1 /* Copyright (c) 2004-2021. The SimGrid Team.
2 * All rights reserved. */
4 /* This program is free software; you can redistribute it and/or modify it
5 * under the terms of the license (GNU LGPL) which comes with this package. */
7 #include "smpi_host.hpp"
9 #include "simgrid/s4u/Engine.hpp"
10 #include "smpi_comm.hpp"
13 XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(smpi);
19 static int universe_size = 0;
23 Instance(int max_no_processes, MPI_Comm comm) : size_(max_no_processes), comm_world_(comm)
25 auto* group = new simgrid::smpi::Group(size_);
26 comm_world_ = new simgrid::smpi::Comm(group, nullptr, false, -1);
27 // FIXME : using MPI_Attr_put with MPI_UNIVERSE_SIZE is forbidden and we make it a no-op (which triggers a warning
28 // as MPI_ERR_ARG is returned). Directly calling Comm::attr_put breaks for now, as MPI_UNIVERSE_SIZE,is <0
29 // instance.comm_world->attr_put<simgrid::smpi::Comm>(MPI_UNIVERSE_SIZE, reinterpret_cast<void*>(instance.size));
31 universe_size += max_no_processes;
35 unsigned int finalized_ranks_ = 0;
42 using simgrid::smpi::app::Instance;
44 static std::map<std::string, Instance, std::less<>> smpi_instances;
46 /** @ingroup smpi_simulation
47 * @brief Registers a running instance of an MPI program.
49 * @param name the reference name of the function.
50 * @param code either the main mpi function
51 * (must have a int ..(int argc, char *argv[]) prototype) or nullptr
52 * (if the function deployment is managed somewhere else —
53 * e.g., when deploying manually or using smpirun)
54 * @param num_processes the size of the instance we want to deploy
56 void SMPI_app_instance_register(const char *name, xbt_main_func_t code, int num_processes)
58 if (code != nullptr) // When started with smpirun, we will not execute a function
59 simgrid::s4u::Engine::get_instance()->register_function(name, code);
61 Instance instance(num_processes, MPI_COMM_NULL);
63 smpi_instances.insert(std::pair<std::string, Instance>(name, instance));
66 void smpi_deployment_register_process(const std::string& instance_id, int rank, const simgrid::s4u::Actor* actor)
68 const Instance& instance = smpi_instances.at(instance_id);
69 instance.comm_world_->group()->set_mapping(actor->get_pid(), rank);
72 void smpi_deployment_unregister_process(const std::string& instance_id)
74 Instance& instance = smpi_instances.at(instance_id);
75 instance.finalized_ranks_++;
77 if (instance.finalized_ranks_ == instance.size_) {
78 simgrid::smpi::Comm::destroy(instance.comm_world_);
79 smpi_instances.erase(instance_id);
83 MPI_Comm* smpi_deployment_comm_world(const std::string& instance_id)
86 .empty()) { // no instance registered, we probably used smpirun. (FIXME: I guess this never happens for real)
89 Instance& instance = smpi_instances.at(instance_id);
90 return &instance.comm_world_;
93 void smpi_deployment_cleanup_instances(){
94 for (auto const& item : smpi_instances) {
95 XBT_INFO("Stalling SMPI instance: %s. Do all your MPI ranks call MPI_Finalize()?", item.first.c_str());
96 Instance instance = item.second;
97 simgrid::smpi::Comm::destroy(instance.comm_world_);
99 smpi_instances.clear();
102 int smpi_get_universe_size()
104 return simgrid::smpi::app::universe_size;
107 /** @brief Auxiliary method to get list of hosts to deploy app */
108 static std::vector<simgrid::s4u::Host*> smpi_get_hosts(simgrid::s4u::Engine* e, const std::string& hostfile)
110 if (hostfile == "") {
111 return e->get_all_hosts();
113 std::vector<simgrid::s4u::Host*> hosts;
114 std::ifstream in(hostfile.c_str());
115 xbt_assert(in, "smpirun: Cannot open the host file: %s", hostfile.c_str());
117 while (std::getline(in, str)) {
119 hosts.emplace_back(e->host_by_name(str));
121 xbt_assert(not hosts.empty(), "smpirun: the hostfile '%s' is empty", hostfile.c_str());
125 /** @brief Read replay configuration from file */
126 static std::vector<std::string> smpi_read_replay(const std::string& replayfile)
128 std::vector<std::string> replay;
129 if (replayfile == "")
132 std::ifstream in(replayfile.c_str());
133 xbt_assert(in, "smpirun: Cannot open the replay file: %s", replayfile.c_str());
135 while (std::getline(in, str)) {
137 replay.emplace_back(str);
143 /** @brief Build argument vector to pass to process */
144 static std::vector<std::string> smpi_deployment_get_args(int rank_id, const std::vector<std::string>& replay, int argc,
147 std::vector<std::string> args{std::to_string(rank_id)};
148 // pass arguments to process only if not a replay execution
149 if (replay.empty()) {
150 for (int i = 0; i < argc; i++) {
151 args.push_back(argv[i]);
154 /* one trace per process */
155 if (replay.size() > 1) {
156 args.push_back(replay[rank_id]);
162 * @brief Deploy an SMPI application from a smpirun call
164 * This used to be done at smpirun script, parsing either the hostfile or the platform XML.
165 * If hostfile isn't provided, get the list of hosts from engine.
167 int smpi_deployment_smpirun(simgrid::s4u::Engine* e, const std::string& hostfile, int np, const std::string& replayfile,
168 int map, int argc, char* argv[])
170 auto hosts = smpi_get_hosts(e, hostfile);
171 auto replay = smpi_read_replay(replayfile);
172 int hosts_size = static_cast<int>(hosts.size());
176 xbt_assert(np > 0, "Invalid number of process (np must be > 0). Check your np parameter, platform or hostfile");
178 if (np > hosts_size) {
179 XBT_INFO("You requested to use %d ranks, but there is only %d processes in your hostfile...", np, hosts_size);
182 for (int i = 0; i < np; i++) {
183 simgrid::s4u::Host* host = hosts[i % hosts_size];
184 std::string rank_id = std::to_string(i);
185 auto args = smpi_deployment_get_args(i, replay, argc, argv);
186 auto actor = simgrid::s4u::Actor::create(rank_id, host, rank_id, args);
187 /* keeping the same behavior as done in smpirun script, print mapping rank/process */
189 XBT_INFO("[rank %d] -> %s", i, host->get_cname());
191 actor->set_property("instance_id", "smpirun");
192 actor->set_property("rank", rank_id);
193 if (not replay.empty())
194 actor->set_property("smpi_replay", "true");
195 /* shared trace file, set it to rank 0 */
196 if (i == 0 && replay.size() == 1)
197 actor->set_property("tracefile", replay[0]);