X-Git-Url: http://bilbo.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/8f52d8360b744ceb8c07a8322b1d7a4194ec73e4..c972ddd5a74b83c183357f707e7dfba87d48e750:/src/smpi/internals/smpi_deployment.cpp diff --git a/src/smpi/internals/smpi_deployment.cpp b/src/smpi/internals/smpi_deployment.cpp index 61d6761a13..93c7e61443 100644 --- a/src/smpi/internals/smpi_deployment.cpp +++ b/src/smpi/internals/smpi_deployment.cpp @@ -10,7 +10,7 @@ #include "smpi_comm.hpp" #include -XBT_LOG_EXTERNAL_CATEGORY(smpi); +XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(smpi); namespace simgrid { namespace smpi { @@ -63,7 +63,7 @@ void SMPI_app_instance_register(const char *name, xbt_main_func_t code, int num_ smpi_instances.insert(std::pair(name, instance)); } -void smpi_deployment_register_process(const std::string& instance_id, int rank, simgrid::s4u::Actor* actor) +void smpi_deployment_register_process(const std::string& instance_id, int rank, const simgrid::s4u::Actor* actor) { const Instance& instance = smpi_instances.at(instance_id); instance.comm_world_->group()->set_mapping(actor->get_pid(), rank); @@ -92,7 +92,7 @@ MPI_Comm* smpi_deployment_comm_world(const std::string& instance_id) void smpi_deployment_cleanup_instances(){ for (auto const& item : smpi_instances) { - XBT_CINFO(smpi, "Stalling SMPI instance: %s. Do all your MPI ranks call MPI_Finalize()?", item.first.c_str()); + XBT_INFO("Stalling SMPI instance: %s. Do all your MPI ranks call MPI_Finalize()?", item.first.c_str()); Instance instance = item.second; simgrid::smpi::Comm::destroy(instance.comm_world_); } @@ -103,3 +103,98 @@ int smpi_get_universe_size() { return simgrid::smpi::app::universe_size; } + +/** @brief Auxiliary method to get list of hosts to deploy app */ +static std::vector smpi_get_hosts(simgrid::s4u::Engine* e, const std::string& hostfile) +{ + if (hostfile == "") { + return e->get_all_hosts(); + } + std::vector hosts; + std::ifstream in(hostfile.c_str()); + xbt_assert(in, "smpirun: Cannot open the host file: %s", hostfile.c_str()); + std::string str; + while (std::getline(in, str)) { + if (not str.empty()) + hosts.emplace_back(e->host_by_name(str)); + } + xbt_assert(not hosts.empty(), "smpirun: the hostfile '%s' is empty", hostfile.c_str()); + return hosts; +} + +/** @brief Read replay configuration from file */ +static std::vector smpi_read_replay(const std::string& replayfile) +{ + std::vector replay; + if (replayfile == "") + return replay; + + std::ifstream in(replayfile.c_str()); + xbt_assert(in, "smpirun: Cannot open the replay file: %s", replayfile.c_str()); + std::string str; + while (std::getline(in, str)) { + if (not str.empty()) + replay.emplace_back(str); + } + + return replay; +} + +/** @brief Build argument vector to pass to process */ +static std::vector smpi_deployment_get_args(int rank_id, const std::vector& replay, int argc, + char* argv[]) +{ + std::vector args{std::to_string(rank_id)}; + // pass arguments to process only if not a replay execution + if (replay.empty()) { + for (int i = 0; i < argc; i++) { + args.push_back(argv[i]); + } + } + /* one trace per process */ + if (replay.size() > 1) { + args.push_back(replay[rank_id]); + } + return args; +} + +/** + * @brief Deploy an SMPI application from a smpirun call + * + * This used to be done at smpirun script, parsing either the hostfile or the platform XML. + * If hostfile isn't provided, get the list of hosts from engine. + */ +int smpi_deployment_smpirun(simgrid::s4u::Engine* e, const std::string& hostfile, int np, const std::string& replayfile, + int map, int argc, char* argv[]) +{ + auto hosts = smpi_get_hosts(e, hostfile); + auto replay = smpi_read_replay(replayfile); + int hosts_size = static_cast(hosts.size()); + if (np == 0) + np = hosts_size; + + xbt_assert(np > 0, "Invalid number of process (np must be > 0). Check your np parameter, platform or hostfile"); + + if (np > hosts_size) { + XBT_INFO("You requested to use %d ranks, but there is only %d processes in your hostfile...", np, hosts_size); + } + + for (int i = 0; i < np; i++) { + simgrid::s4u::Host* host = hosts[i % hosts_size]; + std::string rank_id = std::to_string(i); + auto args = smpi_deployment_get_args(i, replay, argc, argv); + auto actor = simgrid::s4u::Actor::create(rank_id, host, rank_id, args); + /* keeping the same behavior as done in smpirun script, print mapping rank/process */ + if (map != 0) { + XBT_INFO("[rank %d] -> %s", i, host->get_cname()); + } + actor->set_property("instance_id", "smpirun"); + actor->set_property("rank", rank_id); + if (not replay.empty()) + actor->set_property("smpi_replay", "true"); + /* shared trace file, set it to rank 0 */ + if (i == 0 && replay.size() == 1) + actor->set_property("tracefile", replay[0]); + } + return np; +}