src/smpi/internals/smpi_config.cpp

   1 /* Copyright (c) 2008-2023. The SimGrid Team. All rights reserved.          */
   2
   3 /* This program is free software; you can redistribute it and/or modify it
   4  * under the terms of the license (GNU LGPL) which comes with this package. */
   5
   6 #if defined(_GNU_SOURCE)
   7   #define DEFINED_GNUSOURCE 1
   8 #else
   9   #define _GNU_SOURCE
  10 #endif
  11
  12 #if defined(__linux__)
  13   #include <features.h>
  14 //inspired by https://stackoverflow.com/a/70211227
  15   #if not defined(__USE_GNU)
  16     #define __MUSL__
  17   #endif
  18 #endif
  19
  20 #ifndef DEFINED_GNUSOURCE
  21   #undef _GNU_SOURCE
  22 #endif
  23
  24 #include "private.hpp"
  25 #include "smpi_coll.hpp"
  26 #include "smpi_config.hpp"
  27 #include "src/mc/mc.h"
  28 #include "xbt/config.hpp"
  29 #include "xbt/ex.h"
  30 #include "xbt/parse_units.hpp"
  31
  32 #include <cfloat> /* DBL_MAX */
  33 #include <boost/algorithm/string.hpp> /* trim */
  34 #include <boost/tokenizer.hpp>
  35
  36 #include "src/mc/mc_config.hpp"
  37 #include "src/mc/mc_replay.hpp"
  38
  39 #if defined(__APPLE__)
  40 # include <AvailabilityMacros.h>
  41 # ifndef MAC_OS_X_VERSION_10_12
  42 #   define MAC_OS_X_VERSION_10_12 101200
  43 # endif
  44 # ifndef __MAC_11_0
  45 #   define __MAC_11_0 110000
  46 # endif
  47
  48 constexpr bool HAVE_WORKING_MMAP = ((MAC_OS_X_VERSION_MIN_REQUIRED >= MAC_OS_X_VERSION_10_12) && (MAC_OS_X_VERSION_MIN_REQUIRED < __MAC_11_0));
  49 #elif defined(__FreeBSD__) || defined(__FreeBSD_kernel__) || defined(__sun) || defined(__HAIKU__) || defined(__MUSL__)
  50 constexpr bool HAVE_WORKING_MMAP = false;
  51 #else
  52 constexpr bool HAVE_WORKING_MMAP = true;
  53 #endif
  54
  55 SharedMallocType _smpi_cfg_shared_malloc = SharedMallocType::GLOBAL;
  56 SmpiPrivStrategies _smpi_cfg_privatization = SmpiPrivStrategies::NONE;
  57 double _smpi_cfg_host_speed;
  58
  59 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(smpi_config, smpi, "Logging specific to SMPI (config)");
  60
  61 simgrid::config::Flag<std::string> _smpi_cfg_host_speed_string{
  62     "smpi/host-speed", "Speed of the host running the simulation (in flop/s). Used to bench the operations.", "20000f",
  63     [](const std::string& str) {
  64       _smpi_cfg_host_speed = xbt_parse_get_speed("smpi/host-speed", 1, str, "option smpi/host-speed");
  65       xbt_assert(_smpi_cfg_host_speed > 0.0, "Invalid value (%s) for 'smpi/host-speed': it must be positive.",
  66                  str.c_str());
  67     }};
  68
  69 simgrid::config::Flag<bool> _smpi_cfg_simulate_computation{
  70   "smpi/simulate-computation", "Whether the computational part of the simulated application should be simulated.",
  71    true};
  72 simgrid::config::Flag<std::string> _smpi_cfg_shared_malloc_string{
  73     "smpi/shared-malloc", "Whether SMPI_SHARED_MALLOC is enabled. Disable it for debugging purposes.", "global",
  74     [](const std::string& val) {
  75       if ((val == "yes") || (val == "1") || (val == "on") || (val == "global")) {
  76         _smpi_cfg_shared_malloc = SharedMallocType::GLOBAL;
  77       } else if (val == "local") {
  78         _smpi_cfg_shared_malloc = SharedMallocType::LOCAL;
  79       } else if ((val == "no") || (val == "0") || (val == "off")) {
  80         _smpi_cfg_shared_malloc = SharedMallocType::NONE;
  81       } else {
  82         xbt_die("Invalid value '%s' for option smpi/shared-malloc. Possible values: 'on' or 'global', 'local', 'off'",
  83                 val.c_str());
  84       }
  85     }};
  86
  87 simgrid::config::Flag<double> _smpi_cfg_cpu_threshold{
  88   "smpi/cpu-threshold", "Minimal computation time (in seconds) not discarded, or -1 for infinity.", 1e-6,
  89   [](const double& val){
  90     if (val < 0)
  91       _smpi_cfg_cpu_threshold = DBL_MAX;
  92   }};
  93
  94 simgrid::config::Flag<int> _smpi_cfg_async_small_thresh{"smpi/async-small-thresh",
  95                                                         "Maximal size of messages that are to be sent asynchronously, without waiting for the receiver",
  96                                                         0};
  97 simgrid::config::Flag<int> _smpi_cfg_detached_send_thresh{
  98     "smpi/send-is-detached-thresh",
  99     "Threshold of message size where MPI_Send stops behaving like MPI_Isend and becomes MPI_Ssend", 65536};
 100 simgrid::config::Flag<bool> _smpi_cfg_grow_injected_times{"smpi/grow-injected-times",
 101                                                           "Whether we want to make the injected time in MPI_Iprobe and MPI_Test grow, to "
 102                                                           "allow faster simulation. This can make simulation less precise, though.",
 103                                                           true};
 104 simgrid::config::Flag<double> _smpi_cfg_iprobe_cpu_usage{"smpi/iprobe-cpu-usage",
 105                                                         "Maximum usage of CPUs by MPI_Iprobe() calls. We've observed that MPI_Iprobes "
 106                                                         "consume significantly less power than the maximum of a specific application. "
 107                                                         "This value is then (Iprobe_Usage/Max_Application_Usage).",
 108                                                         1.0};
 109
 110 simgrid::config::Flag<bool>  _smpi_cfg_trace_call_location{"smpi/trace-call-location",
 111                                                            "Should filename and linenumber of MPI calls be traced?", false};
 112 simgrid::config::Flag<bool> _smpi_cfg_trace_call_use_absolute_path{"smpi/trace-call-use-absolute-path",
 113                                                                    "Should filenames for trace-call tracing be absolute or not?", false};
 114 simgrid::config::Flag<std::string> _smpi_cfg_comp_adjustment_file{
 115     "smpi/comp-adjustment-file", "A file containing speedups or slowdowns for some parts of the code.", "",
 116     [](const std::string& filename) {
 117       if (not filename.empty()) {
 118         std::ifstream fstream(filename);
 119         xbt_assert(fstream.is_open(), "Could not open file %s. Does it exist?", filename.c_str());
 120         std::string line;
 121         using Tokenizer = boost::tokenizer<boost::escaped_list_separator<char>>;
 122         std::getline(fstream, line); // Skip the header line
 123         while (std::getline(fstream, line)) {
 124           Tokenizer tok(line);
 125           auto it              = tok.begin();
 126           auto end             = std::next(tok.begin());
 127           std::string location = *it;
 128           boost::trim(location);
 129           location2speedup.try_emplace(location, std::stod(*end));
 130         }
 131       }
 132     }};
 133
 134 simgrid::config::Flag<bool> _smpi_cfg_default_errhandler_is_error{
 135   "smpi/errors-are-fatal", "Whether MPI errors are fatal or just return. Default is true", true };
 136 simgrid::config::Flag<bool> _smpi_cfg_pedantic{
 137   "smpi/pedantic", "Activate extra checks that may crash slightly incorrect codes which would not crash on actual implementations", false };
 138 simgrid::config::Flag<double> _smpi_init_sleep(
 139   "smpi/init", "Time to inject inside a call to MPI_Init", 0.0);
 140 #if HAVE_PAPI
 141   simgrid::config::Flag<std::string> _smpi_cfg_papi_events_file{"smpi/papi-events",
 142                                                                 "This switch enables tracking the specified counters with PAPI", ""};
 143 #endif
 144
 145 simgrid::config::Flag<double> _smpi_cfg_auto_shared_malloc_thresh("smpi/auto-shared-malloc-thresh",
 146                                                                   "Threshold size for the automatic sharing of memory",
 147                                                                   0);
 148
 149 simgrid::config::Flag<bool> _smpi_cfg_display_alloc("smpi/display-allocs",
 150                                                     "Whether we should display a memory allocations analysis after simulation.",
 151                                                      false);
 152
 153 simgrid::config::Flag<int> _smpi_cfg_list_leaks("smpi/list-leaks",
 154                                                 "Whether we should display the n first MPI handle leaks (addresses and type only) after simulation",
 155                                                 -1);
 156
 157 double smpi_cfg_host_speed(){
 158   return _smpi_cfg_host_speed;
 159 }
 160
 161 bool smpi_cfg_simulate_computation(){
 162   return _smpi_cfg_simulate_computation;
 163 }
 164
 165 SharedMallocType smpi_cfg_shared_malloc(){
 166   return _smpi_cfg_shared_malloc;
 167 }
 168
 169 double smpi_cfg_cpu_thresh(){
 170   return _smpi_cfg_cpu_threshold;
 171 }
 172
 173 SmpiPrivStrategies smpi_cfg_privatization(){
 174   return _smpi_cfg_privatization;
 175 }
 176
 177 int smpi_cfg_async_small_thresh(){
 178   return _smpi_cfg_async_small_thresh;
 179 }
 180
 181 int smpi_cfg_detached_send_thresh(){
 182   return _smpi_cfg_detached_send_thresh;
 183 }
 184
 185 bool smpi_cfg_grow_injected_times(){
 186   return _smpi_cfg_grow_injected_times;
 187 }
 188
 189 double smpi_cfg_iprobe_cpu_usage(){
 190   return _smpi_cfg_iprobe_cpu_usage;
 191 }
 192
 193 bool smpi_cfg_trace_call_location(){
 194   return _smpi_cfg_trace_call_location;
 195 }
 196
 197 bool smpi_cfg_trace_call_use_absolute_path(){
 198   return _smpi_cfg_trace_call_use_absolute_path;
 199 }
 200
 201 bool smpi_cfg_display_alloc(){
 202   return _smpi_cfg_list_leaks != -1 ? true : _smpi_cfg_display_alloc;
 203 }
 204
 205 std::string smpi_cfg_comp_adjustment_file(){
 206   return _smpi_cfg_comp_adjustment_file;
 207 }
 208 #if HAVE_PAPI
 209 std::string smpi_cfg_papi_events_file(){
 210   return _smpi_cfg_papi_events_file;
 211 }
 212 #endif
 213 double smpi_cfg_auto_shared_malloc_thresh(){
 214   return _smpi_cfg_auto_shared_malloc_thresh;
 215 }
 216
 217 // public version declared in smpi.h (without parameter, and with C linkage)
 218 void smpi_init_options()
 219 {
 220   smpi_init_options_internal(false);
 221 }
 222
 223 void smpi_init_options_internal(bool called_by_smpi_main)
 224 {
 225   static bool smpi_options_initialized = false;
 226   static bool running_with_smpi_main   = false;
 227
 228   if (called_by_smpi_main)
 229     running_with_smpi_main = true;
 230
 231   // return if already called
 232   if (smpi_options_initialized)
 233     return;
 234   simgrid::config::declare_flag<bool>("smpi/display-timing", "Whether we should display the timing after simulation.", false);
 235   simgrid::config::declare_flag<bool>("smpi/keep-temps", "Whether we should keep the generated temporary files.", false);
 236   simgrid::config::declare_flag<std::string>("smpi/tmpdir", "tmp dir for dlopen files", "/tmp");
 237
 238   simgrid::config::declare_flag<std::string>("smpi/coll-selector", "Which collective selector to use", "default");
 239   simgrid::config::declare_flag<std::string>("smpi/gather", "Which collective to use for gather", "");
 240   simgrid::config::declare_flag<std::string>("smpi/allgather", "Which collective to use for allgather", "");
 241   simgrid::config::declare_flag<std::string>("smpi/barrier", "Which collective to use for barrier", "");
 242   simgrid::config::declare_flag<std::string>("smpi/reduce_scatter", "Which collective to use for reduce_scatter", "");
 243   simgrid::config::declare_flag<std::string>("smpi/scatter", "Which collective to use for scatter", "");
 244   simgrid::config::declare_flag<std::string>("smpi/allgatherv", "Which collective to use for allgatherv", "");
 245   simgrid::config::declare_flag<std::string>("smpi/allreduce", "Which collective to use for allreduce", "");
 246   simgrid::config::declare_flag<std::string>("smpi/alltoall", "Which collective to use for alltoall", "");
 247   simgrid::config::declare_flag<std::string>("smpi/alltoallv", "Which collective to use for alltoallv", "");
 248   simgrid::config::declare_flag<std::string>("smpi/bcast", "Which collective to use for bcast", "");
 249   simgrid::config::declare_flag<std::string>("smpi/reduce", "Which collective to use for reduce", "");
 250
 251   const char* default_privatization = std::getenv("SMPI_PRIVATIZATION");
 252   if (default_privatization == nullptr)
 253     default_privatization = "no";
 254
 255   simgrid::config::declare_flag<std::string>(
 256       "smpi/privatization", "How we should privatize global variable at runtime (no, yes, mmap, dlopen).",
 257       default_privatization, [](const std::string& smpi_privatize_option) {
 258         if (smpi_privatize_option == "no" || smpi_privatize_option == "0" || smpi_privatize_option == "OFF")
 259           _smpi_cfg_privatization = SmpiPrivStrategies::NONE;
 260         else if (smpi_privatize_option == "yes" || smpi_privatize_option == "1" || smpi_privatize_option == "ON")
 261           _smpi_cfg_privatization = SmpiPrivStrategies::DEFAULT;
 262         else if (smpi_privatize_option == "mmap")
 263           _smpi_cfg_privatization = SmpiPrivStrategies::MMAP;
 264         else if (smpi_privatize_option == "dlopen")
 265           _smpi_cfg_privatization = SmpiPrivStrategies::DLOPEN;
 266         else
 267           xbt_die("Invalid value for smpi/privatization: '%s'", smpi_privatize_option.c_str());
 268
 269         if (not running_with_smpi_main) {
 270           XBT_DEBUG("Running without smpi_main(); disable smpi/privatization.");
 271           _smpi_cfg_privatization = SmpiPrivStrategies::NONE;
 272         }
 273         if (not HAVE_WORKING_MMAP && _smpi_cfg_privatization == SmpiPrivStrategies::MMAP) {
 274           XBT_INFO("mmap privatization is broken on this platform, switching to dlopen privatization instead.");
 275           _smpi_cfg_privatization = SmpiPrivStrategies::DLOPEN;
 276         }
 277       });
 278
 279   simgrid::config::declare_flag<std::string>(
 280       "smpi/privatize-libs",
 281       "Add libraries (; separated) to privatize (libgfortran for example)."
 282       "You need to provide the full names of the files (libgfortran.so.4), or its full path",
 283       "");
 284   simgrid::config::declare_flag<double>("smpi/shared-malloc-blocksize",
 285                                         "Size of the bogus file which will be created for global shared allocations",
 286                                         1UL << 20);
 287   simgrid::config::declare_flag<std::string>("smpi/shared-malloc-hugepage",
 288                                              "Path to a mounted hugetlbfs, to use huge pages with shared malloc.", "");
 289
 290   simgrid::config::declare_flag<std::string>(
 291       "smpi/os", "Small messages timings (MPI_Send minimum time for small messages)", "0:0:0:0:0");
 292   simgrid::config::declare_flag<std::string>(
 293       "smpi/ois", "Small messages timings (MPI_Isend minimum time for small messages)", "0:0:0:0:0");
 294   simgrid::config::declare_flag<std::string>(
 295       "smpi/or", "Small messages timings (MPI_Recv minimum time for small messages)", "0:0:0:0:0");
 296
 297   simgrid::config::declare_flag<bool>("smpi/barrier-finalization", {"smpi/finalization-barrier"},
 298                                       "Do we add a barrier in MPI_Finalize or not", false);
 299   simgrid::config::declare_flag<bool>("smpi/barrier-collectives",
 300                                       "Inject a barrier in each colllective operation, to detect some deadlocks in "
 301                                       "incorrect MPI codes, which may not be triggered in all cases",
 302                                       false);
 303
 304   smpi_options_initialized = true;
 305 }
 306
 307 void smpi_check_options()
 308 {
 309   if (MC_is_active() || MC_record_replay_is_active()) {
 310     if (_sg_mc_buffering == "zero")
 311       simgrid::config::set_value<int>("smpi/send-is-detached-thresh", 0);
 312     else if (_sg_mc_buffering == "infty")
 313       simgrid::config::set_value<int>("smpi/send-is-detached-thresh", INT_MAX);
 314     else
 315       THROW_IMPOSSIBLE;
 316   }
 317
 318   xbt_assert(smpi_cfg_async_small_thresh() <= smpi_cfg_detached_send_thresh(),
 319              "smpi/async-small-thresh (=%d) should be smaller or equal to smpi/send-is-detached-thresh (=%d)",
 320              smpi_cfg_async_small_thresh(),
 321              smpi_cfg_detached_send_thresh());
 322
 323   if (simgrid::config::is_default("smpi/host-speed") && not MC_is_active()) {
 324     XBT_INFO("You did not set the power of the host running the simulation.  "
 325              "The timings will certainly not be accurate.  "
 326              "Use the option \"--cfg=smpi/host-speed:<flops>\" to set its value.  "
 327              "Check "
 328              "https://simgrid.org/doc/latest/Configuring_SimGrid.html#automatic-benchmarking-of-smpi-code for more "
 329              "information.");
 330   }
 331
 332   simgrid::smpi::colls::set_collectives();
 333   simgrid::smpi::colls::smpi_coll_cleanup_callback = nullptr;
 334 }