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

DEFINITIONS

1. ompi_coll_tuned_alltoall_intra_check_forced_init
2. ompi_coll_tuned_alltoall_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 /* alltoall algorithm variables */
  29 static int coll_tuned_alltoall_forced_algorithm = 0;
  30 static int coll_tuned_alltoall_segment_size = 0;
  31 static int coll_tuned_alltoall_max_requests;
  32 static int coll_tuned_alltoall_tree_fanout;
  33 static int coll_tuned_alltoall_chain_fanout;
  34 
  35 /* valid values for coll_tuned_alltoall_forced_algorithm */
  36 static mca_base_var_enum_value_t alltoall_algorithms[] = {
  37     {0, "ignore"},
  38     {1, "linear"},
  39     {2, "pairwise"},
  40     {3, "modified_bruck"},
  41     {4, "linear_sync"},
  42     {5, "two_proc"},
  43     {0, NULL}
  44 };
  45 
  46 /* The following are used by dynamic and forced rules */
  47 
  48 /* publish details of each algorithm and if its forced/fixed/locked in */
  49 /* as you add methods/algorithms you must update this and the query/map routines */
  50 
  51 /* this routine is called by the component only */
  52 /* this makes sure that the mca parameters are set to their initial values and perms */
  53 /* module does not call this they call the forced_getvalues routine instead */
  54 
  55 int ompi_coll_tuned_alltoall_intra_check_forced_init (coll_tuned_force_algorithm_mca_param_indices_t *mca_param_indices)
  56 {
  57     mca_base_var_enum_t*new_enum;
  58     int cnt;
  59 
  60     for( cnt = 0; NULL != alltoall_algorithms[cnt].string; cnt++ );
  61     ompi_coll_tuned_forced_max_algorithms[ALLTOALL] = cnt;
  62 
  63     (void) mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
  64                                            "alltoall_algorithm_count",
  65                                            "Number of alltoall algorithms available",
  66                                            MCA_BASE_VAR_TYPE_INT, NULL, 0,
  67                                            MCA_BASE_VAR_FLAG_DEFAULT_ONLY,
  68                                            OPAL_INFO_LVL_5,
  69                                            MCA_BASE_VAR_SCOPE_CONSTANT,
  70                                            &ompi_coll_tuned_forced_max_algorithms[ALLTOALL]);
  71 
  72     /* MPI_T: This variable should eventually be bound to a communicator */
  73     coll_tuned_alltoall_forced_algorithm = 0;
  74     (void) mca_base_var_enum_create("coll_tuned_alltoall_algorithms", alltoall_algorithms, &new_enum);
  75     mca_param_indices->algorithm_param_index =
  76         mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
  77                                         "alltoall_algorithm",
  78                                         "Which alltoall algorithm is used. Can be locked down to choice of: 0 ignore, 1 basic linear, 2 pairwise, 3: modified bruck, 4: linear with sync, 5:two proc only.",
  79                                         MCA_BASE_VAR_TYPE_INT, new_enum, 0, MCA_BASE_VAR_FLAG_SETTABLE,
  80                                         OPAL_INFO_LVL_5,
  81                                         MCA_BASE_VAR_SCOPE_ALL,
  82                                         &coll_tuned_alltoall_forced_algorithm);
  83     OBJ_RELEASE(new_enum);
  84     if (mca_param_indices->algorithm_param_index < 0) {
  85         return mca_param_indices->algorithm_param_index;
  86     }
  87 
  88     coll_tuned_alltoall_segment_size = 0;
  89     mca_param_indices->segsize_param_index =
  90         mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
  91                                         "alltoall_algorithm_segmentsize",
  92                                         "Segment size in bytes used by default for alltoall algorithms. Only has meaning if algorithm is forced and supports segmenting. 0 bytes means no segmentation.",
  93                                         MCA_BASE_VAR_TYPE_INT, NULL, 0, MCA_BASE_VAR_FLAG_SETTABLE,
  94                                         OPAL_INFO_LVL_5,
  95                                         MCA_BASE_VAR_SCOPE_ALL,
  96                                         &coll_tuned_alltoall_segment_size);
  97 
  98     coll_tuned_alltoall_tree_fanout = ompi_coll_tuned_init_tree_fanout; /* get system wide default */
  99     mca_param_indices->tree_fanout_param_index =
 100         mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
 101                                         "alltoall_algorithm_tree_fanout",
 102                                         "Fanout for n-tree used for alltoall algorithms. Only has meaning if algorithm is forced and supports n-tree topo based operation.",
 103                                         MCA_BASE_VAR_TYPE_INT, NULL, 0, MCA_BASE_VAR_FLAG_SETTABLE,
 104                                         OPAL_INFO_LVL_5,
 105                                         MCA_BASE_VAR_SCOPE_ALL,
 106                                         &coll_tuned_alltoall_tree_fanout);
 107 
 108     coll_tuned_alltoall_chain_fanout = ompi_coll_tuned_init_chain_fanout; /* get system wide default */
 109     mca_param_indices->chain_fanout_param_index =
 110       mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
 111                                       "alltoall_algorithm_chain_fanout",
 112                                       "Fanout for chains used for alltoall algorithms. Only has meaning if algorithm is forced and supports chain topo based operation.",
 113                                       MCA_BASE_VAR_TYPE_INT, NULL, 0, MCA_BASE_VAR_FLAG_SETTABLE,
 114                                       OPAL_INFO_LVL_5,
 115                                       MCA_BASE_VAR_SCOPE_ALL,
 116                                       &coll_tuned_alltoall_chain_fanout);
 117 
 118     coll_tuned_alltoall_max_requests = 0; /* no limit for alltoall by default */
 119     mca_param_indices->max_requests_param_index =
 120       mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
 121                                       "alltoall_algorithm_max_requests",
 122                                       "Maximum number of outstanding send or recv requests.  Only has meaning for synchronized algorithms.",
 123                                       MCA_BASE_VAR_TYPE_INT, NULL, 0, MCA_BASE_VAR_FLAG_SETTABLE,
 124                                       OPAL_INFO_LVL_5,
 125                                       MCA_BASE_VAR_SCOPE_ALL,
 126                                       &coll_tuned_alltoall_max_requests);
 127     if (mca_param_indices->max_requests_param_index < 0) {
 128         return mca_param_indices->max_requests_param_index;
 129     }
 130 
 131     if (coll_tuned_alltoall_max_requests < 0) {
 132         if( 0 == ompi_comm_rank( MPI_COMM_WORLD ) ) {
 133             opal_output( 0, "Maximum outstanding requests must be positive number greater than 1.  Switching to system level default %d \n",
 134                          ompi_coll_tuned_init_max_requests );
 135         }
 136         coll_tuned_alltoall_max_requests = 0;
 137     }
 138 
 139     return (MPI_SUCCESS);
 140 }
 141 
 142 int ompi_coll_tuned_alltoall_intra_do_this(const void *sbuf, int scount,
 143                                            struct ompi_datatype_t *sdtype,
 144                                            void* rbuf, int rcount,
 145                                            struct ompi_datatype_t *rdtype,
 146                                            struct ompi_communicator_t *comm,
 147                                            mca_coll_base_module_t *module,
 148                                            int algorithm, int faninout, int segsize,
 149                                            int max_requests)
 150 {
 151     OPAL_OUTPUT((ompi_coll_tuned_stream,"coll:tuned:alltoall_intra_do_this selected algorithm %d topo faninout %d segsize %d",
 152                  algorithm, faninout, segsize));
 153 
 154     switch (algorithm) {
 155     case (0):
 156         return ompi_coll_tuned_alltoall_intra_dec_fixed(sbuf, scount, sdtype, rbuf, rcount, rdtype, comm, module);
 157     case (1):
 158         return ompi_coll_base_alltoall_intra_basic_linear(sbuf, scount, sdtype, rbuf, rcount, rdtype, comm, module);
 159     case (2):
 160         return ompi_coll_base_alltoall_intra_pairwise(sbuf, scount, sdtype, rbuf, rcount, rdtype, comm, module);
 161     case (3):
 162         return ompi_coll_base_alltoall_intra_bruck(sbuf, scount, sdtype, rbuf, rcount, rdtype, comm, module);
 163     case (4):
 164         return ompi_coll_base_alltoall_intra_linear_sync(sbuf, scount, sdtype, rbuf, rcount, rdtype, comm, module, max_requests);
 165     case (5):
 166         return ompi_coll_base_alltoall_intra_two_procs(sbuf, scount, sdtype, rbuf, rcount, rdtype, comm, module);
 167     } /* switch */
 168     OPAL_OUTPUT((ompi_coll_tuned_stream,"coll:tuned:alltoall_intra_do_this attempt to select algorithm %d when only 0-%d is valid?",
 169                  algorithm, ompi_coll_tuned_forced_max_algorithms[ALLTOALL]));
 170     return (MPI_ERR_ARG);
 171 }