     block_dsize = dsize * (ptrdiff_t)scount;
 
-    if ((block_dsize < 200) && (communicator_size > 12)) {
+    if ((block_dsize < ompi_coll_tuned_alltoall_small_msg)
+                                              && (communicator_size > 12)) {
         return ompi_coll_tuned_alltoall_intra_bruck(sbuf, scount, sdtype, 
                                                     rbuf, rcount, rdtype,
                                                     comm, module);
 
-    } else if (block_dsize < 3000) {
+    } else if (block_dsize < ompi_coll_tuned_alltoall_intermediate_msg) {
         return ompi_coll_tuned_alltoall_intra_basic_linear(sbuf, scount, sdtype, 
                                                            rbuf, rcount, rdtype, 
                                                            comm, module);