basesmuma: do not call MB directly

opal does not always define MB. It is recommended that opal_atomic_[rw]mb is
called instead. We will need to address the cases where these functions are
no-ops on weak-memory ordered cpus.

cmr=v1.7.5:ticket=trac:4158

This commit was SVN r30534.

The following Trac tickets were found above:
  Ticket 4158 --> https://svn.open-mpi.org/trac/ompi/ticket/4158

ompi/mca/bcol/basesmuma/bcol_basesmuma_allgather.c

@@ -207,7 +207,7 @@ MAIN_PHASE:
     /* we start the recursive k - ing phase */
     for( *iteration = 0; *iteration < pow_k; (*iteration)++) {
         /* announce my arrival */
-        MB();
+        opal_atomic_wmb ();
         my_ctl_pointer->flags[ALLGATHER_FLAG][bcol_id] = ready_flag;
         /* calculate the number of active requests */
         CALC_ACTIVE_REQUESTS(active_requests,exchange_node->rank_exchanges[*iteration],tree_order);
@@ -271,7 +271,7 @@ MAIN_PHASE:
     /* finish off the last piece, send the data back to the extra  */
     if( 0 < exchange_node->n_extra_sources ) {
         /* simply announce my arrival */
-        MB();
+        opal_atomic_wmb ();
         my_ctl_pointer->flags[ALLGATHER_FLAG][bcol_id] = ready_flag;
 
     }
@@ -433,7 +433,7 @@ MAIN_PHASE:
     /* start the recursive k - ing phase */
     for( *iter=*iteration; *iter < pow_k; (*iter)++) {
         /* I am ready at this level */
-        MB();
+        opal_atomic_wmb ();
         my_ctl_pointer->flags[ALLGATHER_FLAG][bcol_id] = ready_flag;
         if( 0 == *active_requests ) {
             /* flip some bits, if we don't have active requests from a previous visit */
@@ -499,7 +499,7 @@ MAIN_PHASE:
     /* finish off the last piece, send the data back to the extra  */
     if( 0 < exchange_node->n_extra_sources ) {
         /* simply announce my arrival */
-        MB();
+        opal_atomic_wmb ();
         my_ctl_pointer->flags[ALLGATHER_FLAG][bcol_id] = ready_flag;
 
     }

ompi/mca/bcol/basesmuma/bcol_basesmuma_allreduce.c

@@ -2,7 +2,7 @@
 /*
  * Copyright (c) 2009-2013 Oak Ridge National Laboratory.  All rights reserved.
  * Copyright (c) 2009-2012 Mellanox Technologies.  All rights reserved.
- * Copyright (c) 2013      Los Alamos National Security, LLC. All rights
+ * Copyright (c) 2013-2014 Los Alamos National Security, LLC. All rights
  *                         reserved.
  * $COPYRIGHT$
  *
@@ -409,7 +409,7 @@ int bcol_basesmuma_allreduce_intra_recursive_doubling(bcol_function_args_t *inpu
     }
 
     /* signal that I have arrived */
-    /* MB(); */
+    /* opal_atomic_wmb (); */
     my_ctl_pointer->sequence_number = sequence_number;
 
     /* If we use this buffer more than once by an sm module in
@@ -581,7 +581,7 @@ int bcol_basesmuma_allreduce_intra_recursive_doubling(bcol_function_args_t *inpu
             /*
              * Signal parent that data is ready
              */
-            MB();
+            opal_atomic_wmb ();
             /*my_ctl_pointer->flags[ALLREDUCE_FLAG] = ready_flag;*/
             my_ctl_pointer->flag = ready_flag;
 

ompi/mca/bcol/basesmuma/bcol_basesmuma_bcast.c

@@ -1,6 +1,8 @@
 /*
  * Copyright (c) 2009-2012 Oak Ridge National Laboratory.  All rights reserved.
  * Copyright (c) 2009-2012 Mellanox Technologies.  All rights reserved.
+ * Copyright (c) 2014      Los Alamos National Security, LLC. All rights
+ *                         reserved.
  * $COPYRIGHT$
  *
  * Additional copyrights may follow
@@ -236,7 +238,7 @@ int bcol_basesmuma_bcast(bcol_function_args_t *input_args,
         memcpy(data_addr, (void *)parent_data_pointer,pack_len);
 		
         /* Signal to children that they may read the data from my shared buffer */
-        MB();
+        opal_atomic_wmb ();
         my_ctl_pointer->flags[BCAST_FLAG][bcol_id] = ready_flag;
     }
 
@@ -438,7 +440,7 @@ int bcol_basesmuma_hdl_zerocopy_bcast(bcol_function_args_t *input_args,
         }
 		
         /* Signal to children that they may read the data from my shared buffer */
-        MB();
+        opal_atomic_wmb ();
         hdl_desc->des_src = hdl_seg;
         hdl_desc->des_src_cnt = 1; 
         for (ridx = 0; ridx < my_fanout_read_tree->n_children; ridx++) {

ompi/mca/bcol/basesmuma/bcol_basesmuma_bcast_prime.c

@@ -165,7 +165,7 @@ int bcol_basesmuma_bcast_k_nomial_knownroot(bcol_function_args_t *input_args,
     } else {
     ready_flag =  my_ctl_pointer->flags[BCAST_FLAG][bcol_id];
     }
-    MB();
+    opal_atomic_wmb ();
     my_ctl_pointer->sequence_number = sequence_number;
   */
 
@@ -178,7 +178,7 @@ int bcol_basesmuma_bcast_k_nomial_knownroot(bcol_function_args_t *input_args,
     /*
      * signal ready flag
      */
-    MB();
+    opal_atomic_wmb ();
     my_ctl_pointer->flags[BCAST_FLAG][bcol_id] = ready_flag;
 
     /* root is finished */
@@ -214,7 +214,7 @@ int bcol_basesmuma_bcast_k_nomial_knownroot(bcol_function_args_t *input_args,
   /* copy the data */
   memcpy(data_addr, (void *) parent_data_pointer, pack_len);
   /* set the memory barrier to ensure completion */
-  MB();
+  opal_atomic_wmb ();
   /* signal that I am done */
   my_ctl_pointer->flags[BCAST_FLAG][bcol_id] = ready_flag;
 
@@ -312,7 +312,7 @@ int bcol_basesmuma_bcast_k_nomial_anyroot(bcol_function_args_t *input_args,
      * set the radix_mask */
     radix_mask = pow_k_group_size;
     /* send to children */
-    MB();
+    opal_atomic_wmb ();
     BASESMUMA_K_NOMIAL_SEND_CHILDREN(radix_mask,
                                      radix,0,
                                      my_rank,group_size, ready_flag);
@@ -347,7 +347,7 @@ int bcol_basesmuma_bcast_k_nomial_anyroot(bcol_function_args_t *input_args,
       radix_mask /= radix;
 
       /* send to children */
-      MB();
+      opal_atomic_wmb ();
       BASESMUMA_K_NOMIAL_SEND_CHILDREN(radix_mask,
                                        radix, relative_rank,
                                        my_rank, group_size, ready_flag);
@@ -515,7 +515,7 @@ int bcol_basesmuma_binary_scatter_allgather_segment(bcol_function_args_t *input_
     /* important that these be set before my children
      * see the ready flag raised
      */
-    MB();
+    opal_atomic_wmb ();
     my_ctl_pointer->flag = ready_flag;
 
     /* root is finished */
@@ -638,7 +638,7 @@ int bcol_basesmuma_binary_scatter_allgather_segment(bcol_function_args_t *input_
     my_ctl_pointer->n_sends = parent_ctl_pointer->n_sends;
 
     /* set the memory barrier */
-    MB();
+    opal_atomic_wmb ();
 
     /* fire the ready flag */
     my_ctl_pointer->flag = ready_flag;
@@ -696,7 +696,7 @@ int bcol_basesmuma_binary_scatter_allgather_segment(bcol_function_args_t *input_
                (size_t)length);
       }
       /* set the memory barrier to ensure completion */
-      MB();
+      opal_atomic_wmb ();
       /* signal that I am done */
       my_ctl_pointer->flag = ready_flag;
       /* set my status */
@@ -746,14 +746,14 @@ int bcol_basesmuma_binary_scatter_allgather_segment(bcol_function_args_t *input_
     local_offset = my_ctl_pointer->offset_zip;
     /* compute the correct length */
     length = length*(1<<(start - 1));
-    /* careful! skip over the MB() to avoid the
+    /* careful! skip over the opal_atomic_wmb () to avoid the
      * cost on every re-entry
      */
     goto Loop;
   }
 
 
-  MB();
+  opal_atomic_wmb ();
   /* I am ready, set the flag */
   my_ctl_pointer->flag = ready_flag;
 
@@ -813,7 +813,7 @@ int bcol_basesmuma_binary_scatter_allgather_segment(bcol_function_args_t *input_
         /* bump the ready flag */
         ready_flag++;
         /* ensure completion */
-        MB();
+        opal_atomic_wmb ();
 
         /* fire the flag for the next level */
         my_ctl_pointer->flag = ready_flag;

ompi/mca/bcol/basesmuma/bcol_basesmuma_gather.c

@@ -2,7 +2,7 @@
 /*
  * Copyright (c) 2009-2012 Oak Ridge National Laboratory.  All rights reserved.
  * Copyright (c) 2009-2012 Mellanox Technologies.  All rights reserved.
- * Copyright (c) 2013      Los Alamos National Security, LLC. All rights
+ * Copyright (c) 2013-2014 Los Alamos National Security, LLC. All rights
  *                         reserved.
  * $COPYRIGHT$
  *
@@ -698,7 +698,7 @@ int bcol_basesmuma_k_nomial_gather(bcol_function_args_t *input_args,
                       exchange_node->rank_exchanges[i][j], buff_offset + exchange_node->payload_info[i][j].r_offset,
                       exchange_node->payload_info[i][j].r_len*pack_len);
                     */
-                    MB();
+                    opal_atomic_wmb ();
                     knt++;
                     if(knt == exchange_node->n_actual_exchanges) {
                         /* this is the trick to break the root out,
@@ -1063,7 +1063,7 @@ int bcol_basesmuma_k_nomial_gather(bcol_function_args_t *input_args,
                               fprintf(stderr,"hello n_actual_exch is %d \n",
                               exchange_node->n_actual_exchanges);
                             */
-                            MB();
+                            opal_atomic_wmb ();
                             my_ctl_pointer->gflag = ready_flag;
 
                             goto LAST_STEP;

ompi/mca/bcol/basesmuma/bcol_basesmuma_lmsg_bcast.c

@@ -215,7 +215,7 @@ int bcol_basesmuma_lmsg_scatter_allgather_portals_bcast_old(bcol_function_args_t
         /* important that these be set before my children
          * see the ready flag raised
          */
-        MB();
+        opal_atomic_wmb ();
         my_ctl_pointer->flag = ready_flag;
 
 		/* Wait for my scatter partner */
@@ -342,7 +342,7 @@ Probe:
         /* important that these be set before my children
          * see the ready flag raised
          */
-        MB();
+        opal_atomic_wmb ();
         my_ctl_pointer->flag = ready_flag;
 
 		wait_for_peers(my_rank, my_lmsg_ctl_pointer->n_sends, data_buffs,
@@ -415,7 +415,7 @@ Probe:
 				msg_posted = true;
 				/* set the memory barrier to ensure completion
 				 * and signal I am done getting scatter data*/
-			MB();
+			opal_atomic_wmb ();
 			my_ctl_pointer->flag = ready_flag;
 
 				wait_for_peers(my_rank, my_lmsg_ctl_pointer->n_sends, data_buffs,
@@ -441,7 +441,7 @@ Probe:
 								remote_offset, length);
 
 				/* signal that I am done reading data from parent */
-		    MB();
+		    opal_atomic_wmb ();
 	        my_ctl_pointer->flag = ready_flag;
 			}
 
@@ -499,7 +499,7 @@ Allgather:
 
 
 	ready_flag++;
-    MB();
+    opal_atomic_wmb ();
     my_ctl_pointer->flag = ready_flag;
 
 	/*
@@ -533,7 +533,7 @@ Allgather:
 								local_sg_offset, length);
 
 		ready_flag++;
-		MB();
+		opal_atomic_wmb ();
 	my_ctl_pointer->flag = ready_flag;
 
 		/* Block until partner completed this level of recursive-doubling stage */
@@ -753,7 +753,7 @@ int bcol_basesmuma_lmsg_scatter_allgather_portals_bcast(bcol_function_args_t *in
 		 /* important that these be set before my children
          * see the ready flag raised
          */
-        MB();
+        opal_atomic_wmb ();
         my_ctl_pointer->flags[BCAST_FLAGS] = ready_flag;
         BASESMUMA_VERBOSE(1,("I am the root(ctl_pointer %x) of the data flag value %d",my_ctl_pointer, my_ctl_pointer->flag));
 		/* Wait for my scatter partner */
@@ -879,7 +879,7 @@ Probe:
 		/* important that these be set before my children
          * see the ready flag raised
          */
-        MB();
+        opal_atomic_wmb ();
         my_ctl_pointer->flags[BCAST_FLAG] = ready_flag;
 
 		wait_for_peers_nb(my_rank, my_ctl_pointer->n_sends, ctl_structs,
@@ -949,7 +949,7 @@ Probe:
 					  );
 				msg_posted = true;
 				/* set the memory barrier to ensure completion */
-			MB();
+			opal_atomic_wmb ();
 		/* signal that I am done */
 			my_ctl_pointer->flags[BCAST_FLAG] = ready_flag;
 
@@ -977,7 +977,7 @@ Probe:
 								remote_offset, length);
 
 				/* signal that I am done reading data from parent */
-		    MB();
+		    opal_atomic_wmb ();
 	        my_ctl_pointer->flags[BCAST_FLAG] = ready_flag;
 			}
 
@@ -1034,7 +1034,7 @@ Allgather:
 	}
 
 	ready_flag++;
-    MB();
+    opal_atomic_wmb ();
     my_ctl_pointer->flags[BCAST_FLAG] = ready_flag;
 
     for( i = 1; i < pow_2_levels; i++) {
@@ -1064,7 +1064,7 @@ Allgather:
 								local_sg_offset, length);
 
 		ready_flag++;
-		MB();
+		opal_atomic_wmb ();
 	my_ctl_pointer->flags[BCAST_FLAG] = ready_flag;
 
 		/* Block until partner is at this level of recursive-doubling stage */
@@ -1257,7 +1257,7 @@ Extra :
 		rc = sm_portals_root_scatter(sg_state);
 
 		/* gvm Fix: Redudant
-		MB();
+		opal_atomic_wmb ();
 		*/
 
 		sg_state->my_ctl_pointer->flag = sg_state->ready_flag;
@@ -1433,7 +1433,7 @@ Scatter_parent_wait:
 
 				/* signal that I am done reading data from parent */
 				/*
-		    MB();
+		    opal_atomic_wmb ();
 				*/
 	        sg_state->my_ctl_pointer->flag = sg_state->ready_flag;
 			}
@@ -1487,7 +1487,7 @@ Allgather:
 	BASESMUMA_VERBOSE(5,("Done with allgather phase"));
 	/* I reached an allgather phase */
 	sg_state->ready_flag++;
-    MB();
+    opal_atomic_wmb ();
     sg_state->my_ctl_pointer->flag = sg_state->ready_flag;
 
 	rc = sm_portals_bcasts_allgather_phase(sg_state);
@@ -1840,7 +1840,7 @@ Allgather:
 	}
 
 	sg_state->ready_flag++;
-    MB();
+    opal_atomic_wmb ();
     sg_state->my_ctl_pointer->flag = sg_state->ready_flag;
 
 	rc = sm_portals_bcasts_allgather_phase(sg_state);

ompi/mca/bcol/basesmuma/bcol_basesmuma_lmsg_knomial_bcast.c

@@ -1,6 +1,8 @@
 /*
  * Copyright (c) 2009-2012 Oak Ridge National Laboratory.  All rights reserved.
  * Copyright (c) 2009-2012 Mellanox Technologies.  All rights reserved.
+ * Copyright (c) 2014      Los Alamos National Security, LLC. All rights
+ *                         reserved.
  * $COPYRIGHT$
  *
  * Additional copyrights may follow
@@ -214,7 +216,7 @@ int bcol_basesmuma_lmsg_bcast_k_nomial_anyroot(bcol_function_args_t *input_args,
 	mca_bcol_basesmuma_portals_get_msg(cs, parent_lmsg_ctl_pointer, userbuf, pack_len);
 
     /* set the memory barrier to ensure completion */
-    MB();
+    opal_atomic_wmb ();
     /* signal that I am done */
     my_ctl_pointer->flag = ready_flag;
 

ompi/mca/bcol/basesmuma/bcol_basesmuma_rd_barrier.c

@@ -1,7 +1,7 @@
 /*
  * Copyright (c) 2009-2013 Oak Ridge National Laboratory.  All rights reserved.
  * Copyright (c) 2009-2012 Mellanox Technologies.  All rights reserved.
- * Copyright (c) 2013      Los Alamos National Security, LLC. All rights
+ * Copyright (c) 2013-2014 Los Alamos National Security, LLC. All rights
  *                         reserved.
  * $COPYRIGHT$
  *
@@ -91,7 +91,7 @@ int bcol_basesmuma_recursive_double_barrier(bcol_function_args_t *input_args,
     my_ctl->flag = -1;
     /* don't need to set this flag anymore */
     my_ctl->sequence_number = sequence_number;
-    /* MB();*/
+    /* opal_atomic_wmb ();*/
 
     if(0 < my_exchange_node->n_extra_sources) {
         if (EXCHANGE_NODE == my_exchange_node->node_type) {

ompi/mca/bcol/basesmuma/bcol_basesmuma_rd_nb_barrier.c

@@ -2,6 +2,8 @@
  * Copyright (c) 2009-2012 UT-Battelle, LLC. All rights reserved.
  * Copyright (c) 2009-2012 Mellanox Technologies. All rights reserved.
  * Copyright (c) 2013 Cisco Systems, Inc.  All rights reserved.
+ * Copyright (c) 2014      Los Alamos National Security, LLC. All rights
+ *                         reserved.
  * $COPYRIGHT$
  *
  * Additional copyrights may follow
@@ -72,7 +74,7 @@ int bcol_basesmuma_rd_nb_barrier_init_admin(
     /* signal that I have arrived */
     my_ctl->flag = -1;
 
-    MB();
+    opal_atomic_wmb ();
 
 	/* don't need to set this flag anymore */
     my_ctl->sequence_number = bank_genaration;