collect output for bcast

src/network-workloads/model-net-mpi-replay.c

@@ -2035,7 +2035,7 @@ static void get_next_mpi_operation_rc(nw_state* s, tw_bf * bf, nw_message * m, t
         {
             if(bf->c1)
             {
-		s->col_latency[s->num_msg_sizes] = 0;
+		        s->col_latency[s->num_msg_sizes] = 0;
                 s->col_msizes[s->num_msg_sizes] = 0;
 		//todo: reverse handler for num_msg_sizes
                 s->num_all_reduce--;
@@ -2050,6 +2050,22 @@ static void get_next_mpi_operation_rc(nw_state* s, tw_bf * bf, nw_message * m, t
         }
         break;
 		case CODES_WK_BCAST:
+        {
+            if(bf->c1)
+            {
+                s->col_latency[s->num_msg_sizes] = 0;
+                s->col_msizes[s->num_msg_sizes] = 0;
+        //todo: reverse handler for num_msg_sizes
+                s->num_all_reduce--;
+                s->col_time = m->rc.saved_send_time; 
+                s->all_reduce_time -= s->col_time;
+            }
+            else
+            {
+               s->col_time = 0; 
+            }
+            codes_issue_next_event_rc(lp);
+        }
 		case CODES_WK_ALLGATHER:
 		case CODES_WK_ALLGATHERV:
 		case CODES_WK_ALLTOALL:
@@ -2183,15 +2199,15 @@ static void get_next_mpi_operation(nw_state* s, tw_bf * bf, nw_message * m, tw_l
                 {
                     bf->c1 = 1;
                     m->rc.saved_delay = s->all_reduce_time;
-		    if(s->num_msg_sizes == 0)
-		    {
-			s->num_msg_sizes += 1;
-		    }
-	 	    else if(s->col_msizes[s->num_msg_sizes-1] != mpi_op.u.collective.num_bytes)
-		    {	
-			s->num_msg_sizes += 1;
-		    }
-		    s->col_latency[s->num_msg_sizes-1] += (tw_now(lp) - s->col_time);
+		        if(s->num_msg_sizes == 0)
+		        {
+			        s->num_msg_sizes += 1;
+		        }
+	 	        else if(s->col_msizes[s->num_msg_sizes-1] != mpi_op.u.collective.num_bytes)
+		        {	
+			        s->num_msg_sizes += 1;
+		        }
+		            s->col_latency[s->num_msg_sizes-1] += (tw_now(lp) - s->col_time);
                     s->col_msizes[s->num_msg_sizes-1] = mpi_op.u.collective.num_bytes;
                     s->all_reduce_time += (tw_now(lp) - s->col_time);
                     m->rc.saved_send_time = s->col_time;
@@ -2208,6 +2224,33 @@ static void get_next_mpi_operation(nw_state* s, tw_bf * bf, nw_message * m, tw_l
 			
             case CODES_WK_REDUCE:
 			case CODES_WK_BCAST:
+            {
+                s->num_cols++;
+                if(s->col_time > 0)
+                {
+                    bf->c1 = 1;
+                    m->rc.saved_delay = s->all_reduce_time;
+                if(s->num_msg_sizes == 0)
+                {
+                    s->num_msg_sizes += 1;
+                }
+                else if(s->col_msizes[s->num_msg_sizes-1] != mpi_op.u.collective.num_bytes)
+                {   
+                    s->num_msg_sizes += 1;
+                }
+                    s->col_latency[s->num_msg_sizes-1] += (tw_now(lp) - s->col_time);
+                    s->col_msizes[s->num_msg_sizes-1] = mpi_op.u.collective.num_bytes;
+                    s->all_reduce_time += (tw_now(lp) - s->col_time);
+                    m->rc.saved_send_time = s->col_time;
+                    s->col_time = 0;
+                    s->num_all_reduce++;
+                }
+                else
+                {
+                   s->col_time = tw_now(lp); 
+                }
+                codes_issue_next_event(lp);
+            }
 			case CODES_WK_ALLGATHER:
 			case CODES_WK_ALLGATHERV:
 			case CODES_WK_ALLTOALL: