Logo AND Algorithmique Numérique Distribuée

Public GIT Repository
sanitize the OOP of kernel::profile
[simgrid.git] / src / smpi / internals / smpi_replay.cpp
index c28efd1..6d00532 100644 (file)
@@ -1,4 +1,4 @@
-/* Copyright (c) 2009-2018. The SimGrid Team. All rights reserved.          */
+/* Copyright (c) 2009-2019. 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. */
@@ -64,8 +64,7 @@ public:
 typedef std::tuple</*sender*/ int, /* reciever */ int, /* tag */int> req_key_t;
 typedef std::unordered_map<req_key_t, MPI_Request, hash_tuple::hash<std::tuple<int,int,int>>> req_storage_t;
 
-
-static void log_timed_action(simgrid::xbt::ReplayAction& action, double clock)
+void log_timed_action(simgrid::xbt::ReplayAction& action, double clock)
 {
   if (XBT_LOG_ISENABLED(smpi_replay, xbt_log_priority_verbose)){
     std::string s = boost::algorithm::join(action, " ");
@@ -135,11 +134,14 @@ 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});
     }
 };
 
-void WaitTestParser::parse(simgrid::xbt::ReplayAction& action, std::string name)
+void WaitTestParser::parse(simgrid::xbt::ReplayAction& action, std::string)
 {
   CHECK_ACTION_PARAMS(action, 3, 0)
   src = std::stoi(action[2]);
@@ -147,7 +149,7 @@ void WaitTestParser::parse(simgrid::xbt::ReplayAction& action, std::string name)
   tag = std::stoi(action[4]);
 }
 
-void SendRecvParser::parse(simgrid::xbt::ReplayAction& action, std::string name)
+void SendRecvParser::parse(simgrid::xbt::ReplayAction& action, std::string)
 {
   CHECK_ACTION_PARAMS(action, 3, 1)
   partner = std::stoi(action[2]);
@@ -157,13 +159,13 @@ void SendRecvParser::parse(simgrid::xbt::ReplayAction& action, std::string name)
     datatype1 = simgrid::smpi::Datatype::decode(action[5]);
 }
 
-void ComputeParser::parse(simgrid::xbt::ReplayAction& action, std::string name)
+void ComputeParser::parse(simgrid::xbt::ReplayAction& action, std::string)
 {
   CHECK_ACTION_PARAMS(action, 1, 0)
   flops = parse_double(action[2]);
 }
 
-void BcastArgParser::parse(simgrid::xbt::ReplayAction& action, std::string name)
+void BcastArgParser::parse(simgrid::xbt::ReplayAction& action, std::string)
 {
   CHECK_ACTION_PARAMS(action, 1, 2)
   size = parse_double(action[2]);
@@ -172,7 +174,7 @@ void BcastArgParser::parse(simgrid::xbt::ReplayAction& action, std::string name)
     datatype1 = simgrid::smpi::Datatype::decode(action[4]);
 }
 
-void ReduceArgParser::parse(simgrid::xbt::ReplayAction& action, std::string name)
+void ReduceArgParser::parse(simgrid::xbt::ReplayAction& action, std::string)
 {
   CHECK_ACTION_PARAMS(action, 2, 2)
   comm_size = parse_double(action[2]);
@@ -182,7 +184,7 @@ void ReduceArgParser::parse(simgrid::xbt::ReplayAction& action, std::string name
     datatype1 = simgrid::smpi::Datatype::decode(action[5]);
 }
 
-void AllReduceArgParser::parse(simgrid::xbt::ReplayAction& action, std::string name)
+void AllReduceArgParser::parse(simgrid::xbt::ReplayAction& action, std::string)
 {
   CHECK_ACTION_PARAMS(action, 2, 1)
   comm_size = parse_double(action[2]);
@@ -191,7 +193,7 @@ void AllReduceArgParser::parse(simgrid::xbt::ReplayAction& action, std::string n
     datatype1 = simgrid::smpi::Datatype::decode(action[4]);
 }
 
-void AllToAllArgParser::parse(simgrid::xbt::ReplayAction& action, std::string name)
+void AllToAllArgParser::parse(simgrid::xbt::ReplayAction& action, std::string)
 {
   CHECK_ACTION_PARAMS(action, 2, 1)
   comm_size = MPI_COMM_WORLD->size();
@@ -251,7 +253,7 @@ void GatherVArgParser::parse(simgrid::xbt::ReplayAction& action, std::string nam
   disps      = std::vector<int>(comm_size, 0);
   recvcounts = std::shared_ptr<std::vector<int>>(new std::vector<int>(comm_size));
 
-  if (name == "gatherV") {
+  if (name == "gatherv") {
     root = (action.size() > 3 + comm_size) ? std::stoi(action[3 + comm_size]) : 0;
     if (action.size() > 4 + comm_size)
       datatype1 = simgrid::smpi::Datatype::decode(action[4 + comm_size]);
@@ -289,7 +291,7 @@ void GatherVArgParser::parse(simgrid::xbt::ReplayAction& action, std::string nam
   recv_size_sum = std::accumulate(recvcounts->begin(), recvcounts->end(), 0);
 }
 
-void ScatterArgParser::parse(simgrid::xbt::ReplayAction& action, std::string name)
+void ScatterArgParser::parse(simgrid::xbt::ReplayAction& action, std::string)
 {
   /* The structure of the scatter action for the rank 0 (total 4 processes) is the following:
         0 gather 68 68 0 0 0
@@ -311,7 +313,7 @@ void ScatterArgParser::parse(simgrid::xbt::ReplayAction& action, std::string nam
     datatype2 = simgrid::smpi::Datatype::decode(action[6]);
 }
 
-void ScatterVArgParser::parse(simgrid::xbt::ReplayAction& action, std::string name)
+void ScatterVArgParser::parse(simgrid::xbt::ReplayAction& action, std::string)
 {
   /* The structure of the scatterv action for the rank 0 (total 4 processes) is the following:
      0 gather 68 10 10 10 68 0 0 0
@@ -339,10 +341,10 @@ void ScatterVArgParser::parse(simgrid::xbt::ReplayAction& action, std::string na
   root          = (action.size() > 3 + comm_size) ? std::stoi(action[3 + comm_size]) : 0;
 }
 
-void ReduceScatterArgParser::parse(simgrid::xbt::ReplayAction& action, std::string name)
+void ReduceScatterArgParser::parse(simgrid::xbt::ReplayAction& action, std::string)
 {
   /* The structure of the reducescatter action for the rank 0 (total 4 processes) is the following:
-       0 reduceScatter 275427 275427 275427 204020 11346849 0
+       0 reducescatter 275427 275427 275427 204020 11346849 0
      where:
        1) The first four values after the name of the action declare the recvcounts array
        2) The value 11346849 is the amount of instructions
@@ -361,10 +363,10 @@ void ReduceScatterArgParser::parse(simgrid::xbt::ReplayAction& action, std::stri
   recv_size_sum = std::accumulate(recvcounts->begin(), recvcounts->end(), 0);
 }
 
-void AllToAllVArgParser::parse(simgrid::xbt::ReplayAction& action, std::string name)
+void AllToAllVArgParser::parse(simgrid::xbt::ReplayAction& action, std::string)
 {
-  /* The structure of the allToAllV action for the rank 0 (total 4 processes) is the following:
-        0 allToAllV 100 1 7 10 12 100 1 70 10 5
+  /* The structure of the alltoallv action for the rank 0 (total 4 processes) is the following:
+        0 alltoallv 100 1 7 10 12 100 1 70 10 5
      where:
       1) 100 is the size of the send buffer *sizeof(int),
       2) 1 7 10 12 is the sendcounts array
@@ -393,17 +395,6 @@ void AllToAllVArgParser::parse(simgrid::xbt::ReplayAction& action, std::string n
   recv_size_sum = std::accumulate(recvcounts->begin(), recvcounts->end(), 0);
 }
 
-template<class T>
-void ReplayAction<T>::execute(simgrid::xbt::ReplayAction& action)
-{
-  // Needs to be re-initialized for every action, hence here
-  double start_time = smpi_process()->simulated_elapsed();
-  args.parse(action, name);
-  kernel(action);
-  if (name != "Init")
-    log_timed_action(action, start_time);
-}
-
 void WaitAction::kernel(simgrid::xbt::ReplayAction& action)
 {
   std::string s = boost::algorithm::join(action, " ");
@@ -433,7 +424,7 @@ void WaitAction::kernel(simgrid::xbt::ReplayAction& action)
     TRACE_smpi_recv(args.src, args.dst, args.tag);
 }
 
-void SendAction::kernel(simgrid::xbt::ReplayAction& action)
+void SendAction::kernel(simgrid::xbt::ReplayAction&)
 {
   int dst_traced = MPI_COMM_WORLD->group()->actor(args.partner)->get_pid();
 
@@ -444,7 +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 {
@@ -454,10 +445,8 @@ void SendAction::kernel(simgrid::xbt::ReplayAction& action)
   TRACE_smpi_comm_out(my_proc_id);
 }
 
-void RecvAction::kernel(simgrid::xbt::ReplayAction& action)
+void RecvAction::kernel(simgrid::xbt::ReplayAction&)
 {
-  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)));
 
@@ -468,28 +457,32 @@ void RecvAction::kernel(simgrid::xbt::ReplayAction& action)
     args.size = status.count;
   }
 
+  bool is_recv = false; // Help analyzers understanding that status is not used unintialized
   if (name == "recv") {
+    is_recv = true;
     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);
+  } else {
+    THROW_IMPOSSIBLE;
   }
 
   TRACE_smpi_comm_out(my_proc_id);
-  // 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()) {
+  if (is_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);
   }
 }
 
-void ComputeAction::kernel(simgrid::xbt::ReplayAction& action)
+void ComputeAction::kernel(simgrid::xbt::ReplayAction&)
 {
   TRACE_smpi_computing_in(my_proc_id, args.flops);
   smpi_execute_flops(args.flops);
   TRACE_smpi_computing_out(my_proc_id);
 }
 
-void TestAction::kernel(simgrid::xbt::ReplayAction& action)
+void TestAction::kernel(simgrid::xbt::ReplayAction&)
 {
   MPI_Request request = req_storage.find(args.src, args.dst, args.tag);
   req_storage.remove(request);
@@ -524,17 +517,17 @@ void InitAction::kernel(simgrid::xbt::ReplayAction& action)
   smpi_process()->simulated_start();
 }
 
-void CommunicatorAction::kernel(simgrid::xbt::ReplayAction& action)
+void CommunicatorAction::kernel(simgrid::xbt::ReplayAction&)
 {
   /* nothing to do */
 }
 
-void WaitAllAction::kernel(simgrid::xbt::ReplayAction& action)
+void WaitAllAction::kernel(simgrid::xbt::ReplayAction&)
 {
   const unsigned int count_requests = req_storage.size();
 
   if (count_requests > 0) {
-    TRACE_smpi_comm_in(my_proc_id, __func__, new simgrid::instr::Pt2PtTIData("waitAll", -1, count_requests, ""));
+    TRACE_smpi_comm_in(my_proc_id, __func__, new simgrid::instr::Pt2PtTIData("waitall", -1, count_requests, ""));
     std::vector<std::pair</*sender*/int,/*recv*/int>> sender_receiver;
     std::vector<MPI_Request> reqs;
     req_storage.get_requests(reqs);
@@ -554,14 +547,14 @@ void WaitAllAction::kernel(simgrid::xbt::ReplayAction& action)
   }
 }
 
-void BarrierAction::kernel(simgrid::xbt::ReplayAction& action)
+void BarrierAction::kernel(simgrid::xbt::ReplayAction&)
 {
   TRACE_smpi_comm_in(my_proc_id, __func__, new simgrid::instr::NoOpTIData("barrier"));
   Colls::barrier(MPI_COMM_WORLD);
   TRACE_smpi_comm_out(my_proc_id);
 }
 
-void BcastAction::kernel(simgrid::xbt::ReplayAction& action)
+void BcastAction::kernel(simgrid::xbt::ReplayAction&)
 {
   TRACE_smpi_comm_in(my_proc_id, "action_bcast",
       new simgrid::instr::CollTIData("bcast", MPI_COMM_WORLD->group()->actor(args.root)->get_pid(),
@@ -572,7 +565,7 @@ void BcastAction::kernel(simgrid::xbt::ReplayAction& action)
   TRACE_smpi_comm_out(my_proc_id);
 }
 
-void ReduceAction::kernel(simgrid::xbt::ReplayAction& action)
+void ReduceAction::kernel(simgrid::xbt::ReplayAction&)
 {
   TRACE_smpi_comm_in(my_proc_id, "action_reduce",
       new simgrid::instr::CollTIData("reduce", MPI_COMM_WORLD->group()->actor(args.root)->get_pid(),
@@ -586,9 +579,9 @@ void ReduceAction::kernel(simgrid::xbt::ReplayAction& action)
   TRACE_smpi_comm_out(my_proc_id);
 }
 
-void AllReduceAction::kernel(simgrid::xbt::ReplayAction& action)
+void AllReduceAction::kernel(simgrid::xbt::ReplayAction&)
 {
-  TRACE_smpi_comm_in(my_proc_id, "action_allReduce", new simgrid::instr::CollTIData("allReduce", -1, args.comp_size, args.comm_size, -1,
+  TRACE_smpi_comm_in(my_proc_id, "action_allreduce", new simgrid::instr::CollTIData("allreduce", -1, args.comp_size, args.comm_size, -1,
         Datatype::encode(args.datatype1), ""));
 
   Colls::allreduce(send_buffer(args.comm_size * args.datatype1->size()),
@@ -598,10 +591,10 @@ void AllReduceAction::kernel(simgrid::xbt::ReplayAction& action)
   TRACE_smpi_comm_out(my_proc_id);
 }
 
-void AllToAllAction::kernel(simgrid::xbt::ReplayAction& action)
+void AllToAllAction::kernel(simgrid::xbt::ReplayAction&)
 {
-  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)));
 
@@ -612,7 +605,7 @@ void AllToAllAction::kernel(simgrid::xbt::ReplayAction& action)
   TRACE_smpi_comm_out(my_proc_id);
 }
 
-void GatherAction::kernel(simgrid::xbt::ReplayAction& action)
+void GatherAction::kernel(simgrid::xbt::ReplayAction&)
 {
   TRACE_smpi_comm_in(my_proc_id, name.c_str(), new simgrid::instr::CollTIData(name, (name == "gather") ? args.root : -1, -1.0, args.send_size, args.recv_size,
         Datatype::encode(args.datatype1), Datatype::encode(args.datatype2)));
@@ -629,15 +622,15 @@ void GatherAction::kernel(simgrid::xbt::ReplayAction& action)
   TRACE_smpi_comm_out(my_proc_id);
 }
 
-void GatherVAction::kernel(simgrid::xbt::ReplayAction& action)
+void GatherVAction::kernel(simgrid::xbt::ReplayAction&)
 {
   int rank = MPI_COMM_WORLD->rank();
 
   TRACE_smpi_comm_in(my_proc_id, name.c_str(), new simgrid::instr::VarCollTIData(
-        name, (name == "gatherV") ? args.root : -1, args.send_size, nullptr, -1, args.recvcounts,
+        name, (name == "gatherv") ? args.root : -1, args.send_size, nullptr, -1, args.recvcounts,
         Datatype::encode(args.datatype1), Datatype::encode(args.datatype2)));
 
-  if (name == "gatherV") {
+  if (name == "gatherv") {
     Colls::gatherv(send_buffer(args.send_size * args.datatype1->size()), args.send_size, args.datatype1,
         (rank == args.root) ? recv_buffer(args.recv_size_sum * args.datatype2->size()) : nullptr,
         args.recvcounts->data(), args.disps.data(), args.datatype2, args.root, MPI_COMM_WORLD);
@@ -651,7 +644,7 @@ void GatherVAction::kernel(simgrid::xbt::ReplayAction& action)
   TRACE_smpi_comm_out(my_proc_id);
 }
 
-void ScatterAction::kernel(simgrid::xbt::ReplayAction& action)
+void ScatterAction::kernel(simgrid::xbt::ReplayAction&)
 {
   int rank = MPI_COMM_WORLD->rank();
   TRACE_smpi_comm_in(my_proc_id, "action_scatter", new simgrid::instr::CollTIData(name, args.root, -1.0, args.send_size, args.recv_size,
@@ -664,7 +657,7 @@ void ScatterAction::kernel(simgrid::xbt::ReplayAction& action)
   TRACE_smpi_comm_out(my_proc_id);
 }
 
-void ScatterVAction::kernel(simgrid::xbt::ReplayAction& action)
+void ScatterVAction::kernel(simgrid::xbt::ReplayAction&)
 {
   int rank = MPI_COMM_WORLD->rank();
   TRACE_smpi_comm_in(my_proc_id, "action_scatterv", new simgrid::instr::VarCollTIData(name, args.root, -1, args.sendcounts, args.recv_size,
@@ -679,10 +672,10 @@ void ScatterVAction::kernel(simgrid::xbt::ReplayAction& action)
   TRACE_smpi_comm_out(my_proc_id);
 }
 
-void ReduceScatterAction::kernel(simgrid::xbt::ReplayAction& action)
+void ReduceScatterAction::kernel(simgrid::xbt::ReplayAction&)
 {
   TRACE_smpi_comm_in(my_proc_id, "action_reducescatter",
-      new simgrid::instr::VarCollTIData("reduceScatter", -1, 0, nullptr, -1, args.recvcounts,
+      new simgrid::instr::VarCollTIData("reducescatter", -1, 0, nullptr, -1, args.recvcounts,
         std::to_string(args.comp_size), /* ugly hack to print comp_size */
         Datatype::encode(args.datatype1)));
 
@@ -694,11 +687,11 @@ void ReduceScatterAction::kernel(simgrid::xbt::ReplayAction& action)
   TRACE_smpi_comm_out(my_proc_id);
 }
 
-void AllToAllVAction::kernel(simgrid::xbt::ReplayAction& action)
+void AllToAllVAction::kernel(simgrid::xbt::ReplayAction&)
 {
   TRACE_smpi_comm_in(my_proc_id, __func__,
       new simgrid::instr::VarCollTIData(
-        "allToAllV", -1, args.send_size_sum, args.sendcounts, args.recv_size_sum, args.recvcounts,
+        "alltoallv", -1, args.send_size_sum, args.sendcounts, args.recv_size_sum, args.recvcounts,
         Datatype::encode(args.datatype1), Datatype::encode(args.datatype2)));
 
   Colls::alltoallv(send_buffer(args.send_buf_size * args.datatype1->size()), args.sendcounts->data(), args.senddisps.data(), args.datatype1,
@@ -709,77 +702,80 @@ 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);
   TRACE_smpi_comm_in(my_proc_id, "smpi_replay_run_init", new simgrid::instr::NoOpTIData("init"));
   TRACE_smpi_comm_out(my_proc_id);
   xbt_replay_action_register("init", [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::InitAction().execute(action); });
-  xbt_replay_action_register("finalize", [](simgrid::xbt::ReplayAction& action) { /* nothing to do */ });
+  xbt_replay_action_register("finalize", [](simgrid::xbt::ReplayAction&) { /* nothing to do */ });
   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("allToAllV", [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::AllToAllVAction().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("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); });
-  xbt_replay_action_register("gatherV",  [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::GatherVAction("gatherV").execute(action); });
-  xbt_replay_action_register("scatterV", [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::ScatterVAction().execute(action); });
-  xbt_replay_action_register("allGather", [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::GatherAction("allGather").execute(action); });
-  xbt_replay_action_register("allGatherV", [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::GatherVAction("allGatherV").execute(action); });
-  xbt_replay_action_register("reduceScatter", [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::ReduceScatterAction().execute(action); });
+  xbt_replay_action_register("gatherv",  [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::GatherVAction("gatherv").execute(action); });
+  xbt_replay_action_register("scatterv", [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::ScatterVAction().execute(action); });
+  xbt_replay_action_register("allgather", [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::GatherAction("allgather").execute(action); });
+  xbt_replay_action_register("allgatherv", [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::GatherVAction("allgatherv").execute(action); });
+  xbt_replay_action_register("reducescatter", [](simgrid::xbt::ReplayAction& action) { simgrid::smpi::replay::ReduceScatterAction().execute(action); });
   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++;
     }
@@ -803,8 +799,8 @@ 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);
 }