1 /* Copyright (c) 2004-2022. 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 "simgrid/s4u/Barrier.hpp"
11 #include "smpi_comm.hpp"
14 XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(smpi);
20 static int universe_size = 0;
24 explicit Instance(int max_no_processes) : size_(max_no_processes)
26 auto* group = new simgrid::smpi::Group(size_);
27 comm_world_ = new simgrid::smpi::Comm(group, nullptr, false, -1);
28 universe_size += max_no_processes;
29 bar_ = s4u::Barrier::create(size_);
33 unsigned int finalized_ranks_ = 0;
40 using simgrid::smpi::app::Instance;
42 static std::map<std::string, Instance, std::less<>> smpi_instances;
44 /** @ingroup smpi_simulation
45 * @brief Registers a running instance of an MPI program.
47 * @param name the reference name of the function.
48 * @param code either the main mpi function
49 * (must have a int ..(int argc, char *argv[]) prototype) or nullptr
50 * (if the function deployment is managed somewhere else —
51 * e.g., when deploying manually or using smpirun)
52 * @param num_processes the size of the instance we want to deploy
54 void SMPI_app_instance_register(const char *name, xbt_main_func_t code, int num_processes)
56 if (code != nullptr) // When started with smpirun, we will not execute a function
57 simgrid::s4u::Engine::get_instance()->register_function(name, code);
59 smpi_instances.try_emplace(name, Instance(num_processes));
62 void smpi_deployment_register_process(const std::string& instance_id, int rank, const simgrid::s4u::Actor* actor)
64 const Instance& instance = smpi_instances.at(instance_id);
65 instance.comm_world_->group()->set_mapping(actor->get_pid(), rank);
68 void smpi_deployment_startup_barrier(const std::string& instance_id)
70 const Instance& instance = smpi_instances.at(instance_id);
71 instance.bar_->wait();
74 void smpi_deployment_unregister_process(const std::string& instance_id)
76 Instance& instance = smpi_instances.at(instance_id);
77 instance.finalized_ranks_++;
79 if (instance.finalized_ranks_ == instance.size_) {
80 simgrid::smpi::Comm::destroy(instance.comm_world_);
81 smpi_instances.erase(instance_id);
85 MPI_Comm* smpi_deployment_comm_world(const std::string& instance_id)
88 .empty()) { // no instance registered, we probably used smpirun. (FIXME: I guess this never happens for real)
91 Instance& instance = smpi_instances.at(instance_id);
92 return &instance.comm_world_;
95 void smpi_deployment_cleanup_instances(){
96 for (auto const& item : smpi_instances) {
97 XBT_INFO("Stalling SMPI instance: %s. Do all your MPI ranks call MPI_Finalize()?", item.first.c_str());
98 Instance instance = item.second;
99 simgrid::smpi::Comm::destroy(instance.comm_world_);
101 smpi_instances.clear();
104 int smpi_get_universe_size()
106 return simgrid::smpi::app::universe_size;
109 /** @brief Auxiliary method to get list of hosts to deploy app */
110 static std::vector<simgrid::s4u::Host*> smpi_get_hosts(const simgrid::s4u::Engine* e, const std::string& hostfile)
112 if (hostfile == "") {
113 return e->get_all_hosts();
115 std::vector<simgrid::s4u::Host*> hosts;
116 std::ifstream in(hostfile.c_str());
117 xbt_assert(in, "smpirun: Cannot open the host file: %s", hostfile.c_str());
119 while (std::getline(in, str)) {
121 hosts.emplace_back(e->host_by_name(str));
123 xbt_assert(not hosts.empty(), "smpirun: the hostfile '%s' is empty", hostfile.c_str());
127 /** @brief Read replay configuration from file */
128 static std::vector<std::string> smpi_read_replay(const std::string& replayfile)
130 std::vector<std::string> replay;
131 if (replayfile == "")
134 std::ifstream in(replayfile.c_str());
135 xbt_assert(in, "smpirun: Cannot open the replay file: %s", replayfile.c_str());
137 while (std::getline(in, str)) {
139 replay.emplace_back(str);
145 /** @brief Build argument vector to pass to process */
146 static std::vector<std::string> smpi_deployment_get_args(int rank_id, const std::vector<std::string>& replay,
147 const std::vector<const char*>& run_args)
149 std::vector<std::string> args{std::to_string(rank_id)};
150 // pass arguments to process only if not a replay execution
152 args.insert(args.end(), begin(run_args), end(run_args));
153 /* one trace per process */
154 if (replay.size() > 1)
155 args.emplace_back(replay[rank_id]);
160 * @brief Deploy an SMPI application from a smpirun call
162 * This used to be done at smpirun script, parsing either the hostfile or the platform XML.
163 * If hostfile isn't provided, get the list of hosts from engine.
165 int smpi_deployment_smpirun(const simgrid::s4u::Engine* e, const std::string& hostfile, int np,
166 const std::string& replayfile, int map, const std::vector<const char*>& run_args)
168 auto hosts = smpi_get_hosts(e, hostfile);
169 auto replay = smpi_read_replay(replayfile);
170 int hosts_size = static_cast<int>(hosts.size());
174 xbt_assert(np > 0, "Invalid number of process (np must be > 0). Check your np parameter, platform or hostfile");
176 if (np > hosts_size) {
177 XBT_INFO("You requested to use %d ranks, but there is only %d processes in your hostfile...", np, hosts_size);
180 for (int i = 0; i < np; i++) {
181 simgrid::s4u::Host* host = hosts[i % hosts_size];
182 std::string rank_id = std::to_string(i);
183 auto args = smpi_deployment_get_args(i, replay, run_args);
184 auto actor = simgrid::s4u::Actor::create(rank_id, host, rank_id, args);
185 /* keeping the same behavior as done in smpirun script, print mapping rank/process */
187 XBT_INFO("[rank %d] -> %s", i, host->get_cname());
189 actor->set_property("instance_id", "smpirun");
190 actor->set_property("rank", rank_id);
191 if (not replay.empty())
192 actor->set_property("smpi_replay", "true");
193 /* shared trace file, set it to rank 0 */
194 if (i == 0 && replay.size() == 1)
195 actor->set_property("tracefile", replay[0]);