Arreglada condicion de carrera en Comunicaciones del bucle

Codes/Main/Main.c

@@ -169,6 +169,7 @@ int checkpoint(int iter, int state, MPI_Request **comm_req) {
     group->numS = config_file->procs[group->grp +1];
 
       results->spawn_start = MPI_Wtime();
+      if(group->myId == ROOT) { printf("Malleability\n");}
     TC(group->numS);
       results->spawn_time[group->grp] = MPI_Wtime() - results->spawn_start;
 
@@ -180,6 +181,8 @@ int checkpoint(int iter, int state, MPI_Request **comm_req) {
     } else {
       state = check_redistribution(iter, comm_req);
     }
+
+      printf("P%d/%d Malleability END state=%d\n", group->myId, group->numP, state);
   }
 
   return state;
@@ -255,16 +258,18 @@ int thread_creation() {
  * El estado de la comunicación es devuelto al finalizar la función. 
  */
 int thread_check(int iter) {
-  if(group->commAsync == MAL_COMM_COMPLETED) {
-    if(pthread_join(async_thread, NULL)) {
-      printf("Error al esperar al hilo\n");
-      MPI_Abort(MPI_COMM_WORLD, -1);
-      return -2;
-    } 
-    return end_redistribution(iter);
-  }
+  int all_completed = 0;
 
-  return MAL_ASYNC_PENDING;
+  // Comprueba que todos los hilos han terminado la distribucion (Mismo valor en commAsync)
+  MPI_Allreduce(&group->commAsync, &all_completed, 1, MPI_INT, MPI_MAX, MPI_COMM_WORLD);
+  if(all_completed != MAL_COMM_COMPLETED) return MAL_ASYNC_PENDING; // Continue only if asynchronous send has ended 
+
+  if(pthread_join(async_thread, NULL)) {
+    printf("Error al esperar al hilo\n");
+    MPI_Abort(MPI_COMM_WORLD, -1);
+    return -2;
+  } 
+  return end_redistribution(iter);
 }
 
 /*
@@ -430,6 +435,7 @@ void iterate(double *matrix, int n, int async_comm) {
   }
 
   if(config_file->comm_tam) {
+    printf("P%d/%d Bcast\n", group->myId, group->numP);
     MPI_Bcast(group->compute_comm_array, config_file->comm_tam, MPI_CHAR, ROOT, MPI_COMM_WORLD);
   }
 
@@ -566,8 +572,8 @@ void obtain_op_times() {
   for(i=0; i<qty; i++) {
     result += computePiSerial(config_file->matrix_tam);
   }
-  printf("Creado Top con valor %lf\n", result);
-  fflush(stdout);
+  //printf("Creado Top con valor %lf\n", result);
+  //fflush(stdout);
 
   config_file->Top = (MPI_Wtime() - start_time) / qty; //Tiempo de una operacion
   MPI_Bcast(&(config_file->Top), 1, MPI_DOUBLE, ROOT, MPI_COMM_WORLD); 

Codes/compila.sh

@@ -6,6 +6,7 @@ if [ $# -gt 0 ]
 then
   if [ $1 = "-e" ]
   then
+    echo "Creado ejecutable para ejecuciones"
     cp a.out bench.out
   fi
 fi

Codes/malleability/CommDist.c

@@ -150,14 +150,7 @@ void send_sync_arrays(struct Dist_data dist_data, char *array, int rootBcast, in
     }
     //print_counts(dist_data, counts.counts, counts.displs, numP_child, "Padres");
     /* COMUNICACION DE DATOS */
-    //int myId;
-    //MPI_Comm_rank(MPI_COMM_WORLD, &myId);
-    //if(myId == 0) { printf("TEST PREALL SEND\n"); fflush(stdout); }
-    //MPI_Barrier(dist_data.intercomm);
     MPI_Alltoallv(array, counts.counts, counts.displs, MPI_CHAR, NULL, counts.zero_arr, counts.zero_arr, MPI_CHAR, dist_data.intercomm);
-    //MPI_Barrier(dist_data.intercomm);
-    //if(myId == 0) { printf("TEST POSTALL SEND\n"); fflush(stdout); }
-    
 }
 
 /*
@@ -182,13 +175,7 @@ void recv_sync_arrays(struct Dist_data dist_data, char *array, int root, int num
     //print_counts(dist_data, counts.counts, counts.displs, numP_parents, "Hijos");
 
     /* COMUNICACION DE DATOS */
-    //int myId;
-    //MPI_Comm_rank(MPI_COMM_WORLD, &myId);
-    //if(myId == 0) { printf("TEST PREALL RECV\n"); fflush(stdout); }
-    //MPI_Barrier(dist_data.intercomm);
     MPI_Alltoallv(&aux, counts.zero_arr, counts.zero_arr, MPI_CHAR, array, counts.counts, counts.displs, MPI_CHAR, dist_data.intercomm);
-    //MPI_Barrier(dist_data.intercomm);
-    //if(myId == 0) { printf("TEST POSTALL RECV\n"); fflush(stdout); }
 }
 
 
@@ -223,6 +210,7 @@ int send_async(char *array, int qty, int myId, int numP, int root, MPI_Comm inte
 
     getIds_intercomm(dist_data, numP_child, &idS); // Obtener rango de Id hijos a los que este proceso manda datos
 
+    // MAL_USE_THREAD sigue el camino sincrono
     if(parents_wait == MAL_USE_NORMAL) {
       *comm_req = (MPI_Request *) malloc(sizeof(MPI_Request));
       *comm_req[0] = MPI_REQUEST_NULL;
@@ -275,6 +263,7 @@ void recv_async(char **array, int qty, int myId, int numP, int root, MPI_Comm in
 
     getIds_intercomm(dist_data, numP_parents, &idS); // Obtener el rango de Ids de padres del que este proceso recibira datos
 
+    // MAL_USE_THREAD sigue el camino sincrono
     if(parents_wait == MAL_USE_POINT) {
       comm_req = (MPI_Request *) malloc(numP_parents * sizeof(MPI_Request));
       for(i=0; i<numP_parents; i++){
@@ -394,12 +383,12 @@ void recv_async_point_arrays(struct Dist_data dist_data, char *array, int root,
     if(idI == 0) {
       set_counts(0, numP_parents, dist_data, counts.counts);
       idI++;
-      MPI_Isend(array, counts.counts[0], MPI_CHAR, 0, 99, dist_data.intercomm, &(comm_req[0]));
+      MPI_Irecv(array, counts.counts[0], MPI_CHAR, 0, 99, dist_data.intercomm, &(comm_req[0]));
     }
     for(i=idI; i<idE; i++) {
       set_counts(i, numP_parents, dist_data, counts.counts);
       counts.displs[i] = counts.displs[i-1] + counts.counts[i-1];
-      MPI_Isend(array+counts.displs[i], counts.counts[0], MPI_CHAR, i, 99, dist_data.intercomm, &(comm_req[0]));
+      MPI_Irecv(array+counts.displs[i], counts.counts[0], MPI_CHAR, i, 99, dist_data.intercomm, &(comm_req[0]));
     }
     //print_counts(dist_data, counts.counts, counts.displs, numP_parents, "Hijos");
 }