Sleep, Time delay, Zen Timer?..in need of time accuracy to 1 ms

[marvg007]

Hi, first off, to know why I need precise accuracy in this function, down to the exact millisecond is because I am trying to produce a frequency through my parallel port. Sending frequencies through the port would allow me to make multiple channels in which a robot can communicate with the computer(different frequencies through same pin can mean different commands). I timed the call and return of the _outp()(function to write to the port in VSC++6.0) to be around 1/200,000th of a second. I only need my program to be exact to the 1/1000th of a second so that isn't much of a time consumer. On the other hand, the Sleep(ms) function is a MAJOR time consumer. I can only achieve at most 100Hz doing this:

while(1){

_outp(0x00378, 0x01); //Switch pin 0 to High
Sleep(1);
_outp(0x00378, 0x00); //Switch pin 0 to Low
Sleep(1);
}
The Supposed result should be that it pauses for 1 ms between switching from on to off, meaning that it does a cycle(on off) in 2 ms. This theoretically mean that it would give a frequency of 1000/2 per second or 500Hz.But it gives me 100Hz, 100 cycles a second at most!

and no, I don't think the while loop has anything to do with it since I batch process it, doing it 10 times before repeating a while.(I even experimented with no while, doing it 100 times)

I tried making my own delay function using clock() but the calls to clock() take Significantly longer than the calls to Sleep(). I'm thinking Sleep might be coded in assembly.

I need help finding an assembly written function to be used with C++ or such that might be able to help me make near precise frequencies. I know Micheal Abrash's Zen Timer does some pretty accurate calculations. Can someone show me a way to do this? Thanks!

 

[KeithBrautigam]

Here's a few things to try:

1. Multimedia timers are pretty good, but I'm not sure if they're this good. Anyway, you can read about them at

http://msdn.microsoft.com/library/default.asp?URL=/library/psdk/multimed/mmtime_5h83.htm

2. You might be able to use the QueryPerformanceFrequency and QueryPerformanceCounter functions. See

http://msdn.microsoft.com/library/psdk/winui/timers_4z76.htm

3. You might want to play some games with increasing the priority of your process. See the documentation for "SetPriorityClass" at

http://msdn.microsoft.com/library/psdk/winbase/prothred_9z1v.htm

Please note, however, that as the documentation warns you can get into trouble by making your priority too high. And even with a high priority I'm not sure that there's any guarantee that your process won't get preempted and lose control for more than a millisecond in the middle of the operation.

 

[Cagliostro]

Look here:
while(1)
{

_outp(0x00378, 0x01);
//calling a function=lose time on
//calling, executing and returning

Sleep(1);
//lose time on calling
//waiting one millisecond
//lose time on returning

_outp(0x00378, 0x00);
//lose time on
//calling, executing and returning

Sleep(1);
//lose time on calling
//waiting one millisecond
//lose time on returning

}
this is the first problem.
The second:
Sleep(1) sleep not precisely one millisecond. This function is for tens of milliseconds. By the way, Sleep(0) sleep more then one millisesond.

 

[marvg007]

Thanks Keith, I'm going to try the Multimedia Timer built into windows..but I'm not sure even it can handle < 10 ms delays. I'll give it a try!

John,Yep, I know the function calls take up quite some time. The _outp function though, takes only 1/200,000 of a second(call and return), I timed it by making it Occillate through the port and measured the resulting Frequency output(number of times it goes off and on in a second). That's why I'm asking for a better sleep function..sleep just doesn't work exactly as it should..

 

[Dragoon]

One thing I would consider, if you are at all familiar with threads, is put the frequence generator code in a seperate thread, set to time critical priority. Something like this.

AfxBeginThread(FreqThreadProc, NULL, THREAD_PRIORITY_TIME_CRITICAL);

I have used the QueryPerformanceFrequency and QueryPerformanceCounter functions to determine how long some of my code was running, and I was able to get 1ms resolution.

 

[richw42]

> Sleep(1) sleep not precisely one millisecond. This function is for tens of milliseconds. By
> the way, Sleep(0) sleep more then one millisesond.

Take a look at timeBeginPeriod()
In my tests it lets Sleep(1) be much more precise.