Logo AND Algorithmique Numérique Distribuée

Public GIT Repository
[smpi args cleanup] internal changes for clean-up
[simgrid.git] / src / smpi / internals / smpi_replay.cpp
index 5d5bf7f..627d061 100644 (file)
@@ -134,7 +134,10 @@ public:
     /* Sometimes we need to re-insert MPI_REQUEST_NULL but we still need src,dst and tag */
     void addNullRequest(int src, int dst, int tag)
     {
-      store.insert({req_key_t(src, dst, tag), MPI_REQUEST_NULL});
+      store.insert({req_key_t(
+            MPI_COMM_WORLD->group()->actor(src)->get_pid()-1,
+            MPI_COMM_WORLD->group()->actor(dst)->get_pid()-1,
+            tag), MPI_REQUEST_NULL});
     }
 };
 
@@ -432,7 +435,7 @@ void SendAction::kernel(simgrid::xbt::ReplayAction& action)
 
   if (name == "send") {
     Request::send(nullptr, args.size, args.datatype1, args.partner, args.tag, MPI_COMM_WORLD);
-  } else if (name == "Isend") {
+  } else if (name == "isend") {
     MPI_Request request = Request::isend(nullptr, args.size, args.datatype1, args.partner, args.tag, MPI_COMM_WORLD);
     req_storage.add(request);
   } else {
@@ -444,8 +447,6 @@ void SendAction::kernel(simgrid::xbt::ReplayAction& action)
 
 void RecvAction::kernel(simgrid::xbt::ReplayAction& action)
 {
-  int src_traced = MPI_COMM_WORLD->group()->actor(args.partner)->get_pid();
-
   TRACE_smpi_comm_in(my_proc_id, __func__, new simgrid::instr::Pt2PtTIData(name, args.partner, args.size,
         args.tag, Datatype::encode(args.datatype1)));
 
@@ -458,14 +459,15 @@ void RecvAction::kernel(simgrid::xbt::ReplayAction& action)
 
   if (name == "recv") {
     Request::recv(nullptr, args.size, args.datatype1, args.partner, args.tag, MPI_COMM_WORLD, &status);
-  } else if (name == "Irecv") {
+  } else if (name == "irecv") {
     MPI_Request request = Request::irecv(nullptr, args.size, args.datatype1, args.partner, args.tag, MPI_COMM_WORLD);
     req_storage.add(request);
   }
 
   TRACE_smpi_comm_out(my_proc_id);
-  // TODO: Check why this was only activated in the "recv" case and not in the "Irecv" case
+  // TODO: Check why this was only activated in the "recv" case and not in the "irecv" case
   if (name == "recv" && not TRACE_smpi_view_internals()) {
+    int src_traced = MPI_COMM_WORLD->group()->actor(status.MPI_SOURCE)->get_pid();
     TRACE_smpi_recv(src_traced, my_proc_id, args.tag);
   }
 }
@@ -588,8 +590,8 @@ void AllReduceAction::kernel(simgrid::xbt::ReplayAction& action)
 
 void AllToAllAction::kernel(simgrid::xbt::ReplayAction& action)
 {
-  TRACE_smpi_comm_in(my_proc_id, "action_allToAll",
-      new simgrid::instr::CollTIData("allToAll", -1, -1.0, args.send_size, args.recv_size,
+  TRACE_smpi_comm_in(my_proc_id, "action_alltoall",
+      new simgrid::instr::CollTIData("alltoall", -1, -1.0, args.send_size, args.recv_size,
         Datatype::encode(args.datatype1),
         Datatype::encode(args.datatype2)));
 
@@ -697,16 +699,19 @@ void AllToAllVAction::kernel(simgrid::xbt::ReplayAction& action)
 } // Replay Namespace
 }} // namespace simgrid::smpi
 
-static std::vector<simgrid::smpi::replay::RequestStorage> storage;
+static std::unordered_map<aid_t, simgrid::smpi::replay::RequestStorage> storage;
 /** @brief Only initialize the replay, don't do it for real */
-void smpi_replay_init(int* argc, char*** argv)
+void smpi_replay_init(const char* instance_id, int rank, double start_delay_flops)
 {
-  simgrid::smpi::Process::init(argc, argv);
+  if (not smpi_process()->initializing()){
+    simgrid::s4u::Actor::self()->set_property("instance_id", instance_id);
+    simgrid::s4u::Actor::self()->set_property("rank", std::to_string(rank));
+    simgrid::smpi::ActorExt::init();
+  }
   smpi_process()->mark_as_initialized();
   smpi_process()->set_replaying(true);
 
   int my_proc_id = simgrid::s4u::this_actor::get_pid();
-  storage.resize(smpi_process_count());
 
   TRACE_smpi_init(my_proc_id);
   TRACE_smpi_computing_init(my_proc_id);
@@ -717,18 +722,18 @@ void smpi_replay_init(int* argc, char*** argv)
   xbt_replay_action_register("comm_size", [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::CommunicatorAction().execute(action); });
   xbt_replay_action_register("comm_split",[](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::CommunicatorAction().execute(action); });
   xbt_replay_action_register("comm_dup",  [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::CommunicatorAction().execute(action); });
-  xbt_replay_action_register("send",  [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::SendAction("send", storage[simgrid::s4u::this_actor::get_pid()-1]).execute(action); });
-  xbt_replay_action_register("Isend", [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::SendAction("Isend", storage[simgrid::s4u::this_actor::get_pid()-1]).execute(action); });
-  xbt_replay_action_register("recv",  [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::RecvAction("recv", storage[simgrid::s4u::this_actor::get_pid()-1]).execute(action); });
-  xbt_replay_action_register("Irecv", [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::RecvAction("Irecv", storage[simgrid::s4u::this_actor::get_pid()-1]).execute(action); });
-  xbt_replay_action_register("test",  [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::TestAction(storage[simgrid::s4u::this_actor::get_pid()-1]).execute(action); });
-  xbt_replay_action_register("wait",  [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::WaitAction(storage[simgrid::s4u::this_actor::get_pid()-1]).execute(action); });
-  xbt_replay_action_register("waitall", [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::WaitAllAction(storage[simgrid::s4u::this_actor::get_pid()-1]).execute(action); });
+  xbt_replay_action_register("send",  [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::SendAction("send", storage[simgrid::s4u::this_actor::get_pid()]).execute(action); });
+  xbt_replay_action_register("isend", [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::SendAction("isend", storage[simgrid::s4u::this_actor::get_pid()]).execute(action); });
+  xbt_replay_action_register("recv",  [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::RecvAction("recv", storage[simgrid::s4u::this_actor::get_pid()]).execute(action); });
+  xbt_replay_action_register("irecv", [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::RecvAction("irecv", storage[simgrid::s4u::this_actor::get_pid()]).execute(action); });
+  xbt_replay_action_register("test",  [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::TestAction(storage[simgrid::s4u::this_actor::get_pid()]).execute(action); });
+  xbt_replay_action_register("wait",  [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::WaitAction(storage[simgrid::s4u::this_actor::get_pid()]).execute(action); });
+  xbt_replay_action_register("waitall", [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::WaitAllAction(storage[simgrid::s4u::this_actor::get_pid()]).execute(action); });
   xbt_replay_action_register("barrier", [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::BarrierAction().execute(action); });
   xbt_replay_action_register("bcast",   [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::BcastAction().execute(action); });
   xbt_replay_action_register("reduce",  [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::ReduceAction().execute(action); });
   xbt_replay_action_register("allreduce", [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::AllReduceAction().execute(action); });
-  xbt_replay_action_register("allToAll", [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::AllToAllAction().execute(action); });
+  xbt_replay_action_register("alltoall", [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::AllToAllAction().execute(action); });
   xbt_replay_action_register("alltoallv", [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::AllToAllVAction().execute(action); });
   xbt_replay_action_register("gather",   [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::GatherAction("gather").execute(action); });
   xbt_replay_action_register("scatter",  [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::ScatterAction().execute(action); });
@@ -740,34 +745,34 @@ void smpi_replay_init(int* argc, char*** argv)
   xbt_replay_action_register("compute", [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::ComputeAction().execute(action); });
 
   //if we have a delayed start, sleep here.
-  if(*argc>2){
-    double value = xbt_str_parse_double((*argv)[2], "%s is not a double");
-    XBT_VERB("Delayed start for instance - Sleeping for %f flops ",value );
-    smpi_execute_flops(value);
+  if (start_delay_flops > 0) {
+    XBT_VERB("Delayed start for instance - Sleeping for %f flops ", start_delay_flops);
+    smpi_execute_flops(start_delay_flops);
   } else {
-    //UGLY: force a context switch to be sure that all MSG_processes begin initialization
-    XBT_DEBUG("Force context switch by smpi_execute_flops  - Sleeping for 0.0 flops ");
-    smpi_execute_flops(0.0);
+    // Wait for the other actors to initialize also
+    simgrid::s4u::this_actor::yield();
   }
 }
 
 /** @brief actually run the replay after initialization */
-void smpi_replay_main(int* argc, char*** argv)
+void smpi_replay_main(int rank, const char* trace_filename)
 {
   static int active_processes = 0;
   active_processes++;
-  simgrid::xbt::replay_runner(*argc, *argv);
+  storage[simgrid::s4u::this_actor::get_pid()] = simgrid::smpi::replay::RequestStorage();
+  std::string rank_string                      = std::to_string(rank);
+  simgrid::xbt::replay_runner(rank_string.c_str(), trace_filename);
 
   /* and now, finalize everything */
   /* One active process will stop. Decrease the counter*/
-  unsigned int count_requests = storage[simgrid::s4u::this_actor::get_pid() - 1].size();
+  unsigned int count_requests = storage[simgrid::s4u::this_actor::get_pid()].size();
   XBT_DEBUG("There are %ud elements in reqq[*]", count_requests);
   if (count_requests > 0) {
     MPI_Request requests[count_requests];
     MPI_Status status[count_requests];
     unsigned int i=0;
 
-    for (auto const& pair : storage[simgrid::s4u::this_actor::get_pid() - 1].get_store()) {
+    for (auto const& pair : storage[simgrid::s4u::this_actor::get_pid()].get_store()) {
       requests[i] = pair.second;
       i++;
     }
@@ -791,8 +796,8 @@ void smpi_replay_main(int* argc, char*** argv)
 }
 
 /** @brief chain a replay initialization and a replay start */
-void smpi_replay_run(int* argc, char*** argv)
+void smpi_replay_run(const char* instance_id, int rank, double start_delay_flops, const char* trace_filename)
 {
-  smpi_replay_init(argc, argv);
-  smpi_replay_main(argc, argv);
+  smpi_replay_init(instance_id, rank, start_delay_flops);
+  smpi_replay_main(rank, trace_filename);
 }