Title: [217055] trunk/Source/WTF
- Revision
- 217055
- Author
- [email protected]
- Date
- 2017-05-18 11:56:15 -0700 (Thu, 18 May 2017)
Log Message
AutomaticThread should wait longer before timing out
https://bugs.webkit.org/show_bug.cgi?id=172292
Reviewed by Filip Pizlo.
Increased the idle timeout from 1s => 10s.
This reduces the number of thread restarts on JetStream from
~150 => ~0. It also matches other thread pool APIs on Darwin.
Intuitively, it seems wrong for helper threads to idle exit during
hardcore benchmarking.
This patch in combination with a bmalloc fix seems to be a 1%-2% JetStream
speedup on my Mac Pro.
A nice side-benefit is that per-thread traces are easier to read.
* wtf/AutomaticThread.cpp:
(WTF::AutomaticThread::start):
* wtf/AutomaticThread.h:
Modified Paths
Diff
Modified: trunk/Source/WTF/ChangeLog (217054 => 217055)
--- trunk/Source/WTF/ChangeLog 2017-05-18 18:55:45 UTC (rev 217054)
+++ trunk/Source/WTF/ChangeLog 2017-05-18 18:56:15 UTC (rev 217055)
@@ -1,3 +1,27 @@
+2017-05-18 Geoffrey Garen <[email protected]>
+
+ AutomaticThread should wait longer before timing out
+ https://bugs.webkit.org/show_bug.cgi?id=172292
+
+ Reviewed by Filip Pizlo.
+
+ Increased the idle timeout from 1s => 10s.
+
+ This reduces the number of thread restarts on JetStream from
+ ~150 => ~0. It also matches other thread pool APIs on Darwin.
+
+ Intuitively, it seems wrong for helper threads to idle exit during
+ hardcore benchmarking.
+
+ This patch in combination with a bmalloc fix seems to be a 1%-2% JetStream
+ speedup on my Mac Pro.
+
+ A nice side-benefit is that per-thread traces are easier to read.
+
+ * wtf/AutomaticThread.cpp:
+ (WTF::AutomaticThread::start):
+ * wtf/AutomaticThread.h:
+
2017-05-18 Don Olmstead <[email protected]>
[Win] Remove usage of _snprintf
Modified: trunk/Source/WTF/wtf/AutomaticThread.cpp (217054 => 217055)
--- trunk/Source/WTF/wtf/AutomaticThread.cpp 2017-05-18 18:55:45 UTC (rev 217054)
+++ trunk/Source/WTF/wtf/AutomaticThread.cpp 2017-05-18 18:56:15 UTC (rev 217055)
@@ -194,10 +194,11 @@
if (result == PollResult::Stop)
return stopPermanently(locker);
RELEASE_ASSERT(result == PollResult::Wait);
- // Shut the thread down after one second.
+
+ // Shut the thread down after a timeout.
m_isWaiting = true;
bool awokenByNotify =
- m_waitCondition.waitFor(*m_lock, 1_s);
+ m_waitCondition.waitFor(*m_lock, 10_s);
if (verbose && !awokenByNotify && !m_isWaiting)
dataLog(RawPointer(this), ": waitFor timed out, but notified via m_isWaiting flag!\n");
if (m_isWaiting) {
Modified: trunk/Source/WTF/wtf/AutomaticThread.h (217054 => 217055)
--- trunk/Source/WTF/wtf/AutomaticThread.h 2017-05-18 18:55:45 UTC (rev 217054)
+++ trunk/Source/WTF/wtf/AutomaticThread.h 2017-05-18 18:56:15 UTC (rev 217055)
@@ -54,7 +54,7 @@
// to this thread sitting around even when it is not needed.
//
// AutomaticThread is here to help you in these situations. It encapsulates a lock, a condition
-// variable, and a thread. It will automatically shut the thread down after 1 second of inactivity.
+// variable, and a thread. It will automatically shut the thread down after a timeout of inactivity.
// You use AutomaticThread by subclassing it, and put any state that is needed between [1] and [2]
// in the subclass.
//
@@ -101,7 +101,7 @@
class WTF_EXPORT_PRIVATE AutomaticThread : public ThreadSafeRefCounted<AutomaticThread> {
public:
// Note that if you drop all of your references to an AutomaticThread then as soon as there is a
- // second during which it doesn't get woken up, it will simply die on its own. This is a
+ // timeout during which it doesn't get woken up, it will simply die on its own. This is a
// permanent kind of death where the AutomaticThread object goes away, rather than the temporary
// kind of death where AutomaticThread lives but its underlying thread dies. All you have to do
// to prevent permanent death is keep a ref to AutomaticThread. At time of writing, every user of
_______________________________________________
webkit-changes mailing list
[email protected]
https://lists.webkit.org/mailman/listinfo/webkit-changes
