Improve EventCount used by the non-blocking threadpool.

The current algorithm requires threads to commit/cancel waiting in order
they called Prewait. Spinning caused by that serialization can consume
lots of CPU time on some workloads. Restructure the algorithm to not
require that serialization and remove spin waits from Commit/CancelWait.
Note: this reduces max number of threads from 2^16 to 2^14 to leave
more space for ABA counter (which is now 22 bits).
Implementation details are explained in comments.

1 files changed, 5 insertions, 5 deletions

diff --git a/unsupported/test/cxx11_eventcount.cpp b/unsupported/test/cxx11_eventcount.cpp
index 2f1418684..3ca8598c7 100644
--- a/unsupported/test/cxx11_eventcount.cpp
+++ b/unsupported/test/cxx11_eventcount.cpp
@@ -30,11 +30,11 @@ static void test_basic_eventcount()
   EventCount ec(waiters);
   EventCount::Waiter& w = waiters[0];
   ec.Notify(false);
-  ec.Prewait(&w);
+  VERIFY(ec.Prewait());
   ec.Notify(true);
   ec.CommitWait(&w);
-  ec.Prewait(&w);
-  ec.CancelWait(&w);
+  VERIFY(ec.Prewait());
+  ec.CancelWait();
 }
 
 // Fake bounded counter-based queue.
@@ -112,7 +112,7 @@ static void test_stress_eventcount()
         unsigned idx = rand_reentrant(&rnd) % kQueues;
         if (queues[idx].Pop()) continue;
         j--;
-        ec.Prewait(&w);
+        if (!ec.Prewait()) continue;
         bool empty = true;
         for (int q = 0; q < kQueues; q++) {
           if (!queues[q].Empty()) {
@@ -121,7 +121,7 @@ static void test_stress_eventcount()
           }
         }
         if (!empty) {
-          ec.CancelWait(&w);
+          ec.CancelWait();
           continue;
         }
         ec.CommitWait(&w);