test/mpi: Add persistent collective ordering test

MPI-4.0 states that persistent collective operations may be started
out-of-order, as the matching requirements are satisfied during
initialization. Add a simple test to verify this capability.

test/mpi/.gitignore

@@ -145,6 +145,7 @@
 /coll/p_allgatherv
 /coll/p_alltoallv
 /coll/p_alltoallw
+/coll/p_order
 /coll/red3
 /coll/red4
 /coll/red_scat_block

test/mpi/coll/Makefile.am

@@ -104,6 +104,7 @@ noinst_PROGRAMS =      \
     p_neighb_alltoall  \
     p_neighb_alltoallv \
     p_neighb_alltoallw \
+    p_order            \
     red3               \
     red4               \
     red_scat_block     \

test/mpi/coll/p_order.c

@@ -0,0 +1,53 @@
+/*
+ * Copyright (C) by Argonne National Laboratory
+ *     See COPYRIGHT in top-level directory
+ */
+
+#include "mpi.h"
+#include "mpitest.h"
+
+/*
+static char MTEST_Descrip[] = "Test that persistent collectives can be started out of matching order";
+*/
+
+int main(int argc, char **argv)
+{
+    int errs = 0;
+
+    MTest_Init(&argc, &argv);
+
+    int size, rank;
+    MPI_Comm_size(MPI_COMM_WORLD, &size);
+    MPI_Comm_rank(MPI_COMM_WORLD, &rank);
+
+    if (size < 2) {
+        fprintf(stderr, "This test requires at least two processes.");
+        MPI_Abort(MPI_COMM_WORLD, 1);
+    }
+
+    int a = -1;
+    MPI_Request requests[2];
+    MPI_Bcast_init(&a, 1, MPI_INT, 0, MPI_COMM_WORLD, MPI_INFO_NULL, &requests[0]);
+    MPI_Barrier_init(MPI_COMM_WORLD, MPI_INFO_NULL, &requests[1]);
+
+    if (rank == 0) {
+        a = 42;
+        MPI_Start(&requests[0]);
+        MPI_Start(&requests[1]);
+    } else {
+        MPI_Start(&requests[1]);
+        MPI_Start(&requests[0]);
+    }
+
+    MPI_Waitall(2, requests, MPI_STATUSES_IGNORE);
+
+    if (a != 42) {
+        errs++;
+    }
+
+    MPI_Request_free(&requests[0]);
+    MPI_Request_free(&requests[1]);
+
+    MTest_Finalize(errs);
+    return MTestReturnValue(errs);
+}

test/mpi/coll/testlist.in

@@ -197,3 +197,4 @@ p_neighb_allgatherv 4 strict=false
 p_neighb_alltoall 4 strict=false
 p_neighb_alltoallv 4 strict=false
 p_neighb_alltoallw 4 strict=false
+p_order 2 strict=false