1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283
|
/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */
/*
* Copyright (c) 2004-2007 The Trustees of Indiana University and Indiana
* University Research and Technology
* Corporation. All rights reserved.
* Copyright (c) 2004-2015 The University of Tennessee and The University
* of Tennessee Research Foundation. All rights
* reserved.
* Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
* University of Stuttgart. All rights reserved.
* Copyright (c) 2004-2005 The Regents of the University of California.
* All rights reserved.
* Copyright (c) 2008 Sun Microsystems, Inc. All rights reserved.
* Copyright (c) 2008 Cisco Systems, Inc. All rights reserved.
* Copyright (c) 2015 Los Alamos National Security, LLC. All rights
* reserved.
* Copyright (c) 2015-2018 Research Organization for Information Science
* and Technology (RIST). All rights reserved.
* Copyright (c) 2019 Mellanox Technologies. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*
* These symbols are in a file by themselves to provide nice linker
* semantics. Since linkers generally pull in symbols by object
* files, keeping these symbols as the only symbols in this file
* prevents utility programs such as "ompi_info" from having to import
* entire components just to query their version and parameters.
*/
#include "ompi_config.h"
#include "opal/util/output.h"
#include "coll_tuned.h"
#include "mpi.h"
#include "ompi/mca/coll/coll.h"
#include "coll_tuned.h"
#include "coll_tuned_dynamic_file.h"
/*
* Public string showing the coll ompi_tuned component version number
*/
const char *ompi_coll_tuned_component_version_string =
"Open MPI tuned collective MCA component version " OMPI_VERSION;
/*
* Global variable
*/
int ompi_coll_tuned_stream = -1;
int ompi_coll_tuned_priority = 30;
bool ompi_coll_tuned_use_dynamic_rules = false;
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 = 200;
int ompi_coll_tuned_alltoall_intermediate_msg = 3000;
/* Set it to the same value as intermediate msg by default, so it does not affect
* default algorithm selection. Changing this value will force using linear with
* sync algorithm on certain message sizes. */
int ompi_coll_tuned_alltoall_large_msg = 3000;
int ompi_coll_tuned_alltoall_min_procs = 0; /* disable by default */
int ompi_coll_tuned_alltoall_max_requests = 0; /* no limit for alltoall by default */
/* Disable by default */
int ompi_coll_tuned_scatter_intermediate_msg = 0;
int ompi_coll_tuned_scatter_large_msg = 0;
int ompi_coll_tuned_scatter_min_procs = 0;
int ompi_coll_tuned_scatter_blocking_send_ratio = 0;
/* forced algorithm variables */
/* indices for the MCA parameters */
coll_tuned_force_algorithm_mca_param_indices_t ompi_coll_tuned_forced_params[COLLCOUNT] = {{0}};
/* max algorithm values */
int ompi_coll_tuned_forced_max_algorithms[COLLCOUNT] = {0};
/*
* Local function
*/
static int tuned_register(void);
static int tuned_open(void);
static int tuned_close(void);
/*
* Instantiate the public struct with all of our public information
* and pointers to our public functions in it
*/
mca_coll_tuned_component_t mca_coll_tuned_component = {
/* First, fill in the super */
{
/* First, the mca_component_t struct containing meta information
about the component itself */
.collm_version = {
MCA_COLL_BASE_VERSION_2_4_0,
/* Component name and version */
.mca_component_name = "tuned",
MCA_BASE_MAKE_VERSION(component, OMPI_MAJOR_VERSION, OMPI_MINOR_VERSION,
OMPI_RELEASE_VERSION),
/* Component open and close functions */
.mca_open_component = tuned_open,
.mca_close_component = tuned_close,
.mca_register_component_params = tuned_register,
},
.collm_data = {
/* The component is checkpoint ready */
MCA_BASE_METADATA_PARAM_CHECKPOINT
},
/* Initialization / querying functions */
.collm_init_query = ompi_coll_tuned_init_query,
.collm_comm_query = ompi_coll_tuned_comm_query,
},
/* priority of the module */
0,
/* Tuned component specific information */
NULL /* ompi_coll_alg_rule_t ptr */
};
static int tuned_register(void)
{
/* Use a low priority, but allow other components to be lower */
ompi_coll_tuned_priority = 30;
(void) mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"priority", "Priority of the tuned coll component",
MCA_BASE_VAR_TYPE_INT, NULL, 0, MCA_BASE_VAR_FLAG_SETTABLE,
OPAL_INFO_LVL_6,
MCA_BASE_VAR_SCOPE_ALL,
&ompi_coll_tuned_priority);
/* some initial guesses at topology parameters */
ompi_coll_tuned_init_tree_fanout = 4;
(void) mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"init_tree_fanout",
"Initial fanout used in the tree topologies for each communicator. This is only an initial guess, if a tuned collective needs a different fanout for an operation, it build it dynamically. This parameter is only for the first guess and might save a little time",
MCA_BASE_VAR_TYPE_INT, NULL, 0, MCA_BASE_VAR_FLAG_SETTABLE,
OPAL_INFO_LVL_6,
MCA_BASE_VAR_SCOPE_ALL,
&ompi_coll_tuned_init_tree_fanout);
ompi_coll_tuned_init_chain_fanout = 4;
(void) mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"init_chain_fanout",
"Initial fanout used in the chain (fanout followed by pipeline) topologies for each communicator. This is only an initial guess, if a tuned collective needs a different fanout for an operation, it build it dynamically. This parameter is only for the first guess and might save a little time",
MCA_BASE_VAR_TYPE_INT, NULL, 0, MCA_BASE_VAR_FLAG_SETTABLE,
OPAL_INFO_LVL_6,
MCA_BASE_VAR_SCOPE_ALL,
&ompi_coll_tuned_init_chain_fanout);
ompi_coll_tuned_alltoall_small_msg = 200;
(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 = 3000;
(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);
(void) mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"use_dynamic_rules",
"Switch used to decide if we use static (compiled/if statements) or dynamic (built at runtime) decision function rules",
MCA_BASE_VAR_TYPE_BOOL, NULL, 0, MCA_BASE_VAR_FLAG_SETTABLE,
OPAL_INFO_LVL_6,
MCA_BASE_VAR_SCOPE_ALL,
&ompi_coll_tuned_use_dynamic_rules);
ompi_coll_tuned_dynamic_rules_filename = NULL;
(void) mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"dynamic_rules_filename",
"Filename of configuration file that contains the dynamic (@runtime) decision function rules",
MCA_BASE_VAR_TYPE_STRING, NULL, 0, MCA_BASE_VAR_FLAG_SETTABLE,
OPAL_INFO_LVL_6,
MCA_BASE_VAR_SCOPE_ALL,
&ompi_coll_tuned_dynamic_rules_filename);
/* register forced params */
ompi_coll_tuned_allreduce_intra_check_forced_init(&ompi_coll_tuned_forced_params[ALLREDUCE]);
ompi_coll_tuned_alltoall_intra_check_forced_init(&ompi_coll_tuned_forced_params[ALLTOALL]);
ompi_coll_tuned_allgather_intra_check_forced_init(&ompi_coll_tuned_forced_params[ALLGATHER]);
ompi_coll_tuned_allgatherv_intra_check_forced_init(&ompi_coll_tuned_forced_params[ALLGATHERV]);
ompi_coll_tuned_alltoallv_intra_check_forced_init(&ompi_coll_tuned_forced_params[ALLTOALLV]);
ompi_coll_tuned_barrier_intra_check_forced_init(&ompi_coll_tuned_forced_params[BARRIER]);
ompi_coll_tuned_bcast_intra_check_forced_init(&ompi_coll_tuned_forced_params[BCAST]);
ompi_coll_tuned_reduce_intra_check_forced_init(&ompi_coll_tuned_forced_params[REDUCE]);
ompi_coll_tuned_reduce_scatter_intra_check_forced_init(&ompi_coll_tuned_forced_params[REDUCESCATTER]);
ompi_coll_tuned_reduce_scatter_block_intra_check_forced_init(&ompi_coll_tuned_forced_params[REDUCESCATTERBLOCK]);
ompi_coll_tuned_gather_intra_check_forced_init(&ompi_coll_tuned_forced_params[GATHER]);
ompi_coll_tuned_scatter_intra_check_forced_init(&ompi_coll_tuned_forced_params[SCATTER]);
ompi_coll_tuned_exscan_intra_check_forced_init(&ompi_coll_tuned_forced_params[EXSCAN]);
ompi_coll_tuned_scan_intra_check_forced_init(&ompi_coll_tuned_forced_params[SCAN]);
return OMPI_SUCCESS;
}
static int tuned_open(void)
{
int rc;
#if OPAL_ENABLE_DEBUG
if (ompi_coll_base_framework.framework_verbose) {
ompi_coll_tuned_stream = opal_output_open(NULL);
}
#endif /* OPAL_ENABLE_DEBUG */
/* now check that the user hasn't overrode any of the decision functions if dynamic rules are enabled */
/* the user can redo this before every comm dup/create if they like */
/* this is useful for benchmarking and user knows best tuning */
/* as this is the component we only lookup the indices of the mca params */
/* the actual values are looked up during comm create via module init */
/* intra functions first */
/* if dynamic rules allowed then look up dynamic rules config filename, else we leave it an empty filename (NULL) */
/* by default DISABLE dynamic rules and instead use fixed [if based] rules */
if (ompi_coll_tuned_use_dynamic_rules) {
if( ompi_coll_tuned_dynamic_rules_filename ) {
OPAL_OUTPUT((ompi_coll_tuned_stream,"coll:tuned:component_open Reading collective rules file [%s]",
ompi_coll_tuned_dynamic_rules_filename));
rc = ompi_coll_tuned_read_rules_config_file( ompi_coll_tuned_dynamic_rules_filename,
&(mca_coll_tuned_component.all_base_rules), COLLCOUNT);
if( rc >= 0 ) {
OPAL_OUTPUT((ompi_coll_tuned_stream,"coll:tuned:module_open Read %d valid rules\n", rc));
} else {
OPAL_OUTPUT((ompi_coll_tuned_stream,"coll:tuned:module_open Reading collective rules file failed\n"));
mca_coll_tuned_component.all_base_rules = NULL;
}
}
}
OPAL_OUTPUT((ompi_coll_tuned_stream, "coll:tuned:component_open: done!"));
return OMPI_SUCCESS;
}
/* here we should clean up state stored on the component */
/* i.e. alg table and dynamic changeable rules if allocated etc */
static int tuned_close(void)
{
OPAL_OUTPUT((ompi_coll_tuned_stream, "coll:tuned:component_close: called"));
/* dealloc alg table if allocated */
/* dealloc dynamic changeable rules if allocated */
OPAL_OUTPUT((ompi_coll_tuned_stream, "coll:tuned:component_close: done!"));
if( NULL != mca_coll_tuned_component.all_base_rules ) {
ompi_coll_tuned_free_all_rules(mca_coll_tuned_component.all_base_rules, COLLCOUNT);
mca_coll_tuned_component.all_base_rules = NULL;
}
return OMPI_SUCCESS;
}
static void
mca_coll_tuned_module_construct(mca_coll_tuned_module_t *module)
{
mca_coll_tuned_module_t *tuned_module = (mca_coll_tuned_module_t*) module;
for( int i = 0; i < COLLCOUNT; i++ ) {
tuned_module->user_forced[i].algorithm = 0;
tuned_module->com_rules[i] = NULL;
}
}
OBJ_CLASS_INSTANCE(mca_coll_tuned_module_t, mca_coll_base_module_t,
mca_coll_tuned_module_construct, NULL);
|