[Bug 36692] New: Bad performance when combineng SetEvent / WaitForSingleObject for synchronizing worker threads

Bug ID: 36692 Summary: Bad performance when combineng SetEvent / WaitForSingleObject for synchronizing worker threads Product: Wine Version: 1.6.2 Hardware: x86-64 OS: Linux Status: UNCONFIRMED Severity: normal Priority: P2 Component: kernel32 Assignee: wine-bugs@winehq.org Reporter: milasudril@gmail.com

In a 2d simulation program, each worker thread has its own pair of events. One event is used to signal master thread that the worker thread is ready. The other is used by the master thread to signal that the worker thread may continue.

In Master thread:

while(!m_stop) { auto proc_ptr=processors.begin(); while(proc_ptr!=processors.end()) { // Call SetEvent on "start" event object owned by the object pointed // to by proc_ptr proc_ptr->frameNext(); ++proc_ptr; } proc_ptr=processors.begin(); while(proc_ptr!=processors.end()) { // Call WaitForSingleObject on "ready" event object owned by // the object pointed to by proc_ptr proc_ptr->wait(); ++proc_ptr; }

++framecounter; }

In worker thread:

while(!m_stop) { // Wait for master thread signaling start event (Calls WaitForSingleObject) start.wait();

m_model->process(m_framecounter,m_buffers[0].first ,m_buffers[0].second,m_offset); swap(m_buffers[0],m_buffers[1]);

// Signal master thread that we are ready for next frame (Calls SetEvent) ready.set(); }

On Wine, the framerate is half of that on Windows 7 on the same machine Intel(R) Core(TM) i5-3570K CPU @ 3.40GHz. Also there seems to be a huge different workload on differnt cores when running under Wine.

I realize that SetEvent/WaitForSingleObject are heavy functions on Windows too as they need kernel assistance, so it might be hard to make it perform better.

-- Do not reply to this email, post in Bugzilla using the above URL to reply. You are receiving this mail because: You are watching all bug changes.

Show replies by date

wine-bugs＠winehq.org

7 Jun 7 Jun

7:03 p.m.

New subject: [Bug 36692] Bad performance when combineng SetEvent / WaitForSingleObject for synchronizing worker threads

milasudril@gmail.com changed:

What |Removed |Added ---------------------------------------------------------------------------- CC| |milasudril@gmail.com

-- Do not reply to this email, post in Bugzilla using the above URL to reply. You are receiving this mail because: You are watching all bug changes.

wine-bugs＠winehq.org

8:07 p.m.

New subject: [Bug 36692] Bad performance when combineng SetEvent / WaitForSingleObject for synchronizing worker threads

--- Comment #1 from Bruno Jesus 00cpxxx@gmail.com --- Please attach a source code sample that can reproduce the issue.

-- Do not reply to this email, post in Bugzilla using the above URL to reply. You are receiving this mail because: You are watching all bug changes.

wine-bugs＠winehq.org

11:12 p.m.

New subject: [Bug 36692] Bad performance when combineng SetEvent / WaitForSingleObject for synchronizing worker threads

Sebastian Lackner sebastian@fds-team.de changed:

What |Removed |Added ---------------------------------------------------------------------------- CC| |sebastian@fds-team.de

--- Comment #2 from Sebastian Lackner sebastian@fds-team.de --- I would like to point out the following wine-devel discussion about the same topic: http://www.winehq.org/pipermail/wine-devel/2014-April/104088.html

Especially this comment by

* Vincent Povirk, that exaplains why it is so difficult to optimize the speed of SetEvent/WaitForMultipleObjects: http://www.winehq.org/pipermail/wine-devel/2014-April/104090.html

* and this comment by me, which explains the differences between the different sets of synchronization primitives: http://www.winehq.org/pipermail/wine-devel/2014-April/104092.html

At my opinion its very difficult to optimize the speed of these specific functions, but as Bruno already wrote, some (minimal) test code would probably be useful - especially to measure how "bad" the performance really is compared to Windows.

-- Do not reply to this email, post in Bugzilla using the above URL to reply. You are receiving this mail because: You are watching all bug changes.

wine-bugs＠winehq.org

8 Jun 8 Jun

7:34 a.m.

New subject: [Bug 36692] Bad performance when combineng SetEvent / WaitForSingleObject for synchronizing worker threads

--- Comment #3 from milasudril@gmail.com --- Created attachment 48735 --> http://bugs.winehq.org/attachment.cgi?id=48735 Sim2d app

Here is Sim2d prerelease

-- Do not reply to this email, post in Bugzilla using the above URL to reply. You are receiving this mail because: You are watching all bug changes.

wine-bugs＠winehq.org

7:37 a.m.

New subject: [Bug 36692] Bad performance when combineng SetEvent / WaitForSingleObject for synchronizing worker threads

--- Comment #4 from milasudril@gmail.com ---

At my opinion its very difficult to optimize the speed of these specific functions, but as Bruno already wrote, some (minimal) test code would probably be useful - especially to measure how "bad" the performance really is compared to Windows.

I guessed that they could be tricky. On Windows, I measure 200 kframes/s when using empty model code, and that is quite slow too. That is when the _virtual_ Sim2d::ModelState::process returns without processing its data block. For the Gray-Scott model I get 2kframes/s in Windows on 256x256 pixels.

I will push the project to github when I have completed the remaining parts of the code:

* Model parameter dialog * Initial condition loading * State matrix export * System state load * System state save

The tool can be useful as benchmark. For example, it is possible to load different models, and set the size of problem domain. The only blocker for building a native Linux port is the gui wrapper, and this guy is tricky since different notions of the Window concept. Gui is not useful when running benchmarks though.

-- Do not reply to this email, post in Bugzilla using the above URL to reply. You are receiving this mail because: You are watching all bug changes.

wine-bugs＠winehq.org

9 Jun 9 Jun

6:47 a.m.

New subject: [Bug 36692] Bad performance when combineng SetEvent / WaitForSingleObject for synchronizing worker threads

--- Comment #5 from milasudril@gmail.com --- Just an idea: What about taking advantage of anonymous autoreset event?

-- Do not reply to this email, post in Bugzilla using the above URL to reply. You are receiving this mail because: You are watching all bug changes.

wine-bugs＠winehq.org

3:39 p.m.

New subject: [Bug 36692] Bad performance when combineng SetEvent / WaitForSingleObject for synchronizing worker threads

--- Comment #6 from Vincent Povirk madewokherd@gmail.com --- DuplicateHandle can still pull an anonymous event handle out of another process and start using it.

So, you either need the server to be able to tell the process that controls a handle that it's being moved into the server, or you need a more efficient way of doing this that works across processes.

-- Do not reply to this email, post in Bugzilla using the above URL to reply. You are receiving this mail because: You are watching all bug changes.

wine-bugs＠winehq.org

13 Jun 13 Jun

5:01 p.m.

New subject: [Bug 36692] Bad performance when combineng SetEvent / WaitForSingleObject for synchronizing worker threads

--- Comment #7 from milasudril@gmail.com --- (In reply to Vincent Povirk from comment #6)

DuplicateHandle can still pull an anonymous event handle out of another process and start using it.

So, you either need the server to be able to tell the process that controls a handle that it's being moved into the server, or you need a more efficient way of doing this that works across processes.

What about letting DuplicateHandle change the behaviour of the event functions. If an event is created with

CreateEvent(NULL,0,0,NULL);

set vtable pointers associated with the event so they point to an implementation like pevents. When something is done that this implementation cannot do (DuplicateHandle, I guess WaitForMultipleObjects too since that function can take an hetrogenous array of waitable objects), switch to current implementation. This implies that all tricky functions need to copy the state of the event and create it inside wineserver. A drawback is that there will probably be a need for a mutex when this transfer is done.

-- Do not reply to this email, post in Bugzilla using the above URL to reply. You are receiving this mail because: You are watching all bug changes.

wine-bugs＠winehq.org

14 Jun 14 Jun

11:43 a.m.

New subject: [Bug 36692] Bad performance when combineng SetEvent / WaitForSingleObject for synchronizing worker threads

milasudril@gmail.com changed:

What |Removed |Added ---------------------------------------------------------------------------- Severity|normal |enhancement

-- Do not reply to this email, post in Bugzilla using the above URL to reply. You are receiving this mail because: You are watching all bug changes.

wine-bugs＠winehq.org

22 Jun 22 Jun

2:01 p.m.

New subject: [Bug 36692] Bad performance when combineng SetEvent / WaitForSingleObject for synchronizing worker threads

--- Comment #8 from milasudril@gmail.com --- The entire Sim2d is now on my Github

-- Do not reply to this email, post in Bugzilla using the above URL to reply. You are receiving this mail because: You are watching all bug changes.

wine-bugs＠winehq.org

4:49 p.m.

New subject: [Bug 36692] Bad performance when combineng SetEvent / WaitForSingleObject for synchronizing worker threads

--- Comment #9 from Bruno Jesus 00cpxxx@gmail.com --- (In reply to milasudril from comment #8)

The entire Sim2d is now on my Github

Please add the link of your github to the URL field and add the download tag.

-- Do not reply to this email, post in Bugzilla using the above URL to reply. You are receiving this mail because: You are watching all bug changes.

wine-bugs＠winehq.org

23 Jun 23 Jun

9:45 a.m.

New subject: [Bug 36692] Bad performance when combineng SetEvent / WaitForSingleObject for synchronizing worker threads

milasudril@gmail.com changed:

-- Do not reply to this email, post in Bugzilla using the above URL to reply. You are receiving this mail because: You are watching all bug changes.

wine-bugs＠winehq.org

11:41 a.m.

New subject: [Bug 36692] Bad performance when combineng SetEvent / WaitForSingleObject for synchronizing worker threads

--- Comment #10 from milasudril@gmail.com --- (In reply to Bruno Jesus from comment #9)

(In reply to milasudril from comment #8)

...
The entire Sim2d is now on my Github

Please add the link of your github to the URL field and add the download tag.

Done. I also added benchmark app without UI. Hope the build system does not complicate things. For me it works well.

-- Do not reply to this email, post in Bugzilla using the above URL to reply. You are receiving this mail because: You are watching all bug changes.

wine-bugs＠winehq.org

9 Sep 9 Sep

8:38 a.m.

New subject: [Bug 36692] Bad performance when combineng SetEvent / WaitForSingleObject for synchronizing worker threads

--- Comment #11 from Austin English austinenglish@gmail.com --- This is your friendly reminder that there has been no bug activity for over a year. Is this still an issue in current (1.7.51 or newer) wine?

-- Do not reply to this email, post in Bugzilla using the above URL to reply. You are receiving this mail because: You are watching all bug changes.

wine-bugs＠winehq.org

28 Oct 28 Oct

11:12 p.m.

New subject: [Bug 36692] Bad performance when combineng SetEvent / WaitForSingleObject for synchronizing worker threads

Chris Cochran chris.cochran.protoncore@gmail.com changed:

-- Do not reply to this email, post in Bugzilla using the above URL to reply. You are receiving this mail because: You are watching all bug changes.

wine-bugs＠winehq.org

29 Oct 29 Oct

1:48 a.m.

New subject: [Bug 36692] Bad performance when combineng SetEvent / WaitForSingleObject for synchronizing worker threads

--- Comment #12 from Alexandre Julliard julliard@winehq.org --- *** Bug 39521 has been marked as a duplicate of this bug. ***

-- Do not reply to this email, post in Bugzilla using the above URL to reply. You are receiving this mail because: You are watching all bug changes.

wine-bugs＠winehq.org

9:38 p.m.

New subject: [Bug 36692] Bad performance when combineng SetEvent / WaitForSingleObject for synchronizing worker threads

--- Comment #13 from Chris Cochran chris.cochran.protoncore@gmail.com --- Created attachment 52673 --> https://bugs.winehq.org/attachment.cgi?id=52673 ZIP file containing bug test program and testing notes

Enclosed is test software to demonstrate Bug 36692, as requested yesterday by Alexandre Julliard. All the necessary binaries and usage instructions are included in the ZIP. I am the author of all material being submitted and certify that it is harmless and safe to run on all systems.

-- Do not reply to this email, post in Bugzilla using the above URL to reply. You are receiving this mail because: You are watching all bug changes.

wine-bugs＠winehq.org

9:44 p.m.

New subject: [Bug 36692] Bad performance when combineng SetEvent / WaitForSingleObject for synchronizing worker threads

--- Comment #14 from Chris Cochran chris.cochran.protoncore@gmail.com --- See my additional comments in Bug 39521

-- Do not reply to this email, post in Bugzilla using the above URL to reply. You are receiving this mail because: You are watching all bug changes.

wine-bugs＠winehq.org

4 Apr 4 Apr

2:45 a.m.

New subject: [Bug 36692] Many multi-threaded applications have poor performance due to heavy use of synchronization primitives

Zebediah Figura z.figura12@gmail.com changed:

--- Comment #15 from Zebediah Figura z.figura12@gmail.com --- I have written a large set of patches that optimize synchronization primitives by reimplementing them on top of Linux's eventfd primitives. This patch set has shown a significant improvement in a fair number of applications, especially heavily multi-threaded games.

Per the request of the head Wine-Staging maintainer, Alistair Leslie-Hughes, I've added the patch set to Staging. It is not enabled by default, but can be enabled by setting the WINEESYNC environment variable to a non-zero value.

-- Do not reply to this email, post in Bugzilla using the above URL to reply. You are receiving this mail because: You are watching all bug changes.

wine-bugs＠winehq.org

5 Apr 5 Apr

7:07 p.m.

New subject: [Bug 36692] Many multi-threaded applications have poor performance due to heavy use of synchronization primitives

Anastasius Focht focht@gmx.net changed:

What |Removed |Added ---------------------------------------------------------------------------- CC| |focht@gmx.net

-- Do not reply to this email, post in Bugzilla using the above URL to reply. You are receiving this mail because: You are watching all bug changes.

WineHQ Bugzilla

6 Jun 6 Jun

10:56 p.m.

New subject: [Bug 36692] Many multi-threaded applications have poor performance due to heavy use of synchronization primitives

Zebediah Figura z.figura12@gmail.com changed:

What |Removed |Added ---------------------------------------------------------------------------- Status|STAGED |NEW

--- Comment #16 from Zebediah Figura z.figura12@gmail.com --- Currently the wine-staging "eventfd_synchronization" patch set is disabled. Due to the current and ongoing work in ntdll.so, it will need significant work to rebase.

-- Do not reply to this email, post in Bugzilla using the above URL to reply. You are receiving this mail because: You are watching all bug changes.

WineHQ Bugzilla

7 Jun 7 Jun

8:44 a.m.

New subject: [Bug 36692] Many multi-threaded applications have poor performance due to heavy use of synchronization primitives

Anya animegirl@stronzi.org changed:

What |Removed |Added ---------------------------------------------------------------------------- CC| |animegirl@stronzi.org

-- Do not reply to this email, post in Bugzilla using the above URL to reply. You are receiving this mail because: You are watching all bug changes.

WineHQ Bugzilla

6:17 p.m.

New subject: [Bug 36692] Many multi-threaded applications have poor performance due to heavy use of synchronization primitives

Gijs Vermeulen gijsvrm@gmail.com changed:

What |Removed |Added ---------------------------------------------------------------------------- CC| |winter.snowfall23@gmail.com

--- Comment #17 from Gijs Vermeulen gijsvrm@gmail.com --- *** Bug 49331 has been marked as a duplicate of this bug. ***

-- Do not reply to this email, post in Bugzilla using the above URL to reply. You are receiving this mail because: You are watching all bug changes.

WineHQ Bugzilla

6:18 p.m.

New subject: [Bug 36692] Many multi-threaded applications have poor performance due to heavy use of synchronization primitives

Gijs Vermeulen gijsvrm@gmail.com changed:

What |Removed |Added ---------------------------------------------------------------------------- CC| |shtetldik@gmail.com

--- Comment #18 from Gijs Vermeulen gijsvrm@gmail.com --- *** Bug 49337 has been marked as a duplicate of this bug. ***

-- Do not reply to this email, post in Bugzilla using the above URL to reply. You are receiving this mail because: You are watching all bug changes.

WineHQ Bugzilla

8:27 p.m.

New subject: [Bug 36692] Many multi-threaded applications have poor performance due to heavy use of synchronization primitives

rtentser@yandex.ru changed:

What |Removed |Added ---------------------------------------------------------------------------- CC| |rtentser@yandex.ru

-- Do not reply to this email, post in Bugzilla using the above URL to reply. You are receiving this mail because: You are watching all bug changes.

WineHQ Bugzilla

18 Jul 18 Jul

8:49 a.m.

New subject: [Bug 36692] Many multi-threaded applications have poor performance due to heavy use of synchronization primitives

--- Comment #19 from Anya animegirl@stronzi.org --- https://github.com/wine-staging/wine-staging/blob/master/patches/eventfd_syn...

-- Do not reply to this email, post in Bugzilla using the above URL to reply. You are receiving this mail because: You are watching all bug changes.

WineHQ Bugzilla

9 Jul 9 Jul

7:31 p.m.

New subject: [Bug 36692] Many multi-threaded applications have poor performance due to heavy use of synchronization primitives

soredake broaden_acid002@simplelogin.com changed:

-- Do not reply to this email, post in Bugzilla using the above URL to reply. You are receiving this mail because: You are watching all bug changes.

WineHQ Bugzilla

19 Jul 19 Jul

3:07 a.m.

New subject: [Bug 36692] Many multi-threaded applications have poor performance due to heavy use of synchronization primitives

Austin English austinenglish@gmail.com changed:

-- Do not reply to this email, post in Bugzilla using the above URL to reply. You are receiving this mail because: You are watching all bug changes.

WineHQ Bugzilla

3:07 a.m.

New subject: [Bug 36692] Many multi-threaded applications have poor performance due to heavy use of synchronization primitives

Austin English austinenglish@gmail.com changed:

What |Removed |Added ---------------------------------------------------------------------------- CC| |austinenglish@gmail.com

-- Do not reply to this email, post in Bugzilla using the above URL to reply. You are receiving this mail because: You are watching all bug changes.

WineHQ Bugzilla

24 Jan 24 Jan

5:46 p.m.

New subject: [Bug 36692] Many multi-threaded applications have poor performance due to heavy use of synchronization primitives