root/ompi/mca/coll/tuned/coll_tuned_bcast_decision.c

DEFINITIONS

1. ompi_coll_tuned_bcast_intra_check_forced_init
2. ompi_coll_tuned_bcast_intra_do_this

   1 /* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */
   2 /*
   3  * Copyright (c) 2004-2017 The University of Tennessee and The University
   4  *                         of Tennessee Research Foundation.  All rights
   5  *                         reserved.
   6  * Copyright (c) 2015      Research Organization for Information Science
   7  *                         and Technology (RIST). All rights reserved.
   8  * $COPYRIGHT$
   9  *
  10  * Additional copyrights may follow
  11  *
  12  * $HEADER$
  13  */
  14 
  15 #include "ompi_config.h"
  16 
  17 #include "mpi.h"
  18 #include "ompi/constants.h"
  19 #include "ompi/datatype/ompi_datatype.h"
  20 #include "ompi/communicator/communicator.h"
  21 #include "ompi/mca/coll/coll.h"
  22 #include "ompi/mca/coll/base/coll_tags.h"
  23 #include "ompi/mca/pml/pml.h"
  24 #include "coll_tuned.h"
  25 #include "ompi/mca/coll/base/coll_base_topo.h"
  26 #include "ompi/mca/coll/base/coll_base_util.h"
  27 
  28 /* bcast algorithm variables */
  29 static int coll_tuned_bcast_forced_algorithm = 0;
  30 static int coll_tuned_bcast_segment_size = 0;
  31 static int coll_tuned_bcast_tree_fanout;
  32 static int coll_tuned_bcast_chain_fanout;
  33 /* k-nomial tree radix for the bcast algorithm (>= 2) */
  34 static int coll_tuned_bcast_knomial_radix = 4;
  35 
  36 /* valid values for coll_tuned_bcast_forced_algorithm */
  37 static mca_base_var_enum_value_t bcast_algorithms[] = {
  38     {0, "ignore"},
  39     {1, "basic_linear"},
  40     {2, "chain"},
  41     {3, "pipeline"},
  42     {4, "split_binary_tree"},
  43     {5, "binary_tree"},
  44     {6, "binomial"},
  45     {7, "knomial"},
  46     {8, "scatter_allgather"},
  47     {9, "scatter_allgather_ring"},
  48     {0, NULL}
  49 };
  50 
  51 /* The following are used by dynamic and forced rules */
  52 
  53 /* publish details of each algorithm and if its forced/fixed/locked in */
  54 /* as you add methods/algorithms you must update this and the query/map routines */
  55 
  56 /* this routine is called by the component only */
  57 /* this makes sure that the mca parameters are set to their initial values and perms */
  58 /* module does not call this they call the forced_getvalues routine instead */
  59 
  60 int ompi_coll_tuned_bcast_intra_check_forced_init (coll_tuned_force_algorithm_mca_param_indices_t *mca_param_indices)
  61 {
  62     mca_base_var_enum_t *new_enum;
  63     int cnt;
  64 
  65     for( cnt = 0; NULL != bcast_algorithms[cnt].string; cnt++ );
  66     ompi_coll_tuned_forced_max_algorithms[BCAST] = cnt;
  67 
  68     (void) mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
  69                                            "bcast_algorithm_count",
  70                                            "Number of bcast algorithms available",
  71                                            MCA_BASE_VAR_TYPE_INT, NULL, 0,
  72                                            MCA_BASE_VAR_FLAG_DEFAULT_ONLY,
  73                                            OPAL_INFO_LVL_5,
  74                                            MCA_BASE_VAR_SCOPE_CONSTANT,
  75                                            &ompi_coll_tuned_forced_max_algorithms[BCAST]);
  76 
  77     /* MPI_T: This variable should eventually be bound to a communicator */
  78     coll_tuned_bcast_forced_algorithm = 0;
  79     (void) mca_base_var_enum_create("coll_tuned_bcast_algorithms", bcast_algorithms, &new_enum);
  80     mca_param_indices->algorithm_param_index =
  81         mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
  82                                         "bcast_algorithm",
  83                                         "Which bcast algorithm is used. Can be locked down to choice of: 0 ignore, 1 basic linear, 2 chain, 3: pipeline, 4: split binary tree, 5: binary tree, 6: binomial tree, 7: knomial tree, 8: scatter_allgather, 9: scatter_allgather_ring.",
  84                                         MCA_BASE_VAR_TYPE_INT, new_enum, 0, MCA_BASE_VAR_FLAG_SETTABLE,
  85                                         OPAL_INFO_LVL_5,
  86                                         MCA_BASE_VAR_SCOPE_ALL,
  87                                         &coll_tuned_bcast_forced_algorithm);
  88     OBJ_RELEASE(new_enum);
  89     if (mca_param_indices->algorithm_param_index < 0) {
  90         return mca_param_indices->algorithm_param_index;
  91     }
  92 
  93     coll_tuned_bcast_segment_size = 0;
  94     mca_param_indices->segsize_param_index =
  95         mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
  96                                         "bcast_algorithm_segmentsize",
  97                                         "Segment size in bytes used by default for bcast algorithms. Only has meaning if algorithm is forced and supports segmenting. 0 bytes means no segmentation.",
  98                                         MCA_BASE_VAR_TYPE_INT, NULL, 0, MCA_BASE_VAR_FLAG_SETTABLE,
  99                                         OPAL_INFO_LVL_5,
 100                                         MCA_BASE_VAR_SCOPE_ALL,
 101                                         &coll_tuned_bcast_segment_size);
 102 
 103     coll_tuned_bcast_tree_fanout = ompi_coll_tuned_init_tree_fanout; /* get system wide default */
 104     mca_param_indices->tree_fanout_param_index =
 105         mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
 106                                         "bcast_algorithm_tree_fanout",
 107                                         "Fanout for n-tree used for bcast algorithms. Only has meaning if algorithm is forced and supports n-tree topo based operation.",
 108                                         MCA_BASE_VAR_TYPE_INT, NULL, 0, MCA_BASE_VAR_FLAG_SETTABLE,
 109                                         OPAL_INFO_LVL_5,
 110                                         MCA_BASE_VAR_SCOPE_ALL,
 111                                         &coll_tuned_bcast_tree_fanout);
 112 
 113     coll_tuned_bcast_chain_fanout = ompi_coll_tuned_init_chain_fanout; /* get system wide default */
 114     mca_param_indices->chain_fanout_param_index =
 115       mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
 116                                       "bcast_algorithm_chain_fanout",
 117                                       "Fanout for chains used for bcast algorithms. Only has meaning if algorithm is forced and supports chain topo based operation.",
 118                                       MCA_BASE_VAR_TYPE_INT, NULL, 0, MCA_BASE_VAR_FLAG_SETTABLE,
 119                                       OPAL_INFO_LVL_5,
 120                                       MCA_BASE_VAR_SCOPE_ALL,
 121                                       &coll_tuned_bcast_chain_fanout);
 122 
 123     coll_tuned_bcast_knomial_radix = 4;
 124     mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
 125                                     "bcast_algorithm_knomial_radix",
 126                                     "k-nomial tree radix for the bcast algorithm (radix > 1).",
 127                                     MCA_BASE_VAR_TYPE_INT, NULL, 0, MCA_BASE_VAR_FLAG_SETTABLE,
 128                                     OPAL_INFO_LVL_5, MCA_BASE_VAR_SCOPE_ALL,
 129                                     &coll_tuned_bcast_knomial_radix);
 130 
 131     return (MPI_SUCCESS);
 132 }
 133 
 134 int ompi_coll_tuned_bcast_intra_do_this(void *buf, int count,
 135                                         struct ompi_datatype_t *dtype,
 136                                         int root,
 137                                         struct ompi_communicator_t *comm,
 138                                         mca_coll_base_module_t *module,
 139                                         int algorithm, int faninout, int segsize)
 140 {
 141     OPAL_OUTPUT((ompi_coll_tuned_stream,"coll:tuned:bcast_intra_do_this algorithm %d topo faninout %d segsize %d",
 142                  algorithm, faninout, segsize));
 143 
 144     switch (algorithm) {
 145     case (0):
 146         return ompi_coll_tuned_bcast_intra_dec_fixed( buf, count, dtype, root, comm, module );
 147     case (1):
 148         return ompi_coll_base_bcast_intra_basic_linear( buf, count, dtype, root, comm, module );
 149     case (2):
 150         return ompi_coll_base_bcast_intra_chain( buf, count, dtype, root, comm, module, segsize, faninout );
 151     case (3):
 152         return ompi_coll_base_bcast_intra_pipeline( buf, count, dtype, root, comm, module, segsize );
 153     case (4):
 154         return ompi_coll_base_bcast_intra_split_bintree( buf, count, dtype, root, comm, module, segsize );
 155     case (5):
 156         return ompi_coll_base_bcast_intra_bintree( buf, count, dtype, root, comm, module, segsize );
 157     case (6):
 158         return ompi_coll_base_bcast_intra_binomial( buf, count, dtype, root, comm, module, segsize );
 159     case (7):
 160         return ompi_coll_base_bcast_intra_knomial(buf, count, dtype, root, comm, module,
 161                                                   segsize, coll_tuned_bcast_knomial_radix);
 162     case (8):
 163         return ompi_coll_base_bcast_intra_scatter_allgather(buf, count, dtype, root, comm, module, segsize);
 164     case (9):
 165         return ompi_coll_base_bcast_intra_scatter_allgather_ring(buf, count, dtype, root, comm, module, segsize);
 166     } /* switch */
 167     OPAL_OUTPUT((ompi_coll_tuned_stream,"coll:tuned:bcast_intra_do_this attempt to select algorithm %d when only 0-%d is valid?",
 168                  algorithm, ompi_coll_tuned_forced_max_algorithms[BCAST]));
 169     return (MPI_ERR_ARG);
 170 }