How about the traceevent interrupt hook?  Then you can know everytime a different thread is scheduled.  I don't have the
 api in front if me right now, so I probably have the function name wrong. 

----- Original Message -----
From: Lichun Zhu [mailto:community-noreply@qnx.com]
Sent: Thursday, October 20, 2011 09:14 AM
To: ostech-core_os <post89479@community.qnx.com>
Subject: RE: Number of CPU cycles used by a thread

Are you looking for something like the ClockCycles() API call in your program?


-----Original Message-----
From: Gregoire Banderet [mailto:community-noreply@qnx.com] 
Sent: Thursday, October 20, 2011 5:10 AM
To: ostech-core_os
Subject: Re: Number of CPU cycles used by a thread

It's the kernel job to schedule the execution of the threads. So I guess it knows which threads is running at any time.
When you use the system profiler in Momentics you see the execution of the threads, it's then easy to calculate the 
number of microseconds of execution for a specific thread within a given time slot. I just need that programatically.

I'm currently porting some code written for INTime. INTime can reports the number of CPU cycles consumed by a thread 
even with a clock period of 125us.
It will make me very sad if QNX couldn't do what INTime can :-)

-- Greg





_______________________________________________

OSTech
http://community.qnx.com/sf/go/post89474




_______________________________________________

OSTech
http://community.qnx.com/sf/go/post89479

No. ClockCycles only reports the current CPU cycles counter value. The counter is global for all threads. I need one per
 thread.

The VM takes the measurements, not a program running inside the VM.
Every 250us the VM executes a program and I would like to know how many CPU cycles were used for the execution of this 
program.

The VM is really not an issue, I talked about it to explain the situation at customer's site.
I'll look at TraceEvent API.

-- Greg

> -----Message d'origine-----
> De : Colin Burgess [mailto:community-noreply@qnx.com]
> Envoyé : Friday, November 25, 2011 17:57
> À : ostech-core_os
> Objet : Re: Number of CPU cycles used by a thread
> 
> Please note that trying to take accurate timing measurements in a VM is not
> going to work very well!
> 
> On 11-11-25 11:53 AM, Elena Laskavaia wrote:
> > If you customer running instrumented kernel you can hook in to
> > tracelogger API. You would need to enable events on when thread
> > start/stop running and start tracelogging from the code of your
> > utility, then run custom "traceprinter" to aggregate the data
> >
> > On 11/25/2011 02:35 AM, Gregoire Banderet wrote:
> >> Hi,
> >>
> >> I'm not profiling my QNX executable. I'm implementing a feature for a
> client. The timings values are then reported to a Windows IDE through
> TCP/IP.
> >> The QNX executable is a multi-threaded virtual machine running the
> customer's PLC application. Our customers want to do some profiling of their
> application executed by the Virtual machine running on QNX.
> >>
> >> That means I need timings values of a granularity of 1 microseconds and
> this profiling feature must not increase the CPU load (1 or 2 percent of CPU
> power is acceptable).
> >> The system profiler is not available in customer site.
> >>
> >> Is there any way to use the tracing API in order to gather threads data? Is
> there events for the start and end of execution for every threads?
> >>
> >> -- Greg
> >>
> >>> -----Message d'origine-----
> >>> De : Elena Laskavaia [mailto:community-noreply@qnx.com]
> >>> Envoyé : Thursday, November 24, 2011 19:11 À : ostech-core_os Objet
> >>> : Re: Number of CPU cycles used by a thread
> >>>
> >>> You can use system profiler, it will do a tracing with resolution of
> >>> clockcycles, it has view that can aggregate data per thread on
> >>> selected region
> >>>
> >>> On 11/24/2011 09:24 AM, Gregoire Banderet wrote:
> >>>> Hi,
> >>>>
> >>>> I eventually have some time to look into this...
> >>>>
> >>>> I followed the QNX documentation under "Controlling processes via
> >>>> the /proc filesystem", and I use open/devctl, DCMD_PROC_STATUS
> and
> >>>> the
> >>> member procfs_status.sutime.
> >>>> Unfortunately sutime increases by step of the cycle time value :-(
> >>>> That means I cannot reach the granularity (1 microseconds) I need...
> >>>>
> >>>> -- Greg
> >>>>
> >>>>
> >>>>
> >>>> _______________________________________________
> >>>>
> >>>> OSTech
> >>>> http://community.qnx.com/sf/go/post90277
> >>>>
> >>>
> >>>
> >>> _______________________________________________
> >>>
> >>> OSTech
> >>> http://community.qnx.com/sf/go/post90287
> >>>
> >>
> >> Please be advised that this email may contain confidential
> >> information.  If you are not the intended recipient, please notify us
> >> by email by...