-/* Copyright (c) 2004-2017. The SimGrid Team.
+/* Copyright (c) 2004-2019. The SimGrid Team.
* All rights reserved. */
/* This program is free software; you can redistribute it and/or modify it
* under the terms of the license (GNU LGPL) which comes with this package. */
-#include "SmpiHost.hpp"
+#include "smpi_host.hpp"
#include "private.hpp"
-#include "simgrid/msg.h" /* barrier */
+#include "simgrid/s4u/Engine.hpp"
#include "smpi_comm.hpp"
#include <map>
+XBT_LOG_EXTERNAL_CATEGORY(smpi);
+
namespace simgrid {
namespace smpi {
namespace app {
+static int universe_size = 0;
+
class Instance {
public:
- Instance(const char* name, int max_no_processes, int process_count, MPI_Comm comm, msg_bar_t finalization_barrier)
- : name(name)
- , size(max_no_processes)
- , present_processes(0)
- , index(process_count)
- , comm_world(comm)
- , finalization_barrier(finalization_barrier)
+ Instance(const std::string& name, int max_no_processes, MPI_Comm comm)
+ : name_(name), size_(max_no_processes), comm_world_(comm)
{
+ MPI_Group group = new simgrid::smpi::Group(size_);
+ comm_world_ = new simgrid::smpi::Comm(group, nullptr, 0, -1);
+ // FIXME : using MPI_Attr_put with MPI_UNIVERSE_SIZE is forbidden and we make it a no-op (which triggers a warning
+ // as MPI_ERR_ARG is returned). Directly calling Comm::attr_put breaks for now, as MPI_UNIVERSE_SIZE,is <0
+ // instance.comm_world->attr_put<simgrid::smpi::Comm>(MPI_UNIVERSE_SIZE, reinterpret_cast<void*>(instance.size));
+
+ universe_size += max_no_processes;
}
- const char* name;
- int size;
- int present_processes;
- int index; // Badly named. This should be "no_processes_when_registering" ;)
- MPI_Comm comm_world;
- msg_bar_t finalization_barrier;
+ const std::string name_;
+ int size_;
+ std::vector<simgrid::s4u::ActorPtr> present_processes_;
+ unsigned int finalized_ranks_ = 0;
+ MPI_Comm comm_world_;
};
}
}
-namespace s4u {
-extern std::map<std::string, simgrid::s4u::Host*> host_list;
-}
}
using simgrid::smpi::app::Instance;
static std::map<std::string, Instance> smpi_instances;
-extern int process_count; // How many processes have been allocated over all instances?
-extern int* index_to_process_data;
-/** \ingroup smpi_simulation
- * \brief Registers a running instance of a MPI program.
+/** @ingroup smpi_simulation
+ * @brief Registers a running instance of a MPI program.
*
- * FIXME : remove MSG from the loop at some point.
- * \param name the reference name of the function.
- * \param code the main mpi function (must have a int ..(int argc, char *argv[])) prototype
- * \param num_processes the size of the instance we want to deploy
+ * @param name the reference name of the function.
+ * @param code either the main mpi function
+ * (must have a int ..(int argc, char *argv[]) prototype) or nullptr
+ * (if the function deployment is managed somewhere else —
+ * e.g., when deploying manually or using smpirun)
+ * @param num_processes the size of the instance we want to deploy
*/
void SMPI_app_instance_register(const char *name, xbt_main_func_t code, int num_processes)
{
- SIMIX_function_register(name, code);
+ if (code != nullptr) // When started with smpirun, we will not execute a function
+ simgrid::s4u::Engine::get_instance()->register_function(name, code);
- static int already_called = 0;
+ static bool already_called = false;
if (not already_called) {
- already_called = 1;
- for (auto const& item : simgrid::s4u::host_list) {
- simgrid::s4u::Host* host = item.second;
- host->extension_set(new simgrid::smpi::SmpiHost(host));
- }
+ already_called = true;
+ for (auto const& host : simgrid::s4u::Engine::get_instance()->get_all_hosts())
+ host->extension_set(new simgrid::smpi::Host(host));
}
- Instance instance(name, num_processes, process_count, MPI_COMM_NULL, MSG_barrier_init(num_processes));
-
- process_count+=num_processes;
+ Instance instance(std::string(name), num_processes, MPI_COMM_NULL);
smpi_instances.insert(std::pair<std::string, Instance>(name, instance));
}
-//get the index of the process in the process_data array
-void smpi_deployment_register_process(const char* instance_id, int rank, int index)
+void smpi_deployment_register_process(const std::string& instance_id, int rank, simgrid::s4u::ActorPtr actor)
{
- if (smpi_instances.empty()) { // no instance registered, we probably used smpirun.
- index_to_process_data[index]=index;
- return;
- }
-
Instance& instance = smpi_instances.at(instance_id);
- if (instance.comm_world == MPI_COMM_NULL) {
- MPI_Group group = new simgrid::smpi::Group(instance.size);
- instance.comm_world = new simgrid::smpi::Comm(group, nullptr);
- }
- instance.present_processes++;
- index_to_process_data[index] = instance.index + rank;
- instance.comm_world->group()->set_mapping(index, rank);
+ instance.present_processes_.push_back(actor);
+ instance.comm_world_->group()->set_mapping(actor, rank);
}
-//get the index of the process in the process_data array
-MPI_Comm* smpi_deployment_comm_world(const char* instance_id)
+void smpi_deployment_unregister_process(const std::string& instance_id)
{
- if (smpi_instances.empty()) { // no instance registered, we probably used smpirun.
- return nullptr;
- }
Instance& instance = smpi_instances.at(instance_id);
- return &instance.comm_world;
+
+ instance.finalized_ranks_++;
+ if (instance.finalized_ranks_ == instance.present_processes_.size()) {
+ instance.present_processes_.clear();
+ simgrid::smpi::Comm::destroy(instance.comm_world_);
+ smpi_instances.erase(instance_id);
+ }
}
-msg_bar_t smpi_deployment_finalization_barrier(const char* instance_id)
+MPI_Comm* smpi_deployment_comm_world(const std::string& instance_id)
{
if (smpi_instances.empty()) { // no instance registered, we probably used smpirun.
return nullptr;
}
Instance& instance = smpi_instances.at(instance_id);
- return instance.finalization_barrier;
+ return &instance.comm_world_;
}
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());
Instance instance = item.second;
- if (instance.comm_world != MPI_COMM_NULL)
- delete instance.comm_world->group();
- delete instance.comm_world;
- MSG_barrier_destroy(instance.finalization_barrier);
+ instance.present_processes_.clear();
+ simgrid::smpi::Comm::destroy(instance.comm_world_);
}
+ smpi_instances.clear();
+}
+
+int smpi_get_universe_size()
+{
+ return simgrid::smpi::app::universe_size;
}
