From 0080ee69d552c8c50a16c0aaafe2ab97b97c7c7e Mon Sep 17 00:00:00 2001
From: Nadav Har'El <nyh@cloudius-systems.com>
Date: Thu, 18 Jul 2013 08:00:27 +0300
Subject: [PATCH] Micro-benchmark for waking condvar on which no-one is waiting

This patch adds to tst-condvar two benchmark for measuring
condvar::wake_all() on a condvar that nobody is waiting on.

The first benchmark does these wakes from a single thread, measuring
26ns before commit 3509b19bb5c7acc5d97559a44e66e007133b647b, and
only 3ns after it.

The second benchmark does wake_all() loops from two threads on two
different CPUs. Before the aforementioned commit, this frequently
involved a contented mutex and context switches, with as much as
30,000 ns delay. After that commit, this benchmark measures 3ns,
the same as the single-threaded benchmark.
---
 tests/tst-condvar.cc | 38 ++++++++++++++++++++++++++++++++++++++
 1 file changed, 38 insertions(+)

diff --git a/tests/tst-condvar.cc b/tests/tst-condvar.cc
index 42d7064df6..677922055a 100644
--- a/tests/tst-condvar.cc
+++ b/tests/tst-condvar.cc
@@ -1,5 +1,8 @@
+#include <atomic>
+
 #include "sched.hh"
 #include "debug.hh"
+#include <drivers/clock.hh>
 
 #include <osv/condvar.h>
 
@@ -131,6 +134,41 @@ int main(int argc, char **argv)
     }
     assert_idle(&cond);
 
+    debug("Measuring unwaited wake_all (one thread): ");
+    int iterations = 100000000;
+    condvar cv;
+    auto start = nanotime();
+    for (int i = 0; i < iterations; i++) {
+        cv.wake_all();
+    }
+    auto end = nanotime();
+    debug ("%d ns\n", (end-start)/iterations);
+
+
+    debug("Measuring unwaited wake_all (two threads): ");
+    iterations = 100000000;
+    unsigned int nthreads = 2;
+    assert(sched::cpus.size() >= nthreads);
+    sched::thread *threads2[nthreads];
+    std::atomic<u64> time(0);
+    for(unsigned int i = 0; i < nthreads; i++) {
+        threads2[i]= new sched::thread([iterations, &cv, &time] {
+            auto start = nanotime();
+            for (int j = 0; j < iterations; j++) {
+                cv.wake_all();
+            }
+            auto end = nanotime();
+            time += (end-start);
+        }, sched::thread::attr(sched::cpus[i]));
+        threads2[i]->start();
+    }
+    for(unsigned int i = 0; i < nthreads; i++) {
+        threads2[i]->join();
+        delete threads2[i];
+    }
+    debug ("%d ns\n", time/iterations/nthreads);
+
+
     debug("condvar tests succeeded\n");
     return 0;
 }