Change default --bind-to

For hybrid runs (OpenMP with OpenMPI) the process binding under MPI can
not be 'core' which prohibits OpenMP from running on more than one core.

examples/gpmdcov/run.sh

@@ -12,6 +12,7 @@ if ${MPIRUN} --version > /dev/null 2>&1; then
     ${MPIRUN} \
         -np ${NODES} \
         --map-by ${MAP} \
+        --bind-to board \
         --hostfile ~/hostfile \
         --mca plm_rsh_no_tree_spawn 1 \
         --mca orte_base_help_aggregate 0 \

examples/gpmdcov/test_openmp.c

@@ -14,27 +14,28 @@ int main(int argc, char **argv)
     int counter;
     int provided;
 
-    //MPI_Init(&argc, &argv);
-    MPI_Init_thread(&argc, &argv, MPI_THREAD_MULTIPLE, &provided);
+    MPI_Init(&argc, &argv);
+    //MPI_Init_thread(&argc, &argv, MPI_THREAD_MULTIPLE, &provided);
+    //MPI_Init_thread(&argc, &argv, MPI_THREAD_SERIALIZED, &provided);
     MPI_Comm_rank(MPI_COMM_WORLD, &rank);
     MPI_Comm_size(MPI_COMM_WORLD, &size);
 
     if (rank == 0) {
         printf("MPI size %d\n", size);
+        printf("MPI_THREAD_SERIALIZED = %d\n", MPI_THREAD_SERIALIZED);
         printf("MPI_THREAD_MULTIPLE = %d\n", MPI_THREAD_MULTIPLE);
         printf("provided = %d\n", provided);
     }
 
 #pragma omp parallel
     {
 #pragma omp master
-        printf("Running on %d threads\n", omp_get_num_threads());
+        printf("Rank %d, running on %d threads\n", rank, omp_get_num_threads());
     }
 
-    printf("MPI rank %d\n", rank);
 #pragma omp parallel private(counter)
     {
-        printf("Thread %d starting up\n", omp_get_thread_num());
+        printf("Rank %d, thread %d starting up\n", rank, omp_get_thread_num());
         while (1) {
             counter++;
             counter = counter%100;