QTimer is very inaccurate
-
@Pavankumar-S-V said in QTimer is very inaccurate:
So i tried to limit the cpu usage of main thread using linux utility "cpulimit" and limited the cpu usage of the main thread to 70%,80%..
That's not what you should do and I don't know why you think it could improve anything.
What I and @JonB asked is: is your main thread too busy? Did you check that? As @JonB pointed out if you're doing too much stuff in the slot then the next timeout will not cause a new slot call untill the current one finishes. You should really check what you are doing in main thread (why is the CPU usage so high?). It's your application you should know what you are doing. You can also use a profiler to see which parts of your code cause most CPU load. -
@JonB, @jsulm :
Yes understood. But as I mentioned earlier, I'm not doing any time consuming tasks inside the main thread or timerEvnt() from application point of view. So, I could not understand why the cpu load was too high. Qt might be doing something internally that might be causing high cpu load. I wanted to know what could it be.The Reason why I tried limiting the cpu usage of main thread:
Note: One of the main task of timerEvnt() is to exchange data with UI objects of the currently displayed screen.
Because of the inaccuracy of the QTimer, I am seeing sluggishness when interacting with the UI(This is the actual issue that I'm facing). So, I'm trying to improve the accuracy of the Qtimer and see if it can improve the UI responsiveness.I read that "QSGRenderThread" thread is responsible for managing and controlling rendering operations for Qt Quick applications and also helps prevent the user interface from freezing or becoming unresponsive during resource-intensive graphical tasks.
As my main thread was using too much cpu , I thought "QSGRenderThread" is starving and reducing the cpu usage of main thread may help to get more cpu time for "QSGRenderThread" thereby increasing the performance of the graphics. But this did not happen.@jsulm Also you mentioned that profiling might help me get a good picture of what is happening in the application. Can you please share any link with good description of how to profile Embedded QT applications?
-
@Pavankumar-S-V You should really find out why you have such a high CPU load.
Either use a profiler or disable some code (comment out) until CPU load gets lower then you know which part of your code causes the CPU load."Because of the inaccuracy of the QTimer, I am seeing sluggishness when interacting with the UI" - I doubt it is "inaccuracy" of QTimer - as already explained if you have such high CPU load then it can delay slot calls. So, before blaming Qt check why you have so high CPU load...
-
@jsulm Ok, I will check which part of my application could be causing high cpu load.
If possible, can you please share any link with good description of how to profile Embedded QT applications?
Thanks a lot for the suggestions@jsulm , @JonB can creating a separate thread for timerEvnt() with a msleep(20) will help me to get good accuracy?
-
@Pavankumar-S-V said in QTimer is very inaccurate:
@jsulm , @JonB can creating a separate thread for timerEvnt() with a msleep(20) will help me to get good accuracy?
No.
-
@Pavankumar-S-V said in QTimer is very inaccurate:
can creating a separate thread for timerEvnt() with a msleep(20) will help me to get good accuracy?
I would say it depends whether the slot will be called in main thread or in the same thread where the timer is running. If in the same thread as timer then this could improve the situation (at least if the CPU has more than one core). But even then you should investigate the CPU load.
Profiling with GCC: https://www.thegeekstuff.com/2012/08/gprof-tutorial/
-
@jsulm
If you want a timer to tick at 20ms and do some processing I cannot see howmsleep(20)can do any good anywhere. At best it would occupy the whole desired 20ms interval, leaving no time to do any further processing...? Though you know better than I. -
@Pavankumar-S-V
You might like to look at the topic Small example showing QTimer slowing to a halt. The important take away is the OP's last post there, https://forum.qt.io/topic/94657/small-example-showing-qtimer-slowing-to-a-halt/16, where the gist isCurrently I have moved the logging functionality on a separate thread and this seems to work fine.
Something in your application might benefit from being moved to a thread if the main thread is too busy to keep up.
But to me this still does not resolve your CPU usage issue. If anything takes near 100% CPU time you have a potential problem keeping up....
-
@jsulm : I tried to profile the application using "gprof" as per your suggestion.
I followed the same procedure that is mentioned in the link shared by you. But the file that holds the profiled data(gmon.out) is not getting generated after running and exiting from the application.
Just to confirm, I followed the following steps:
1) Enabled -pg, -O0 in the compiler, linker and built the qt project.
2) I made the following modification to the main() so that the application returns from main() gracefully(This is one of the condition for gprof):int main(int argc, char *argv[]) { QApplication *app; app = new QApplication(argc, argv); MainClass maincls; maincls.view = new QQuickView; maincls.view->setSource(QUrl::fromLocalFile("Path of Loader.qml")); maincls.view->show(); //starting all the user threads maincls.view->installEventFilter(&maincls); exitTimer = new QTimer(); exitTimer->setSingleShot(true); QObject::connect(exitTimer, &QTimer::timeout, app, &QApplication::quit, Qt::QueuedConnection); exitTimer->start(85000); //main event loop should run for 85 sec and return to the main function after that. app->exec(); qDebug()<<"Exiting qt;"; //This message is printed successfully return 0; }The above procedure should have generated "gmon.out" file but it has not. I have used gprof before for C based application and successfully profiled it. But I'm facing this issue when profiling qt application using grpof.
- Please let me know if I'm missing something during profiling.
- Please let me know if gprof is a right profiling tool to profile the qt application. As per my experience with gprof before, it is not very suitable for multi threaded applications and also provides info related to only user space.
- Please guide me to the right way for profiling embedded qt applications as there is very limited information available online related to qt profiling. As I'm working on improving the performance of the qt based device, it would be very useful for me to profile the app.
Thank you
-
@Pavankumar-S-V said in QTimer is very inaccurate:
Enabled -pg, -O0 in the compiler, linker and built the qt project.
Can you show how?
-
@jsulm : In the .pro file of qt project, I added these flags like below:
QMAKE_CFLAGS += -Werror -O0 -pg
QMAKE_CXXFLAGS += -Werror -O0 -pgQMAKE_LFLAGS += -pg
-
@Pavankumar-S-V Did you do a complete rebuild?
- Delete build folder
- Run qmake
- Build
-
@jsulm : Yes I did a complete re-build.
We have a Makefile that calls qmake to generate a .mk for the build.
I have checked in that generated .mk file and made sure that these flags are reflected in CFLAGS, CXXFLAGS. So, these flags are indeed considered for the build as per my understanding. -
@Pavankumar-S-V
Run your executable understrace. Do you see any mention ofgmon.out?So, these flags are indeed considered for the build as per my understanding.
Probably. But I would inspect the actual commands issued, why not do so?
-
@Pavankumar-S-V can you try Qt::DirectConnection in signal and slot connection of your timer ?
Note that this may make your CPU very busy.
-
@JonB Thanks for your suggestion. I execution my qt app with strace((strace -c)) and following is the summary report:
% time seconds usecs/call calls errors syscall ------ ----------- ----------- --------- --------- ---------------- 60.56 2.065174 185 11146 2 futex 9.86 0.336357 9 33782 semop 6.68 0.227773 21 10454 9734 openat 4.65 0.158571 26 5944 11 write 4.64 0.158276 29 5451 read 2.37 0.080926 19 4138 mprotect 1.96 0.066935 185 360 munmap 1.43 0.048862 26 1812 1 mmap2 1.28 0.043518 14 2966 4 statx 1.18 0.040082 6 5979 259 rt_sigreturn 1.01 0.034415 12 2783 2677 recvmsg 0.53 0.017979 9 1961 cacheflush 0.52 0.017783 12 1395 1 poll 0.37 0.012461 10 1174 close 0.35 0.011980 11 1002 getcwd 0.35 0.011798 25 457 memfd_create 0.28 0.009419 672 14 clone 0.25 0.008616 21 397 395 readlink 0.24 0.008117 2029 4 uname 0.23 0.007893 9 813 _llseek 0.21 0.007148 26 273 78 stat64 0.21 0.007088 15 466 ftruncate 0.18 0.006122 19 318 130 access 0.16 0.005416 51 105 1 ioctl 0.15 0.005156 15 326 brk 0.09 0.003161 16 186 fstat64 0.07 0.002225 11 192 lstat64 0.06 0.001929 48 40 getdents64 0.05 0.001867 15 118 madvise 0.03 0.000971 57 17 sendmsg 0.02 0.000535 13 40 fcntl64 0.01 0.000254 127 2 socketpair 0.00 0.000165 23 7 semget 0.00 0.000127 6 20 rt_sigaction 0.00 0.000120 40 3 fallocate 0.00 0.000114 57 2 unlink 0.00 0.000095 47 2 sendfile 0.00 0.000078 5 15 getpid 0.00 0.000061 61 1 linkat 0.00 0.000046 46 1 statfs 0.00 0.000035 2 14 sched_get_priority_min 0.00 0.000034 34 1 sched_setaffinity 0.00 0.000033 2 14 sched_get_priority_max 0.00 0.000028 14 2 execve 0.00 0.000028 28 1 mremap 0.00 0.000027 3 7 sched_setscheduler 0.00 0.000024 24 1 semctl 0.00 0.000022 11 2 socket 0.00 0.000022 11 2 set_tls 0.00 0.000020 20 1 sched_getaffinity 0.00 0.000016 16 1 chmod 0.00 0.000016 4 4 gettid 0.00 0.000014 4 3 geteuid32 0.00 0.000013 13 1 setitimer 0.00 0.000012 2 6 rt_sigprocmask 0.00 0.000010 5 2 getuid32 0.00 0.000005 2 2 tgkill 0.00 0.000004 4 1 getgid32 0.00 0.000004 4 1 getegid32 0.00 0.000000 0 1 1 mkdir 0.00 0.000000 0 1 sigreturn 0.00 0.000000 0 10 writev 0.00 0.000000 0 1 fdatasync 0.00 0.000000 0 1 ugetrlimit 0.00 0.000000 0 1 set_tid_address 0.00 0.000000 0 1 connect 0.00 0.000000 0 1 set_robust_list 0.00 0.000000 0 1 eventfd2 ------ ----------- ----------- --------- --------- ---------------- 100.00 3.409980 94250 13294 totalAs per the above report, i don't see anything related to gprof.
- Do you think any specific system call should have been called that indicates the generation of gmon.out?
- Please let me know if gprof is a right profiling tool to profile the qt application. As per my experience with gprof before, it is not very suitable for multi threaded applications and also provides info related to only user space.
- Please let me know how to proceed furthur.
-
@JoeCFD : I tried to use"Qt::DirectConnection" in signal slot connection as per your suggestion. But somehow the whole application is crashing.
Also, QT documentation recommends to use QueuedConnection for quit() slot(as per below explaination):
It's good practice to always connect signals to this slot using a QueuedConnection. If a signal connected (non-queued) to this slot is emitted before control enters the main event loop (such as before "int main" calls exec()), the slot has no effect and the application never exits. Using a queued connection ensures that the slot will not be invoked until after control enters the main event loop.
