coll/libnbc: fix MPI_Ineighbor_alltoallw when in place.

This is a one off and a quick fix, and it unlikely works with complex types
(e.g. negative lower bound).
libnbc considers MPI_Ineighbor_alltoallw is in place when send and recv buffers are identical,
and even if they do not overlap because of distinct displacements.

Thanks Jun Kudo for the bug report.

Author: ggouaillardet

ompi/mca/coll/libnbc/nbc_ineighbor_alltoallw.c

@@ -1,9 +1,11 @@
 /*
- * Copyright (c) 2006 The Trustees of Indiana University and Indiana
- *                    University Research and Technology
- *                    Corporation.  All rights reserved.
- * Copyright (c) 2006 The Technical University of Chemnitz. All
- *                    rights reserved.
+ * Copyright (c) 2006      The Trustees of Indiana University and Indiana
+ *                         University Research and Technology
+ *                         Corporation.  All rights reserved.
+ * Copyright (c) 2006      The Technical University of Chemnitz. All
+ *                         rights reserved.
+ * Copyright (c) 2016      Research Organization for Information Science
+ *                         and Technology (RIST). All rights reserved.
  *
  * Author(s): Torsten Hoefler <htor@cs.indiana.edu>
  *
@@ -100,15 +102,16 @@ int ompi_coll_libnbc_ineighbor_alltoallw(void *sbuf, int *scounts, MPI_Aint *sdi
       if(res != NBC_OK) return res;
 
       if(inplace) { /* we need an extra buffer to be deadlock-free */
-        int sumrbytes=0;
+        MPI_Aint sumrbytes=0, offset=0;
         for(i=0; i<indegree; ++i) sumrbytes += rcounts[i]*rcvexts[i];
         handle->tmpbuf = malloc(sumrbytes);
 
         for(i = 0; i < indegree; i++) {
           if(srcs[i] != MPI_PROC_NULL) {
-            res = NBC_Sched_recv((char*)0+rdisps[i], true, rcounts[i], rtypes[i], srcs[i], schedule);
+            res = NBC_Sched_recv((char*)0+offset, true, rcounts[i], rtypes[i], srcs[i], schedule);
             if (NBC_OK != res) { printf("Error in NBC_Sched_recv() (%i)\n", res); return res; }
           }
+          offset += rcounts[i]*rcvexts[i];
         }
         for(i = 0; i < outdegree; i++) {
           if(dsts[i] != MPI_PROC_NULL) {
@@ -117,11 +120,15 @@ int ompi_coll_libnbc_ineighbor_alltoallw(void *sbuf, int *scounts, MPI_Aint *sdi
           }
         }
         /* unpack from buffer */
+        offset = 0;
         for(i = 0; i < indegree; i++) {
-          res = NBC_Sched_barrier(schedule);
-          if (NBC_OK != res) { printf("Error in NBC_Sched_barrier() (%i)\n", res); return res; }
-          res = NBC_Sched_copy((char*)0+rdisps[i], true, rcounts[i], rtypes[i], (char*)rbuf+rdisps[i], false, rcounts[i], rtypes[i], schedule);
-          if (NBC_OK != res) { printf("Error in NBC_Sched_copy() (%i)\n", res); return res; }
+          if(srcs[i] != MPI_PROC_NULL) {
+            res = NBC_Sched_barrier(schedule);
+            if (NBC_OK != res) { printf("Error in NBC_Sched_barrier() (%i)\n", res); return res; }
+            res = NBC_Sched_copy((char*)0+offset, true, rcounts[i], rtypes[i], (char*)rbuf+rdisps[i], false, rcounts[i], rtypes[i], schedule);
+            if (NBC_OK != res) { printf("Error in NBC_Sched_copy() (%i)\n", res); return res; }
+          }
+          offset += rcounts[i] * rcvexts[i];
         }
       } else { /* non INPLACE case */
         /* simply loop over neighbors and post send/recv operations */