[SMPI] Move process_data to map<Actor, smpi::Process>

[simgrid.git] / src / smpi / internals / smpi_global.cpp

diff --git a/src/smpi/internals/smpi_global.cpp b/src/smpi/internals/smpi_global.cpp
index 0a66fcf..fa1605b 100644 (file)
--- a/src/smpi/internals/smpi_global.cpp
+++ b/src/smpi/internals/smpi_global.cpp
@@ -1,4 +1,4 @@
-/* Copyright (c) 2007-2017. The SimGrid Team. All rights reserved.          */
+/* Copyright (c) 2007-2018. 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. */
 
 /* 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. */


@@ -8,6 +8,7 @@
 #include "private.hpp"
 #include "simgrid/s4u/Host.hpp"
 #include "simgrid/s4u/Mailbox.hpp"
 #include "private.hpp"
 #include "simgrid/s4u/Host.hpp"
 #include "simgrid/s4u/Mailbox.hpp"

+#include "simgrid/s4u/forward.hpp"

 #include "smpi_coll.hpp"
 #include "smpi_comm.hpp"
 #include "smpi_group.hpp"
 #include "smpi_coll.hpp"
 #include "smpi_comm.hpp"
 #include "smpi_group.hpp"


@@ -50,12 +51,13 @@ struct papi_process_data {
 };
 
 #endif
 };
 
 #endif

+using simgrid::s4u::Actor;
+using simgrid::s4u::ActorPtr;

 std::unordered_map<std::string, double> location2speedup;
 
 std::unordered_map<std::string, double> location2speedup;
 

-simgrid::smpi::Process **process_data = nullptr;
+static std::map</*process_id*/ ActorPtr, simgrid::smpi::Process*> process_data;

 int process_count = 0;
 int smpi_universe_size = 0;
 int process_count = 0;
 int smpi_universe_size = 0;

-int* index_to_process_data = nullptr;

 extern double smpi_total_benched_time;
 xbt_os_timer_t global_timer;
 /**
 extern double smpi_total_benched_time;
 xbt_os_timer_t global_timer;
 /**


@@ -64,7 +66,7 @@ xbt_os_timer_t global_timer;
  * "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
  * "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 (basically, the one their SMPI instance uses).
+ * different communicators (the one their SMPI instance uses).

  *
  * See smpi_comm.cpp and the functions therein for details.
  */
  *
  * See smpi_comm.cpp and the functions therein for details.
  */


@@ -82,6 +84,12 @@ static simgrid::config::Flag<double> smpi_init_sleep(
 
 void (*smpi_comm_copy_data_callback) (smx_activity_t, void*, size_t) = &smpi_comm_copy_buffer_callback;
 
 
 void (*smpi_comm_copy_data_callback) (smx_activity_t, void*, size_t) = &smpi_comm_copy_buffer_callback;
 

+void smpi_add_process(ActorPtr actor)
+{
+  process_data.insert({actor, new simgrid::smpi::Process(actor, nullptr)});
+  // smpi_deployment_register_process("master_mpi", 0, actor);
+}
+

 int smpi_process_count()
 {
   return process_count;
 int smpi_process_count()
 {
   return process_count;


@@ -89,16 +97,16 @@ int smpi_process_count()
 
 simgrid::smpi::Process* smpi_process()
 {
 
 simgrid::smpi::Process* smpi_process()
 {

-  smx_actor_t me = SIMIX_process_self();
+  ActorPtr me = Actor::self();

   if (me == nullptr) // This happens sometimes (eg, when linking against NS3 because it pulls openMPI...)
     return nullptr;
   if (me == nullptr) // This happens sometimes (eg, when linking against NS3 because it pulls openMPI...)
     return nullptr;

-  simgrid::msg::ActorExt* msgExt = static_cast<simgrid::msg::ActorExt*>(me->userdata);
+  simgrid::msg::ActorExt* msgExt = static_cast<simgrid::msg::ActorExt*>(me->getImpl()->userdata);

   return static_cast<simgrid::smpi::Process*>(msgExt->data);
 }
 
   return static_cast<simgrid::smpi::Process*>(msgExt->data);
 }
 

-simgrid::smpi::Process* smpi_process_remote(int index)
+simgrid::smpi::Process* smpi_process_remote(ActorPtr actor)

 {
 {

-  return process_data[index_to_process_data[index]];
+  return process_data.at(actor);

 }
 
 MPI_Comm smpi_process_comm_self(){
 }
 
 MPI_Comm smpi_process_comm_self(){


@@ -241,7 +249,7 @@ static void smpi_check_options(){
 }
 
 int smpi_enabled() {
 }
 
 int smpi_enabled() {

-  return process_data != nullptr;
+  return not process_data.empty();

 }
 
 void smpi_global_init()
 }
 
 void smpi_global_init()


@@ -339,31 +347,6 @@ void smpi_global_init()
     }
   }
 #endif
     }
   }
 #endif

-
-  if (index_to_process_data == nullptr) {
-    index_to_process_data = new int[SIMIX_process_count()];
-  }
-
-  bool smpirun = 0;
-  if (process_count == 0) { // The program has been dispatched but no other
-                            // SMPI instances have been registered. We're using smpirun.
-    smpirun = true;
-    SMPI_app_instance_register(smpi_default_instance_name, nullptr,
-                               SIMIX_process_count()); // This call has a side effect on process_count...
-    MPI_COMM_WORLD = *smpi_deployment_comm_world(smpi_default_instance_name);
-  }
-  smpi_universe_size = process_count;
-  process_data       = new simgrid::smpi::Process*[process_count];
-  for (int i = 0; i < process_count; i++) {
-    if (smpirun) {
-      process_data[i] = new simgrid::smpi::Process(i, smpi_deployment_finalization_barrier(smpi_default_instance_name));
-      smpi_deployment_register_process(smpi_default_instance_name, i, i);
-    } else {
-      // TODO We can pass a nullptr here because Process::set_data() assigns the
-      // barrier from the instance anyway. This is ugly and should be changed
-      process_data[i] = new simgrid::smpi::Process(i, nullptr);
-    }
-  }

 }
 
 void smpi_global_destroy()
 }
 
 void smpi_global_destroy()


@@ -371,19 +354,18 @@ void smpi_global_destroy()
   smpi_bench_destroy();
   smpi_shared_destroy();
   smpi_deployment_cleanup_instances();
   smpi_bench_destroy();
   smpi_shared_destroy();
   smpi_deployment_cleanup_instances();

-  for (int i = 0, count = smpi_process_count(); i < count; i++) {
-    if(process_data[i]->comm_self()!=MPI_COMM_NULL){
-      simgrid::smpi::Comm::destroy(process_data[i]->comm_self());
+  for (auto& pair : process_data) {
+    auto& process = pair.second;
+    if (process->comm_self() != MPI_COMM_NULL) {
+      simgrid::smpi::Comm::destroy(process->comm_self());

     }
     }

-    if(process_data[i]->comm_intra()!=MPI_COMM_NULL){
-      simgrid::smpi::Comm::destroy(process_data[i]->comm_intra());
+    if (process->comm_intra() != MPI_COMM_NULL) {
+      simgrid::smpi::Comm::destroy(process->comm_intra());

     }
     }

-    xbt_os_timer_free(process_data[i]->timer());
-    xbt_mutex_destroy(process_data[i]->mailboxes_mutex());
-    delete process_data[i];
+    xbt_os_timer_free(process->timer());
+    xbt_mutex_destroy(process->mailboxes_mutex());

   }
   }

-  delete[] process_data;
-  process_data = nullptr;
+  process_data.clear();

 
   if (simgrid::smpi::Colls::smpi_coll_cleanup_callback != nullptr)
     simgrid::smpi::Colls::smpi_coll_cleanup_callback();
 
   if (simgrid::smpi::Colls::smpi_coll_cleanup_callback != nullptr)
     simgrid::smpi::Colls::smpi_coll_cleanup_callback();


@@ -394,43 +376,11 @@ void smpi_global_destroy()
     xbt_os_timer_free(global_timer);
   }
 
     xbt_os_timer_free(global_timer);
   }
 

-  xbt_free(index_to_process_data);

   if(smpi_privatize_global_variables == SMPI_PRIVATIZE_MMAP)
     smpi_destroy_global_memory_segments();
   smpi_free_static();
 }
 
   if(smpi_privatize_global_variables == SMPI_PRIVATIZE_MMAP)
     smpi_destroy_global_memory_segments();
   smpi_free_static();
 }
 

-extern "C" {
-
-static void smpi_init_logs(){
-
-  /* Connect log categories.  See xbt/log.c */
-
-  XBT_LOG_CONNECT(smpi);  /* Keep this line as soon as possible in this function: xbt_log_appender_file.c depends on it
-                             DO NOT connect this in XBT or so, or it will be useless to xbt_log_appender_file.c */
-  XBT_LOG_CONNECT(instr_smpi);
-  XBT_LOG_CONNECT(smpi_bench);
-  XBT_LOG_CONNECT(smpi_coll);
-  XBT_LOG_CONNECT(smpi_colls);
-  XBT_LOG_CONNECT(smpi_comm);
-  XBT_LOG_CONNECT(smpi_datatype);
-  XBT_LOG_CONNECT(smpi_dvfs);
-  XBT_LOG_CONNECT(smpi_group);
-  XBT_LOG_CONNECT(smpi_host);
-  XBT_LOG_CONNECT(smpi_kernel);
-  XBT_LOG_CONNECT(smpi_mpi);
-  XBT_LOG_CONNECT(smpi_memory);
-  XBT_LOG_CONNECT(smpi_op);
-  XBT_LOG_CONNECT(smpi_pmpi);
-  XBT_LOG_CONNECT(smpi_process);
-  XBT_LOG_CONNECT(smpi_request);
-  XBT_LOG_CONNECT(smpi_replay);
-  XBT_LOG_CONNECT(smpi_rma);
-  XBT_LOG_CONNECT(smpi_shared);
-  XBT_LOG_CONNECT(smpi_utils);
-}
-}
-

 static void smpi_init_options(){
   // return if already called
   if (smpi_cpu_threshold > -1)
 static void smpi_init_options(){
   // return if already called
   if (smpi_cpu_threshold > -1)


@@ -461,17 +411,16 @@ static void smpi_init_options(){
     if (smpi_cpu_threshold < 0)
       smpi_cpu_threshold = DBL_MAX;
 
     if (smpi_cpu_threshold < 0)
       smpi_cpu_threshold = DBL_MAX;
 

-    const char* val = xbt_cfg_get_string("smpi/shared-malloc").c_str();
-    if (not strcasecmp(val, "yes") || not strcmp(val, "1") || not strcasecmp(val, "on") ||
-        not strcasecmp(val, "global")) {
+    std::string val = xbt_cfg_get_string("smpi/shared-malloc");
+    if ((val == "yes") || (val == "1") || (val == "on") || (val == "global")) {

       smpi_cfg_shared_malloc = shmalloc_global;
       smpi_cfg_shared_malloc = shmalloc_global;

-    } else if (not strcasecmp(val, "local")) {
+    } else if (val == "local") {

       smpi_cfg_shared_malloc = shmalloc_local;
       smpi_cfg_shared_malloc = shmalloc_local;

-    } else if (not strcasecmp(val, "no") || not strcmp(val, "0") || not strcasecmp(val, "off")) {
+    } else if ((val == "no") || (val == "0") || (val == "off")) {

       smpi_cfg_shared_malloc = shmalloc_none;
     } else {
       xbt_die("Invalid value '%s' for option smpi/shared-malloc. Possible values: 'on' or 'global', 'local', 'off'",
       smpi_cfg_shared_malloc = shmalloc_none;
     } else {
       xbt_die("Invalid value '%s' for option smpi/shared-malloc. Possible values: 'on' or 'global', 'local', 'off'",

-              val);
+              val.c_str());

     }
 }
 
     }
 }
 


@@ -533,6 +482,9 @@ int smpi_main(const char* executable, int argc, char *argv[])
 
   SMPI_switch_data_segment = &smpi_switch_data_segment;
 
 
   SMPI_switch_data_segment = &smpi_switch_data_segment;
 

+  // TODO This will not be executed in the case where smpi_main is not called,
+  // e.g., not for smpi_msg_masterslave. This should be moved to another location
+  // that is always called -- maybe close to Actor::onCreation?

   simgrid::s4u::Host::onCreation.connect([](simgrid::s4u::Host& host) {
     host.extension_set(new simgrid::smpi::SmpiHost(&host));
   });
   simgrid::s4u::Host::onCreation.connect([](simgrid::s4u::Host& host) {
     host.extension_set(new simgrid::smpi::SmpiHost(&host));
   });


@@ -542,7 +494,6 @@ int smpi_main(const char* executable, int argc, char *argv[])
   SIMIX_comm_set_copy_data_callback(smpi_comm_copy_buffer_callback);
 
   smpi_init_options();
   SIMIX_comm_set_copy_data_callback(smpi_comm_copy_buffer_callback);
 
   smpi_init_options();

-

   if (smpi_privatize_global_variables == SMPI_PRIVATIZE_DLOPEN) {
 
     std::string executable_copy = executable;
   if (smpi_privatize_global_variables == SMPI_PRIVATIZE_DLOPEN) {
 
     std::string executable_copy = executable;


@@ -631,9 +582,13 @@ int smpi_main(const char* executable, int argc, char *argv[])
 
   }
 
 
   }
 

+  SMPI_init();

   SIMIX_launch_application(argv[2]);
   SIMIX_launch_application(argv[2]);

+  SMPI_app_instance_register(smpi_default_instance_name, nullptr,
+                               SIMIX_process_count()); // This call has a side effect on process_count...
+  MPI_COMM_WORLD = *smpi_deployment_comm_world(smpi_default_instance_name);
+  smpi_universe_size = process_count;

 
 

-  SMPI_init();

 
   /* Clean IO before the run */
   fflush(stdout);
 
   /* Clean IO before the run */
   fflush(stdout);


@@ -659,9 +614,10 @@ int smpi_main(const char* executable, int argc, char *argv[])
     }
   }
   int ret   = 0;
     }
   }
   int ret   = 0;

-  for (int i = 0, count = smpi_process_count(); i < count; i++) {
-    if(process_data[i]->return_value()!=0){
-      ret=process_data[i]->return_value();//return first non 0 value
+  for (auto& pair : process_data) {
+    auto& smpi_process = pair.second;
+    if (smpi_process->return_value() != 0) {
+      ret = smpi_process->return_value(); // return first non 0 value

       break;
     }
   }
       break;
     }
   }


@@ -674,7 +630,9 @@ int smpi_main(const char* executable, int argc, char *argv[])
 
 // Called either directly from the user code, or from the code called by smpirun
 void SMPI_init(){
 
 // Called either directly from the user code, or from the code called by smpirun
 void SMPI_init(){

-  smpi_init_logs();
+  simgrid::s4u::Actor::onCreation.connect([](simgrid::s4u::ActorPtr actor) {
+    smpi_add_process(actor);
+  });

   smpi_init_options();
   smpi_global_init();
   smpi_check_options();
   smpi_init_options();
   smpi_global_init();
   smpi_check_options();