test/bench: add bcast bench test

Adding bcast bench test. It will first measure the barrier overhead and
account for the barrier usage in measurement. It also runs auto-warm-up
by using bcast to synchronize all processes.

test/mpi/bench/.gitignore

@@ -1,3 +1,4 @@
 /*.c
 /p2p_bw
 /p2p_latency
+/bcast

test/mpi/bench/Makefile.am

@@ -11,7 +11,8 @@ LDADD += -lm
 ## correctly
 noinst_PROGRAMS = \
     p2p_latency \
-    p2p_bw
+    p2p_bw \
+    bcast
 
 .def.c:
 	mydef_page $<

test/mpi/bench/bcast.def

@@ -0,0 +1,20 @@
+include: macros/bench_frame.def
+include: macros/bench_coll.def
+include: macros/mtest.def
+
+page: bcast, bench_frame
+    data: buf, size, MPI_CHAR
+
+    tf_barrier = bench_barrier(comm)
+    $for int size = 1; size < MAX_BUFSIZE; size *= 2
+        &call coll_warmup, 0
+            MPI_Bcast($(data), 0, comm)
+            MPI_Barrier(comm)
+        &call run_stat, NUM_REPEAT, tf_latency
+            &call measure, iter
+                MPI_Bcast($(data), 0, comm)
+                MPI_Barrier(comm)
+            tf_latency = (tf_dur / iter) - tf_barrier
+        $if grank == 0
+            $call report_latency, size, 1
+

test/mpi/bench/macros/bench_coll.def

@@ -0,0 +1,32 @@
+subcode: coll_warmup(root)
+    $if grank == 0
+        &call warm_up, iter, tf_dur
+            MPI_Bcast(&iter, 1, MPI_INT, $(root), comm)
+            &call measure, iter
+                BLOCK
+        tn_zero = 0
+        MPI_Bcast(&tn_zero, 1, MPI_INT, $(root), comm)
+    $else
+        $while 1
+            MPI_Bcast(&iter, 1, MPI_INT, $(root), comm)
+            $if iter == 0 
+                break
+            $for 0:iter
+                BLOCK
+
+    MPI_Bcast(&iter, 1, MPI_INT, $(root), comm)
+
+fncode: bench_barrier(comm)
+    $local int iter
+    &call coll_warmup, 0
+        MPI_Barrier(comm)
+
+    &call run_stat, NUM_REPEAT, tf_latency
+        &call measure, iter
+            MPI_Barrier(comm)
+        tf_latency = (tf_dur / iter)
+
+    $if grank == 0
+        printf("Barrier latency %.3f +/- %.3f us\n", sum1 * 1e6, sum2 * 1e6)
+    return sum1
+

test/mpi/bench/testlist

@@ -1,2 +1,3 @@
 p2p_latency 2 resultTest=TestBench
 p2p_bw 2 resultTest=TestBench
+bcast 16 resultTest=TestBench