Fix bug in shrink when serializing failed procs

The function to convert the group of failed procs to a bitarray was
incorrectly quiting early if one of the globally known failed processes
was not in the communciator being dealt with.
Signed-off-by: Ken Raffenetti <raffenet@mcs.anl.gov>

src/mpi/comm/comm_shrink.c

@@ -50,7 +50,7 @@ int MPIR_Comm_shrink(MPID_Comm *comm_ptr, MPID_Comm **newcomm_ptr)
     int mpi_errno = MPI_SUCCESS;
     MPID_Group *global_failed, *comm_grp, *new_group_ptr;
     int attempts = 0;
-    mpir_errflag_t errflag = MPIR_ERR_NONE, tmp_errflag = MPIR_ERR_NONE;
+    mpir_errflag_t errflag = MPIR_ERR_NONE;
 
     MPID_MPI_STATE_DECL(MPID_STATE_MPIR_COMM_SHRINK);
     MPID_MPI_FUNC_ENTER(MPID_STATE_MPIR_COMM_SHRINK);
@@ -59,7 +59,9 @@ int MPIR_Comm_shrink(MPID_Comm *comm_ptr, MPID_Comm **newcomm_ptr)
     MPIR_Comm_group_impl(comm_ptr, &comm_grp);
 
     do {
-        mpi_errno = MPID_Comm_get_all_failed_procs(comm_ptr, &global_failed, MPIR_SHRINK_TAG);
+        errflag = MPIR_ERR_NONE;
+
+        MPID_Comm_get_all_failed_procs(comm_ptr, &global_failed, MPIR_SHRINK_TAG);
         /* Ignore the mpi_errno value here as it will definitely communicate
          * with failed procs */
 
@@ -68,10 +70,15 @@ int MPIR_Comm_shrink(MPID_Comm *comm_ptr, MPID_Comm **newcomm_ptr)
         if (MPID_Group_empty != global_failed) MPIR_Group_release(global_failed);
 
         mpi_errno = MPIR_Comm_create_group(comm_ptr, new_group_ptr, MPIR_SHRINK_TAG, newcomm_ptr);
-        errflag = mpi_errno || *newcomm_ptr == NULL;
+        if (*newcomm_ptr == NULL) {
+            errflag = MPIR_ERR_PROC_FAILED;
+        } else if (mpi_errno) {
+            errflag = MPIR_ERR_GET_CLASS(mpi_errno);
+            MPIR_Comm_release(*newcomm_ptr, 0);
+        }
 
         mpi_errno = MPIR_Allreduce_group(MPI_IN_PLACE, &errflag, 1, MPI_INT, MPI_MAX, comm_ptr,
-            new_group_ptr, MPIR_SHRINK_TAG, &tmp_errflag);
+            new_group_ptr, MPIR_SHRINK_TAG, &errflag);
         MPIR_Group_release(new_group_ptr);
 
         if (errflag) MPIU_Object_set_ref(new_group_ptr, 0);

src/mpid/ch3/src/mpid_comm_get_all_failed_procs.c

@@ -33,7 +33,8 @@ static int *group_to_bitarray(MPID_Group *group, MPID_Comm *orig_comm) {
     MPIR_Group_translate_ranks_impl(group, group->size, group_ranks,
                                     orig_comm->local_group, comm_ranks);
 
-    for (i = 0; i < group->size && comm_ranks[i] != MPI_UNDEFINED; i++) {
+    for (i = 0; i < group->size ; i++) {
+        if (comm_ranks[i] == MPI_UNDEFINED) continue;
         index = comm_ranks[i] / 32;
         mask = 0x80000000 >> comm_ranks[i] % 32;
         bitarray[index] |= mask;
@@ -128,7 +129,7 @@ int MPID_Comm_get_all_failed_procs(MPID_Comm *comm_ptr, MPID_Group **failed_grou
             /* Send the list to each rank to be processed locally */
             mpi_errno = MPIC_Send(bitarray, bitarray_size, MPI_UINT32_T, i,
                 tag, comm_ptr->handle, &errflag);
-            if (mpi_errno) errflag = 1;
+            if (mpi_errno) errflag = MPIR_ERR_PROC_FAILED;
         }
 
         /* Convert the bitarray into a group */
@@ -137,12 +138,10 @@ int MPID_Comm_get_all_failed_procs(MPID_Comm *comm_ptr, MPID_Group **failed_grou
         /* Send my bitarray to rank 0 */
         mpi_errno = MPIC_Send(bitarray, bitarray_size, MPI_UINT32_T, 0,
             tag, comm_ptr->handle, &errflag);
-        if (mpi_errno) errflag = 1;
 
         /* Get the resulting bitarray back from rank 0 */
         mpi_errno = MPIC_Recv(remote_bitarray, bitarray_size, MPI_UINT32_T, 0,
             tag, comm_ptr->handle, MPI_STATUS_IGNORE, &errflag);
-        if (mpi_errno) errflag = 1;
 
         /* Convert the bitarray into a group */
         *failed_group = bitarray_to_group(comm_ptr, remote_bitarray);