Reduced amount of arguments needed in MaM. Reduced tasks performed by users of MaM.

Codes/Main/Main.c

@@ -66,7 +66,7 @@ int main(int argc, char *argv[]) {
     }
 
     init_group_struct(argv, argc, myId, numP);
-    im_child = MAM_Init(myId, numP, ROOT, comm, argv[0], nodelist, num_cpus, num_nodes);
+    im_child = MAM_Init(ROOT, comm, argv[0], nodelist, num_cpus, num_nodes);
 
     if(!im_child) { //TODO REFACTOR Simplificar inicio
       init_application();
@@ -82,8 +82,6 @@ int main(int argc, char *argv[]) {
     // EMPIEZA LA EJECUCION-------------------------------
     //
     do {
-      MPI_Comm_size(comm, &(group->numP));
-      MPI_Comm_rank(comm, &(group->myId));
 
       if(group->grp != 0) {
         obtain_op_times(0); //Obtener los nuevos valores de tiempo para el computo
@@ -107,15 +105,11 @@ int main(int argc, char *argv[]) {
       if(res==1) { // Se ha llegado al final de la aplicacion
         MPI_Barrier(comm);
         results->exec_time = MPI_Wtime() - results->exec_start - results->wasted_time;
+        print_local_results();
       }
 
-      print_local_results();
       reset_results_index(results);
 
-      if(res!=1) {
-        if(comm != MPI_COMM_WORLD) MPI_Comm_free(&comm);
-        comm = new_comm;
-      }
       group->grp = group->grp + 1;
     } while(config_file->n_groups > group->grp && config_file->groups[group->grp].sm == MALL_SPAWN_MERGE);
 
@@ -154,7 +148,7 @@ int main(int argc, char *argv[]) {
  * de procesos. En caso contrario se devuelve 0.
  */
 int work() {
-  int iter, maxiter, state, res, commited, targets_qty;
+  int iter, maxiter, state, res, commited;
   int wait_completed = MAM_CHECK_COMPLETION;
 
   maxiter = config_file->groups[group->grp].iters;
@@ -182,11 +176,13 @@ int work() {
   compute_results_iter(results, group->myId, group->numP, ROOT, config_file->n_stages, config_file->capture_method, comm);
   if(config_file->n_groups == group->grp + 1) { res=1; }
   else {
-    MAM_Get_comm(&new_comm, &targets_qty);
+    MAM_Get_comm(&new_comm);
     send_config_file(config_file, ROOT, new_comm);
     results_comm(results, ROOT, config_file->n_resizes, new_comm);
-    MPI_Comm_free(&new_comm);
-    MAM_Commit(&commited, &new_comm); 
+    print_local_results();
+    MAM_Commit(&commited, &comm); 
+    MPI_Comm_size(comm, &(group->numP));
+    MPI_Comm_rank(comm, &(group->myId));
   }
   if(state == MAM_ZOMBIE) res=state;
   return res;
@@ -220,7 +216,7 @@ double iterate(int async_comm) {
 
   // Se esta realizando una redistribucion de datos asincrona
   if(async_comm == MAM_PENDING) { 
-  // TODO Que diferencie entre ambas en el IO
+    // TODO Que diferencie entre ambas en el IO
     results->iters_async += 1;
   }
 
@@ -315,7 +311,6 @@ int print_local_results() {
   int ptr_local, ptr_out, err;
   char *file_name;
 
-  //compute_results_iter(results, group->myId, group->numP, ROOT, config_file->n_stages, config_file->capture_method, comm);
   if(group->myId == ROOT) {
     ptr_out = dup(1);
 
@@ -550,11 +545,11 @@ void init_originals() {
 }
 
 void init_targets() {
-  int commited, targets_qty;
+  int commited;
   size_t i, entries;
   void *value = NULL;
 
-  MAM_Get_comm(&new_comm, &targets_qty);
+  MAM_Get_comm(&new_comm);
 
   malleability_get_data(&value, 0, 1, 0);
   group->grp = *((int *)value);
@@ -564,10 +559,11 @@ void init_targets() {
   results = malloc(sizeof(results_data));
   init_results_data(results, config_file->n_resizes, config_file->n_stages, config_file->groups[group->grp].iters);
   results_comm(results, ROOT, config_file->n_resizes, new_comm);
-  MPI_Comm_free(&new_comm);
-  
   MAM_Commit(&commited, &comm);
 
+  MPI_Comm_size(comm, &(group->numP));
+  MPI_Comm_rank(comm, &(group->myId));
+
   // TODO Refactor - Que sea una unica funcion
   // Obtiene las variables que van a utilizar los hijos
 

Codes/malleability/malleabilityManager.c

@@ -47,7 +47,7 @@ malleability_data_t *dist_a_data;
  * la comunicacion los procesos hijo estan preparados para ejecutar la
  * aplicacion.
  */
-int MAM_Init(int myId, int numP, int root, MPI_Comm comm, char *name_exec, char *nodelist, int num_cpus, int num_nodes) {
+int MAM_Init(int root, MPI_Comm comm, char *name_exec, char *nodelist, int num_cpus, int num_nodes) {
   MPI_Comm dup_comm, thread_comm;
 
   #if USE_MAL_DEBUG
@@ -67,8 +67,8 @@ int MAM_Init(int myId, int numP, int root, MPI_Comm comm, char *name_exec, char
   MPI_Comm_set_name(dup_comm, "MPI_COMM_MAM");
   MPI_Comm_set_name(thread_comm, "MPI_COMM_MAM_THREAD");
 
-  mall->myId = myId;
-  mall->numP = numP;
+  MPI_Comm_rank(comm, &(mall->myId));
+  MPI_Comm_size(comm, &(mall->numP));
   mall->root = root;
   mall->root_parents = -1;
   mall->comm = dup_comm;
@@ -136,8 +136,8 @@ void MAM_Finalize() {
 
   MAM_Free_main_datatype();
   free_malleability_times();
-  if(mall->comm != MPI_COMM_WORLD) MPI_Comm_free(&(mall->comm));
-  if(mall->thread_comm != MPI_COMM_WORLD) MPI_Comm_free(&(mall->thread_comm));
+  if(mall->comm != MPI_COMM_WORLD && mall->comm != MPI_COMM_NULL) MPI_Comm_free(&(mall->comm));
+  if(mall->thread_comm != MPI_COMM_WORLD && mall->thread_comm != MPI_COMM_NULL) MPI_Comm_free(&(mall->thread_comm));
   free(mall);
   free(mall_conf);
 
@@ -265,31 +265,26 @@ int MAM_Checkpoint(int *mam_state, int wait_completed) {
 
 /*
  * Returns an intracommunicator to allow users to perform their
- * own redistributions. The user must free this communicator
- * when is not longer needed.
+ * own redistributions. The user may free this communicator
+ * when is not longer needed. It will be freed by MaM when
+ * commiting the reconfiguration.
  *
- * This is a blocking function, must be called by all processes involved in the
- * reconfiguration.
- * TODO Hacer en otro sitio la creacion del comunicador y borrar en commit.
  *
  * The communicator is only returned if the state of reconfiguration
  * is completed (MALL_COMPLETED / MAM_COMPLETED). Otherwise MALL_DENIED is obtained.
  */
-int MAM_Get_comm(MPI_Comm *comm, int *targets_qty) {
+int MAM_Get_comm(MPI_Comm *comm) {
   if(!(state == MALL_COMPLETED || state == MALL_ZOMBIE)) {
     return MALL_DENIED;
   }
-
-  MPI_Comm_dup(mall->user_comm, comm);
-  MPI_Comm_set_name(*comm, "MPI_MAM_DUP");
-  *targets_qty = mall->numC;
+  *comm = mall->user_comm;
   return 0;
 }
 
 /*
  * TODO
  */
-void MAM_Commit(int *mam_state, MPI_Comm *new_comm) {
+void MAM_Commit(int *mam_state, MPI_Comm *updated_comm) {
   if(!(state == MALL_COMPLETED || state == MALL_ZOMBIE)) {
     *mam_state = MALL_DENIED;
     return;
@@ -307,13 +302,25 @@ void MAM_Commit(int *mam_state, MPI_Comm *new_comm) {
       zombies_collect_suspended(mall->comm, mall->myId, mall->numP, mall->numC, mall->root);
     }
   }
+
   // Reset/Free unneded communicators
-  if(mall->user_comm != MPI_COMM_WORLD) MPI_Comm_free(&(mall->user_comm));
+  if(*updated_comm != MPI_COMM_WORLD && *updated_comm != MPI_COMM_NULL) MPI_Comm_free(updated_comm);
+  if(mall->user_comm != MPI_COMM_WORLD && mall->user_comm != MPI_COMM_NULL) MPI_Comm_free(&(mall->user_comm));
   if(mall_conf->spawn_method == MALL_SPAWN_MERGE) { malleability_comms_update(mall->intercomm); }
   if(mall->intercomm != MPI_COMM_NULL && mall->intercomm != MPI_COMM_WORLD) { 
     MPI_Comm_disconnect(&(mall->intercomm)); //FIXME Error en OpenMPI + Merge
   }
 
+  // Zombies KILL
+  if(state == MALL_ZOMBIE) {
+    MAM_Finalize();
+    MPI_Finalize(); 
+    #if USE_MAL_DEBUG
+      DEBUG_FUNC("Is terminating as zombie", mall->myId, mall->numP); fflush(stdout);
+    #endif
+    exit(0);
+  }
+
   MPI_Comm_rank(mall->comm, &(mall->myId));
   MPI_Comm_size(mall->comm, &(mall->numP));
   mall->root = mall->root_parents == -1 ? mall->root : mall->root_parents;
@@ -322,8 +329,8 @@ void MAM_Commit(int *mam_state, MPI_Comm *new_comm) {
   *mam_state = MAM_COMMITED;
 
   // Set new communicator
-  if(mall_conf->spawn_method == MALL_SPAWN_BASELINE) { *new_comm = MPI_COMM_WORLD; }
-  else if(mall_conf->spawn_method == MALL_SPAWN_MERGE) { MPI_Comm_dup(mall->comm, new_comm); }
+  if(mall_conf->spawn_method == MALL_SPAWN_BASELINE) { *updated_comm = MPI_COMM_WORLD; }
+  else if(mall_conf->spawn_method == MALL_SPAWN_MERGE) { MPI_Comm_dup(mall->comm, updated_comm); }
   #if USE_MAL_DEBUG
     if(mall->myId == mall->root) DEBUG_FUNC("Reconfiguration has been commited", mall->myId, mall->numP); fflush(stdout); MPI_Barrier(MPI_COMM_WORLD);
   #endif

Codes/malleability/malleabilityManager.h

@@ -9,11 +9,11 @@
 #include <mpi.h>
 #include "malleabilityStates.h"
 
-int MAM_Init(int myId, int numP, int root, MPI_Comm comm, char *name_exec, char *nodelist, int num_cpus, int num_nodes);
+int MAM_Init(int root, MPI_Comm comm, char *name_exec, char *nodelist, int num_cpus, int num_nodes);
 void MAM_Finalize();
 int MAM_Checkpoint(int *mam_state, int wait_completed);
-int MAM_Get_comm(MPI_Comm *comm, int *targets_qty);
-void MAM_Commit(int *mam_state, MPI_Comm *new_comm);
+int MAM_Get_comm(MPI_Comm *comm);
+void MAM_Commit(int *mam_state, MPI_Comm *updated_comm);
 
 void MAM_Set_configuration(int spawn_method, int spawn_strategies, int spawn_dist, int red_method, int red_strategies);
 void MAM_Set_target_number(int numC); // TODO TO BE DEPRECATED