Re: question about thread scheduling

now the control thread get running every 4ms, but I got another question on
thread scheduling. Since the thread doing neural network stuff want to use
up the CPU time allocated to it as much as possible, it is an optimization
problem, the longer it runs, the better the result, but at some point, it
will be preempted by higher priority thread: control thread,
and enter a suspended state, how can I get the thread doing neural network
stuff start over again when next time it is awakened. Sicne normally it
would start from where it is preempted, right?

thanks for help
zhiqiang
"Zhiqiang Li" <zhiqiang.li@xxxxxxxxxx> schrieb im Newsbeitrag
news:%23NGlnpiaHHA.1296@xxxxxxxxxxxxxxxxxxxxxxx
thanks for your help!
I will try what you suggested, the reason that I didn't use the sleep
method from the beginning is that I want a better resolution of timer,
that interrupt exactly every 4ms, but using sleep will only get an
accuracy of 4-5ms, am I right?
your suggestion is helpful to bring back on the right way!
thanks
Zhiqiang
"Henrik Viklund" <henrik.viklund@xxxxxxxxx> schrieb im Newsbeitrag
news:1174309826.982542.173700@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
But why complicate things so much? Why mess with the system tick and
introducing periodic sysinterrupts?

Have I missed something here?!? As far as I can tell, there's no need
whatsoever to tamper with the timer tick. Why can't you just leave the
scheduler alone and use one high-prio thread that fire an event every
4 ms and let the control loop run i a second thread that is trigged by
the 4ms event?

Also, as voidcoder hinted, the moment you introduce debug prints in
your real-time threads you're compromizing the real-time integrity of
that thread.

Henrik Viklund
http://www.addlogic.se

On Mar 19, 10:55 am, "Zhiqiang Li" <zhiqiang...@xxxxxxxxxx> wrote:
my control thread is triggered only once, not many times(after I have
used
flag FOREVER to remove the while(!pISTStruct->abort){} statement). So,
I
still have problems with rescheduling.
have changed my program to:
DWORD WINAPI OptimizationThread( LPVOID lpvParam )
{
// Do all interrupt processing to complete the interaction
// with the board so we can receive another interrupt.
ISTStruct* pISTStruct=(ISTStruct*)lpvParam ;
SYSTEMTIME st;
#ifdef FOREVER
while(!pISTStruct->abort){
#endif
//wait for the 4ms timer interrupt event...
#ifdef SYNCHR_EVENT
WaitForSingleObject(pISTStruct->hEvent, INFINITE);
#endif
if(pISTStruct->abort){
return 0;
}
GetSystemTime(&st);
DEBUGMSG(TRUE, (L"Current SystemTime%d \r\n", st.wMilliseconds));
#ifdef SYNCHR_EVENT
InterruptDone(SYSINTR_SOFT4MS);
#endif
#ifdef FOREVER
}
#endif
return 1;}

I also modify the code inside "ULONG PeRPISR(void) " in fwpc.c to enforce
it
to return SYSINTR_RESCHED(here FLAG_4MS is defined somewhere in
oal_timer.h
as" #ifndef FLAG_4MS #define FLAG_4MS #endif"):
#ifdef FLAG_4MS
//for the scheduler to reschedule every 4ms
ulRet = SYSINTR_RESCHED;
#endif

I also have problems with debugging code inside PeRPISR(void), I can
not
perform actions like step into, and so on.
For example, when I attach the device, the only available item under
Debug
is Break All, and Stop Debugging. When I click on Break All, it leads me
to
dbgbrk.c. Then I can click on Start, and nothing happened. I think there
must be something wrong with setting breakpoint inside PeRPISR(void),
there
is only circle with a mark on the line, not a filled point.

thanks in advance
Zhiqiang
"voidcoder" <voidco...@xxxxxxxxx> schrieb im
Newsbeitragnews:%23IS6h0ZaHHA.1220@xxxxxxxxxxxxxxxxxxxxxxx





Aren't you, by chance, debug printing every 4ms
from your thread?

Zhiqiang Li wrote:
thanks for your reply.
Now I am just considering getting one thread running every 4ms, later
I
will divide my work into more threads, as you have suggested.

I have got a lot of same debug message, it seems that now the thread
is
rescheduled. But after launching the remote kernel tracker, my debub
messages disappear. And I do not know why.

"Henrik Viklund" <henrik.vikl...@xxxxxxxxx> schrieb im Newsbeitrag
news:1174209635.140915.224560@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
On 17 Mar, 11:12, "Zhiqiang Li" <zhiqiang...@xxxxxxxxxx> wrote:
Hi Remi,
thanks for your advice.
Let me tell you what I exactly want to do:

In essential, it is a control problem, and I must send a voltage to
the
moter every 4ms(this can be configured to another value in other
cases),
and
this voltage is calculated through a process model, which is based on
neural
network. During the 4ms, I must use the CUP resource as much as
possible,
since algorithms on neural networks are always computational
intensive,
and
I have to stop my calculation before is reach the 4ms deadline, so I
am
afraid that I can not use any sleep function to kill the precious CPU
time:)

Um, sounds like you've not grasped all of the fundamentals of
multitasking (or I have totally misunderstood what you're trying to
do). If you have enough cycles to make a 4ms deadline it should be
trivial to achieve what you need using a couple of threads and
Sleep(...). If the NN run in a different thread as the control loop
you simply schedule the control loop with a higher priority than the
NN loop. If they run in the same thread, you could just create a pace-
thread that signals an event every 4 ms (using Sleep(...)) that the
control/NN thread wait for. When trigged, the ctrl thread run to
completion and go back to waiting for the event again.

Henrik Viklund
http://www.addlogic.se

I am using CE 6.0 on a CEPC, and I have modifid the value of
g_dwBSPMsPerIntr to 4, so that it will generate an timer interrupt
every
4ms. I found that if I want to let the scheduler run also every 4ms,
I
have
to modify the code inside "PeRPISR", so that it will generated a
SYSINTR_RESCHED every 4ms.

I am now study this source code, and trying to modify it to meet my
requirement.

best wishes
Zhiqiang
"Remi de Gravelaine" <gravelaine at aton dash sys dot fr> schrieb im
Newsbeitragnews:OHk45G9ZHHA.4520@xxxxxxxxxxxxxxxxxxxxxxx

Zhiqiang,
You should try to tell us *exactly* what you want to do.
You should also avoid multiplying the discussion threads if you want
people to take care of you from the beginning.
If I understood what you want to do, you have implemented a polling
loop
to do some job (what kind of job?) every 4ms.
You have put this loop in a high-priority thread started by the
XXX_Init
function of a stream driver.
You are using CE 6.0 on a CEPC.
The loop should thus look like:
while (!g_bSuicide)
{
Sleep(3);
<do some job>
}
There is no reason that such a loop exits except when g_bSuicide is
set
to
TRUE (something that can be done in XXX_Deinit) or the <do some job>
code
executes a break of goto intruction.
Now, why Sleep(3) instead of Sleep(4)? Because Sleep(3) will sleep
for
*at
least* 3 ms and because Sleep is synchonized on timer ticks, that
fire
every ms (producing the SYSINTR_RESCHED you are focusing on.) So,
Sleep(3)
returns from sleeping right after the third tick and a new 1ms slice
begins. You <do some job> during a portion of this slice and reenter
Sleep(3). Sleep(3) puts your thread to sleep for 3 timer ticks and
that
means the time remaining in the current tick + 3 ms.
HTH
Remi- Dölj citerad text -
- Visa citerad text -- Hide quoted text -

- Show quoted text -





.

Relevant Pages

  • CFS patch (v6) -- dynamic RT priorities?
    ... Kernel module 2 controls the hardware and receives messages from the driver via a message ring. ... When the control thread puts a message into the ring for the driver thread, he calls "up" on a semaphore that the driver exported. ... I expect all 5 threads to be the highest priority tasks in the system and ALWAYS be scheduled if they are ready. ... When I have 3 of the 4 control threads running at a very high frequency of interrupts "s on the sync semaphore when the driver puts a message into the control thread's ring), it appears that the priority of the 3 are somehow getting "elevated" by the scheduler? ...
    (Linux-Kernel)
  • Re: question about thread scheduling
    ... the control thread sets when it wants the NN to start over. ... thread scheduling. ... Since the thread doing neural network stuff want to ... will be preempted by higher priority thread: ...
    (microsoft.public.windowsce.platbuilder)
  • Re: question about thread scheduling
    ... the control thread sets when it wants the NN to start over. ... thread scheduling. ... Since the thread doing neural network stuff want ... will be preempted by higher priority thread: ...
    (microsoft.public.windowsce.platbuilder)
Loading