🚀 go-pugleaf

Bob Gonder wrote:
 > "Boguslaw Fries" <B.Fries@luca-online.de> wrote:
 >
 >>I need a function, which sleeps my thread for 5 - 10 milisecond but not
 >>more!
 >>Standad function Sleep( 10 ) - in cfact sleeps my thread fo 30 milisecond
 >>
 >>Any suggestions?
 >
 > Three suggestions:
 > 1) Get rid of Windows and use a real-time OS.
 > 2) Up your priority and only Sleep(5). You'll come back sometime after
 > the 5, and with a high enough priority, you stand a chance of comming
 > back within 10 (I'd guess).
 > 3) Aren't there some high precision timers that would trigger you when
 > they go off?

Are you refering to the *WaitableTimer? Yes, it can be set to go off at 100 
nanosecond intervals. I have no idea how accurate it really is. I have never 
used it.

If you refer to the WinAPI *PerformanceCounter, it won't go off, it just 
counts time. Something like this:

DWORD sleeptime=10;
LARGE_INTEGER Before, After, Freq;

if( QueryPerformanceFrequency(&Freq) ) {
CriticalSection
   QueryPerformanceCounter(&Before);
   Sleep(sleeptime);
   QueryPerformanceCounter(&After);
   /*
   ** Diff in seconds:
   ** (double)(After.QuadPart-Before.QuadPart)/Freq.QuadPart -
   **         (sleeptime/1000.0)
   */
}

You could boost the thread priority to reduce the impact of context switching 
to get a more accurate diff and you could also loop many times to get the 
average and min/max diffs. In such a loop, you could also slowly adjust 
sleeptime until you get the sleeptime to be used in subsequent Sleep()s to get 
the pause duration you need.

The *PerformanceCounter is very accurate. A friend and I used it to measure 
the frequency of pulses (from the DSR pin on the serial port connected to a 
tone generator - not recommended by any serial port manufacturer :-) and had 
an oscilloscope as a reference counter. From what I recall, we got a very 
little differance (I don't recall the % but we decided it was negligible) in 
the result on the scope and using the *PerformanceCounter on NT4 up to 
something like 300000pulses/s. Not a very scientific statement perhaps, but I 
wouldn't hesitate using it for similar tasks in the future. :)

Kind regards,
Ted Lyngmo

-- 
Replace MY_FIRST_NAME in my email address with my first name to send a private
email to me.