From 0107aafc87b61c0bcf64982c4fbd75a369137b3d Mon Sep 17 00:00:00 2001
From: Noah Evans <nevans@sandia.gov>
Date: Mon, 5 Dec 2016 10:26:37 -0700
Subject: [PATCH] Add thread ring benchmark

Add a thread ring benchmark to tune performance. Begins to address issue #48.
---
 test/benchmarks/.gitignore                 |  1 +
 test/benchmarks/Makefile.am                |  6 ++-
 test/benchmarks/generic/time_thread_ring.c | 58 ++++++++++++++++++++++
 3 files changed, 64 insertions(+), 1 deletion(-)
 create mode 100644 test/benchmarks/generic/time_thread_ring.c

diff --git a/test/benchmarks/.gitignore b/test/benchmarks/.gitignore
index 1b73419d3..57df6af94 100644
--- a/test/benchmarks/.gitignore
+++ b/test/benchmarks/.gitignore
@@ -42,6 +42,7 @@ time_task_spawn
 time_thrcrt_bench
 time_thrcrt_bench_pthread
 time_threading
+time_thread_ring
 time_uts_aligned
 time_uts_donecount
 time_uts_donecount2
diff --git a/test/benchmarks/Makefile.am b/test/benchmarks/Makefile.am
index 418417411..fbec33612 100644
--- a/test/benchmarks/Makefile.am
+++ b/test/benchmarks/Makefile.am
@@ -19,7 +19,9 @@ generic_benchmarks = \
                      time_halo_swap_all \
                      time_prodcons_comm \
                      time_qt_loops \
-                     time_qt_loopaccums
+                     time_qt_loopaccums \
+                     time_thread_ring
+
 thesis_benchmarks = \
                     time_allpairs \
                     time_wavefront
@@ -254,6 +256,8 @@ time_syncvar_producerconsumer_SOURCES = generic/time_syncvar_producerconsumer.c
 
 time_threading_SOURCES = generic/time_threading.c
 
+time_thread_ring_SOURCES = generic/time_thread_ring.c
+
 if COMPILE_OMP_BENCHMARKS
 time_threading_omp_SOURCES = generic/time_threading.omp.c
 time_threading_omp_CFLAGS = @OPENMP_CFLAGS@
diff --git a/test/benchmarks/generic/time_thread_ring.c b/test/benchmarks/generic/time_thread_ring.c
new file mode 100644
index 000000000..10871f363
--- /dev/null
+++ b/test/benchmarks/generic/time_thread_ring.c
@@ -0,0 +1,58 @@
+#include <stdio.h>
+#include <stdlib.h>
+#include <assert.h>
+#include <qthread/qthread.h>
+#include <qthread/qloop.h>
+#include <qthread/qtimer.h>
+
+
+// native qthreads version of thread-ring, based on Chapel's release version
+//   https://github.com/chapel-lang/chapel/blob/master/test/release/examples/benchmarks/shootout/threadring.chpl
+
+//static int n = 1000, ntasks = 503;
+static int n = 50000000, ntasks = 503;
+static aligned_t *mailbox;
+
+static void passTokens(size_t start, size_t stop, void* arg) {
+  uint64_t id = start;
+
+  // printf("entering tokens\n");
+  uint64_t numPasses = 0;
+  do {
+    // printf("numpasses %lld n %d\n", numPasses, n);
+    qthread_readFE(&numPasses, &mailbox[id]);
+    qthread_writeEF_const(&mailbox[(id+1)%ntasks], numPasses+1);
+    if (numPasses == n) {
+      printf("%ld\n", (long)id+1);
+    }
+  } while (numPasses < n);
+  //  printf("exiting tokens\n");
+}
+
+static void init() {
+  assert(qthread_initialize() == 0);
+  printf("%i threads...\n", qthread_num_workers());
+
+  // init array of syncvars (code grabbed from stress/feb_stream.c)
+  mailbox = malloc(ntasks * sizeof(aligned_t));
+  for (int i=0; i<ntasks; i++) {
+    mailbox[i] = 0;
+    qthread_empty(&mailbox[i]);
+  }
+}
+
+int main(int argc, char **argv) {
+  qtimer_t timer = qtimer_create();
+  double ring_time;
+  init();
+
+  qtimer_start(timer);
+  qthread_writeEF_const(&mailbox[0], 0);
+  qt_loop_simple(0, ntasks, passTokens, NULL);
+  qtimer_stop(timer);
+  ring_time = qtimer_secs(timer);
+
+  printf("\tThread ring time: %f usecs, %f/sec\n", 1000000 * ring_time / ntasks, ntasks / ring_time);
+
+  return 0;
+}