test: add allreduce test of MPIX_BFLOAT16

Author: hzhou

test/mpi/configure.ac

@@ -723,6 +723,16 @@ AC_CACHE_CHECK([whether long long is supported],pac_cv_have_long_long,[
 if test "$pac_cv_have_long_long" = yes ; then
     AC_DEFINE(HAVE_LONG_LONG,1,[Define if compiler supports long long])
 fi
+
+#
+# Check for _Float16
+AC_MSG_CHECKING([whether _Float16 is supported])
+AC_COMPILE_IFELSE([AC_LANG_PROGRAM([],[_Float16 a;])],[pac_cv_have_float16=yes],[pac_cv_have_float16=no])
+AC_MSG_RESULT($pac_cv_have_float16)
+if test "$pac_cv_have_float16" = "yes" ; then
+    AC_DEFINE(HAVE_FLOAT16,1,[Define if _Float16 is supported])
+fi
+
 #
 # Check for const and restrict (used in some of the performance tests)
 AC_C_CONST

test/mpi/impls/mpich/misc/Makefile.am

@@ -12,4 +12,5 @@ EXTRA_DIST = testlist
 ## correctly
 noinst_PROGRAMS = \
 	type_iov \
+	type_float16 \
 	gpu_query

test/mpi/impls/mpich/misc/testlist

@@ -1 +1,2 @@
 gpu_query 1
+type_float16 4

test/mpi/impls/mpich/misc/type_float16.c

@@ -0,0 +1,44 @@
+/*
+ * Copyright (C) by Argonne National Laboratory
+ *     See COPYRIGHT in top-level directory
+ */
+
+#include "mpitest.h"
+#include <stdio.h>
+#include <stdlib.h>
+#include <assert.h>
+
+#define N 4
+MPI_Comm comm = MPI_COMM_WORLD;
+
+int main(int argc, char *argv[])
+{
+    int errs = 0;
+    MTest_Init(&argc, &argv);
+
+    /* Test MPIX_BFLOAT16 */
+    unsigned short buf[N];
+    for (int i = 0; i < N; i++) {
+        buf[i] = 0x3f80;        /* 1.0 in bfloat16 */
+    }
+    MPI_Allreduce(MPI_IN_PLACE, buf, N, MPIX_BFLOAT16, MPI_SUM, comm);
+
+    MTestPrintfMsg(1, "Allreduce %d MPIX_BFLOAT16 (val=0.1), result: 0x%x, 0x%x, ...\n",
+                   N, buf[0], buf[1]);
+
+    /* Test MPIX_C_FLOAT16 */
+#ifdef HAVE_FLOAT16
+    _Float16 buf_f16[N];
+    for (int i = 0; i < N; i++) {
+        buf_f16[i] = 1.0;
+    }
+    MPI_Allreduce(MPI_IN_PLACE, buf_f16, N, MPIX_C_FLOAT16, MPI_SUM, comm);
+
+    MTestPrintfMsg(1, "Allreduce %d MPIX_C_FLOAT16 (val=0.1), result: %f, %f, ...\n",
+                   N, (float) buf_f16[0], (float) buf_f16[1]);
+#endif
+
+    MTest_Finalize(errs);
+
+    return MTestReturnValue(errs);
+}