attempt to fix en issue when timing functions are called before initialization

[simgrid.git] / src / smpi / mpi / smpi_comm.cpp

diff --git a/src/smpi/mpi/smpi_comm.cpp b/src/smpi/mpi/smpi_comm.cpp
index e1cf7e6..1722cec 100644 (file)
--- a/src/smpi/mpi/smpi_comm.cpp
+++ b/src/smpi/mpi/smpi_comm.cpp
@@ -1,4 +1,4 @@
-/* Copyright (c) 2010-2020. The SimGrid Team. All rights reserved.          */
+/* Copyright (c) 2010-2021. 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. */
@@ -16,8 +16,18 @@
 
 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(smpi_comm, smpi, "Logging specific to SMPI (comm)");
 
-simgrid::smpi::Comm mpi_MPI_COMM_UNINITIALIZED;
-MPI_Comm MPI_COMM_UNINITIALIZED=&mpi_MPI_COMM_UNINITIALIZED;
+simgrid::smpi::Comm smpi_MPI_COMM_UNINITIALIZED;
+MPI_Comm MPI_COMM_UNINITIALIZED=&smpi_MPI_COMM_UNINITIALIZED;
+/**
+ * Setting MPI_COMM_WORLD to MPI_COMM_UNINITIALIZED (it's a variable)
+ * is important because the implementation of MPI_Comm checks
+ * "this == MPI_COMM_UNINITIALIZED"? If yes, it uses smpi_process()->comm_world()
+ * instead of "this".
+ * This is basically how we only have one global variable but all processes have
+ * different communicators (the one their SMPI instance uses).
+ *
+ */
+MPI_Comm MPI_COMM_WORLD = MPI_COMM_UNINITIALIZED;
 
 /* Support for cartesian topology was added, but there are 2 other types of topology, graph et dist graph. In order to
  * support them, we have to add a field SMPI_Topo_type, and replace the MPI_Topology field by an union. */
@@ -34,6 +44,8 @@ Comm::Comm(MPI_Group group, MPI_Topology topo, bool smp, int in_id)
   errhandler_->ref();
   //First creation of comm is done before SIMIX_run, so only do comms for others
   if(in_id==MPI_UNDEFINED && smp==0 && this->rank()!=MPI_UNDEFINED ){
+    this->add_f();
+    group->c2f();
     int id;
     if(this->rank()==0){
       static int global_id_ = 0;
@@ -319,41 +331,48 @@ void Comm::unref(Comm* comm){
     return;
   }
   comm->refcount_--;
-  Group::unref(comm->group_);
 
   if(comm->refcount_==0){
+    if(simgrid::smpi::F2C::lookup() != nullptr)
+      F2C::free_f(comm->c2f());
     comm->cleanup_smp();
     comm->cleanup_attr<Comm>();
     if (comm->info_ != MPI_INFO_NULL)
       simgrid::smpi::Info::unref(comm->info_);
-    if (comm->errhandler_ != MPI_ERRHANDLER_NULL)
+    if(comm->errhandlers_!=nullptr){
+      for (int i=0; i<comm->size(); i++)
+       if (comm->errhandlers_[i]!=MPI_ERRHANDLER_NULL)
+          simgrid::smpi::Errhandler::unref(comm->errhandlers_[i]);
+      delete[] comm->errhandlers_;
+    } else if (comm->errhandler_ != MPI_ERRHANDLER_NULL)
       simgrid::smpi::Errhandler::unref(comm->errhandler_);
-    delete comm;
   }
+  Group::unref(comm->group_);
+  if(comm->refcount_==0)
+    delete comm;
 }
 
 MPI_Comm Comm::find_intra_comm(int * leader){
   //get the indices of all processes sharing the same simix host
-  auto& actor_list        = sg_host_self()->pimpl_->actor_list_;
   int intra_comm_size     = 0;
   int min_index           = INT_MAX; // the minimum index will be the leader
-  for (auto& actor : actor_list) {
+  sg_host_self()->pimpl_->foreach_actor([this, &intra_comm_size, &min_index](auto& actor) {
     int index = actor.get_pid();
     if (this->group()->rank(actor.get_ciface()) != MPI_UNDEFINED) { // Is this process in the current group?
       intra_comm_size++;
       if (index < min_index)
         min_index = index;
     }
-  }
+  });
   XBT_DEBUG("number of processes deployed on my node : %d", intra_comm_size);
   auto* group_intra = new Group(intra_comm_size);
   int i = 0;
-  for (auto& actor : actor_list) {
+  sg_host_self()->pimpl_->foreach_actor([this, group_intra, &i](auto& actor) {
     if (this->group()->rank(actor.get_ciface()) != MPI_UNDEFINED) {
       group_intra->set_mapping(actor.get_ciface(), i);
       i++;
     }
-  }
+  });
   *leader=min_index;
   return new Comm(group_intra, nullptr, true);
 }
@@ -549,18 +568,36 @@ void Comm::set_info(MPI_Info info)
 
 MPI_Errhandler Comm::errhandler()
 {
-  if (errhandler_ != MPI_ERRHANDLER_NULL)
-    errhandler_->ref();
-  return errhandler_;
+  if (this != MPI_COMM_WORLD){
+    if (errhandler_ != MPI_ERRHANDLER_NULL)
+      errhandler_->ref();
+    return errhandler_;
+  } else {
+    if(errhandlers_==nullptr)
+      return MPI_ERRORS_ARE_FATAL;
+    else {
+      if(errhandlers_[this->rank()] != MPI_ERRHANDLER_NULL)
+        errhandlers_[this->rank()]->ref();
+      return errhandlers_[this->rank()];
+    }
+  }
 }
 
 void Comm::set_errhandler(MPI_Errhandler errhandler)
 {
-  if (errhandler_ != MPI_ERRHANDLER_NULL)
-    simgrid::smpi::Errhandler::unref(errhandler_);
-  errhandler_ = errhandler;
-  if (errhandler_ != MPI_ERRHANDLER_NULL)
-    errhandler_->ref();
+  if(this != MPI_COMM_WORLD){
+    if (errhandler_ != MPI_ERRHANDLER_NULL)
+      simgrid::smpi::Errhandler::unref(errhandler_);
+    errhandler_ = errhandler;
+  }else{
+    if(errhandlers_==nullptr)
+      errhandlers_= new MPI_Errhandler[this->size()]{MPI_ERRHANDLER_NULL};
+    if(errhandlers_[this->rank()] != MPI_ERRHANDLER_NULL)
+      simgrid::smpi::Errhandler::unref(errhandlers_[this->rank()]);
+    errhandlers_[this->rank()]=errhandler;
+  }
+  if (errhandler != MPI_ERRHANDLER_NULL)
+    errhandler->ref();
 }
 
 MPI_Comm Comm::split_type(int type, int /*key*/, const Info*)